food-for-thought

In linear accelerator, injected electrons gain energy from an RF field. Ordinarily, the electrons would then enter a free electron laser and are immediately disposed of in a beam dump. But in an energy recovery linear accelerator (ERL), the electrons circle back into the RF field and lend their energy to newly injected electrons before exiting to a beam dump.

Taste Is All That's Left

September 1, 2017

when one party’s drones are destroyed by drones of another, it will have no other choice but to surrender

development of AI raises “colossal opportunities and threats that are difficult to predict now.”

the one who becomes the leader in this sphere will be the ruler of the world.

Frontier models are now generally very competent at understanding a terminal (or terminal-style) coding environment, and acting within it. Anthropic recently cutting down Claude Code’s system prompt by 80% is a clear sign of this

SIMD allows a CPU to operate on multiple values in parallel. For example, instead of comparing one byte at a time, a CPU can compare 4, 8, or even more bytes with a single instruction.

expert-designed harness can coordinate a large portion of auto-research loop, experimented in the form of writing research papers

AI companies are racing to build AIs that are smarter than humans in every way. In AI 2027, we predicted that this would result in either extinction or irreversible concentration of power.1Plan A is our positive vision for what should happen instead.

The scenario itself was written iteratively: we wrote the first period (up to mid-2025), then the following period, etc. until we reached the ending. We then scrapped this and did it again.

Write a future history of 2022. Condition on it, and write a future history of 2023. Repeat for 2024, 2025, etc. (I'm posting 2022-2026 now so I can get feedback that will help me write 2027+. I intend to keep writing until the story reaches singularity/extinction/utopia/etc.)

What 2026 looks like

rumoured list of papers that Ilya Sutskever gave to John Carmack

Intelligence vs. Cost per Intelligence Index TaskArtificial Analysis Intelligence Index · Weighted average cost (USD) per Artificial Analysis Intelligence Index task

Everything is Logarithms

Logarithms are Vectors

Moving forward, the industry cannot continue to train bigger and bigger models since their intelligence not only plateaus but often will get worse

GPT-5.5 and DeepSeek V4 Pro are two of the clearest hallucination leaders, despite being absolutely huge. Because of their immense size they simply did not learn how to say “I don’t know”

.gitignore Isn’t the Only Way To Ignore Files in Git

You can do things like change the local context window and watch performance improve or degrade, and really dig into how your tokens are processed on the GPU. You can change the system prompt, the quantizations. You can pit models against each other. You can also change and introspect the harness side. The possibilities are endless, and the tools only keep getting better.

YASA's 37 kg stackable 750R motor delivers 800Nm and >5kW/kg with an axial length of 98 mm

proposed lightweight radiator for the dissipation of waste heat generated by power plants, propulsion or spacecraft systems in space.

Our preliminary findings show that the iterative Code Review stage accounts for the majority of token consumption for an average of 59.4% of tokens. Furthermore, we observe that input tokens consistently constitute the largest share of consumption for an average of 53.9%, providing empirical evidence for potentially significant inefficiencies in agentic collaboration.

So what’s actually different between GPT and Claude and Gemini and LLaMA? Public details vary, and the proprietary models do not publish every architectural choice. But at the level this post is covering, they broadly sit in the same transformer-family design space.

CS336: Language Modeling from Scratch

The twenty claims in this corpus with the widest spread between the highest- and lowest-bucket frontier verdicts. These are claims where the panel doesn't just disagree — it disagrees substantively, with at least one model picking a verdict ≥2 buckets away from another.

Flipper One — we need your help

Ultimate collection of online dos games

Live coding session with Boris Cherny and Jarred Sumner

Thermodynamic Computing System for AI Applications

Openrouter for comparable models is 1/3rd the price and ~2x the speed

The harness is built from five extension points—CLAUDE.md files, hooks, skills, plugins, and MCP servers—each serving a different function. The order in which teams build them matters

The sefirot are the ten spheres on the Tree of Life. Each sefirah (singular of sefirot) represents a different aspect of the Divine, as well as aspects of human consciousness and existence. These are, from top to bottom

software bug that seems to disappear or alter its behavior when one attempts to study it