Featured post what's this? ✨ Distorted Reality When you divide performance by salary, being expensive looks like underperforming. The metric quietly shifts from impact to intelligence per dollar, and everyone reading it ends up believing something that is not true.
No cost, no value | jola.dev Why generating code makes it meaningless, why writing code by hand has value, and why toil is a critical part of the human experience.
How Claude Could Watermark Text Without Invisible Characters | James O'Reilly Anthropic started watermarking text Claude generates. My two cents on how something like that could actually work.
The FastLanes Unified Transport Layout · blog.dave.tf An explanation of how the UTL layout enables ridiculously fast SIMD delta-decoding
Nothing failed. That's why no one caught it. Generation got cheap, understanding got expensive, and nothing in your toolchain measures the expensive half.
Let Them Write RFCs Or: why I spent hundreds of millions of tokens to write four versions of the same markdown file. I’m working on a complex project that’s just …
The fastest double-to-string algorithm you’ve never heard of The fastest double-to-string algorithm you’ve never heard of
Unauthenticated Root Access to the Enphase IQ Gateway The public disclosure of an RCE in the Enphase IQ Gateway, and the journey that led me here.
Notes from China, Q3 2026 Context: I recently spent 2.5 weeks in China visiting universities, research institutes, clinical sites, companies of all sizes, and investors. Most meetings focused on biotech, neurotech, and AI. Conventional wisdom my trip corroborated US cities feel me...
Fallacy of Decentralization Decentralized networks work well when everyone is following the protocol. But there are always bad actors in large decentralized networks, they attack the...
Time ChatGPT 5.6 Sol Has Been Attempting to Make a Trusted Ethernet Lan VNC-type Client That Doesn't Drop Keystrokes Unsuccessfully. On the Pro Max $200/month Subscription, ChatGPT Sol 5.6 Pro using Codex has been attempting to build a VNC type viewer and server that doesn't drop every second keystroke.
Layer Scope: How I used $20 of compute to make a new way to look at LLMs Reusing an LLM's own final layers as a cheap probe for reading its middle-layer hidden states.
We didn't Kill Physical Media. You did. The recent news from Sony about physical media genuinely pissed me off. Not because I didn't see it coming—I did. We've all seen it coming for years. This has been the direction of travel ever since…
Compression is prediction | ngrok blog Compression and LLMs are trying to solve the exact same problem: predicting what comes next. Learn the fundamentals of compression and how better prediction leads to better shrinkage.
WAL Levels in Postgres and Effective WAL Level in PG19 · Burak Sen A high-level overview of WAL levels and upcoming effective WAL level in PG19
Tiled Rasterization for Large DOM Captures Capture one canonical SVG, move a tile-sized window across it, and rasterize each tile into its own canvas.
My Most Autonomous Loop Has Been Dead for Two Months My fully autonomous pipeline ran for months, quietly died, and nobody noticed - including me. What that taught me about the loop-engineering hype.
Google Search Is Dying. What Comes Next Is Worse | The Walrus As AI eats the web, the internet’s collective memory is disappearing
GitHub - antirez/h3.c: MiniMax H3 inference engine for Mac computers MiniMax H3 inference engine for Mac computers. Contribute to antirez/h3.c development by creating an account on GitHub.
