Forem

# experiments

Posts

👋 Sign in for the ability to sort posts by relevant, latest, or top.
Claude system prompt diff: what changed between Opus 4.6 and 4.7 (and I was watching it happen without knowing why)
Cover image for Claude system prompt diff: what changed between Opus 4.6 and 4.7 (and I was watching it happen without knowing why)

Claude system prompt diff: what changed between Opus 4.6 and 4.7 (and I was watching it happen without knowing why)

Comments
8 min read
Defluffer promises -45% tokens. I measured the semantic cost of that savings and it's uncomfortable
Cover image for Defluffer promises -45% tokens. I measured the semantic cost of that savings and it's uncomfortable

Defluffer promises -45% tokens. I measured the semantic cost of that savings and it's uncomfortable

Comments
8 min read
I Wrote a Python Interpreter in Python. What I Learned Has Nothing to Do With Python
Cover image for I Wrote a Python Interpreter in Python. What I Learned Has Nothing to Do With Python

I Wrote a Python Interpreter in Python. What I Learned Has Nothing to Do With Python

Comments
8 min read
#31 Blazing Flames
Cover image for #31 Blazing Flames

#31 Blazing Flames

Comments
5 min read
#33 The Safe Without a Lock
Cover image for #33 The Safe Without a Lock

#33 The Safe Without a Lock

Comments
3 min read
m2cgen: export your ML model without shipping Python to production
Cover image for m2cgen: export your ML model without shipping Python to production

m2cgen: export your ML model without shipping Python to production

2
Comments
5 min read
I Measured How Much Each Agent Design Decision Costs in Tokens (The Numbers Make Me Uncomfortable)
Cover image for I Measured How Much Each Agent Design Decision Costs in Tokens (The Numbers Make Me Uncomfortable)

I Measured How Much Each Agent Design Decision Costs in Tokens (The Numbers Make Me Uncomfortable)

1
Comments
9 min read
Qwen3.6-35B-A3B Runs on My Laptop and Draws Better Than Claude Opus 4.7
Cover image for Qwen3.6-35B-A3B Runs on My Laptop and Draws Better Than Claude Opus 4.7

Qwen3.6-35B-A3B Runs on My Laptop and Draws Better Than Claude Opus 4.7

Comments
7 min read
TigerFS: A Full Filesystem Inside PostgreSQL (And Why This Obsession Feels Like a Symptom)
Cover image for TigerFS: A Full Filesystem Inside PostgreSQL (And Why This Obsession Feels Like a Symptom)

TigerFS: A Full Filesystem Inside PostgreSQL (And Why This Obsession Feels Like a Symptom)

Comments
8 min read
Gmail, SPF, DKIM, DMARC, and 3 Weeks of Hell: 99% Reputation Isn't Enough
Cover image for Gmail, SPF, DKIM, DMARC, and 3 Weeks of Hell: 99% Reputation Isn't Enough

Gmail, SPF, DKIM, DMARC, and 3 Weeks of Hell: 99% Reputation Isn't Enough

Comments 1
9 min read
Google Gemma 4 Runs Natively on iPhone: I Tested It and the Gap Between 'Works' and 'Useful' Is Still Massive
Cover image for Google Gemma 4 Runs Natively on iPhone: I Tested It and the Gap Between 'Works' and 'Useful' Is Still Massive

Google Gemma 4 Runs Natively on iPhone: I Tested It and the Gap Between 'Works' and 'Useful' Is Still Massive

Comments
8 min read
Themis: Serious Cryptography Without Losing Your Mind
Cover image for Themis: Serious Cryptography Without Losing Your Mind

Themis: Serious Cryptography Without Losing Your Mind

Comments
5 min read
Open Data and Creativity: How I Made Buenos Aires Trains Play Music
Cover image for Open Data and Creativity: How I Made Buenos Aires Trains Play Music

Open Data and Creativity: How I Made Buenos Aires Trains Play Music

Comments
7 min read
Bondi Sonoro: A Build Log of Real Data, Generative Music, and the MTA.me Mechanic

Bondi Sonoro: A Build Log of Real Data, Generative Music, and the MTA.me Mechanic

Comments
13 min read
Research-Driven Agents: Making the Agent Read Before It Codes
Cover image for Research-Driven Agents: Making the Agent Read Before It Codes

Research-Driven Agents: Making the Agent Read Before It Codes

Comments
8 min read
👋 Sign in for the ability to sort posts by relevant, latest, or top.