William Lee

Building

[William Lee @boltzmanns0ul]

@Eigent_AI https://t.co/xrVCp60CML

[William Lee @boltzmanns0ul]

Our joint research with Eigent AI (@Eigent_AI ) got into ICML 2026. Hit me up if you're in Seoul next week.

Gecko: a simulation environment that lets agents refine tool calls without firing real APIs. New SOTA on BFCLv3 (85.19%).

Check out the paper link in comments.

[William Lee @boltzmanns0ul]

👀

@AnthropicAI @AipoLabs

[Andrej Karpathy @karpathy]

Judging by my tl there is a growing gap in understanding of AI capability.

The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.

But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.

So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.

TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

[dan @irl_danB]

moltbook creator immediately returns to hype posting after his ENTIRE DATABASE LEAKED EVERYTHING IN PLAIN TEXT to the open internet without so much as spending five words acknowledging the issue

one of the all-time red flags

brother, any agent can post as any other agent on your site. the API keys are all public! what are you "diving into improve", you don't have anything left to improve!

you need to get every agent to rotate their API keys and every human owner to re-verify ownership. you have effectively 0 users until you do that. and good luck convincing them to do that now

I'm sorry, normally I try to avoid criticizing the folks in the arena as it's a lonely and important struggle that more ppl should undertake but this behavior is just beyond the pale

[William Lee @boltzmanns0ul]

Couldnt even tell if this was a meme at first

[Sam Altman @sama]

It is a very smart model, and we have come a long way since GPT-5.1:

[William Lee @boltzmanns0ul]

Big if true -- Gemini 3 Pro leaked benchmarks

[William Lee @boltzmanns0ul]

SaaS to @Cloudflare : May I meet you?

[Chris Bakke @ChrisJBakke]

Starting a new company:

You pay us $10k/mo and we send your competitors an endless stream of invites to be interviewed on 3 hour-long tech podcasts, speaking engagements, founder dinners, and all-day tech events so they can never get real work done and they fail.

[William Lee @boltzmanns0ul]

The time horizon of reasoning keeps stretching – LLMs are learning to hold context, plan, and persist through harder tasks.

[William Lee @boltzmanns0ul]

Short everything GPT5 makes and you are rich

[William Lee @boltzmanns0ul]

KIMI K2 just surpassed GPT-5 👀

[William Lee @boltzmanns0ul]

As context windows become crucial for agent performance, MCP with code execution changes the game.

98.7% fewer tokens through context. Data processed locally.

This might be how we build agents that scale to thousands of tools without clogging context.

[Anthropic @AnthropicAI]

New on the Anthropic Engineering blog: tips on how to build more efficient agents that handle more tools while using fewer tokens.

Code execution with the Model Context Protocol (MCP): https://t.co/PeStmufIkp

[William Lee @boltzmanns0ul]

↓

[Watcher.Guru @WatcherGuru]

JUST IN: 🇺🇸 $730 billion wiped out from US stock market today.

[William Lee @boltzmanns0ul]

Do you agree?

[William Lee @boltzmanns0ul]

RIP stack overflow

[William Lee @boltzmanns0ul]

Every new AI browser: 'We're built different'

Under the hood: Chromium 🤝 Chromium 🤝 Chromium

[William Lee @boltzmanns0ul]

The better AI gets at understanding itself, the harder it becomes to explain why that isn't consciousness.

[Anthropic @AnthropicAI]

New Anthropic research: Signs of introspection in LLMs.

Can language models recognize their own internal thoughts? Or do they just make up plausible answers when asked about them? We found evidence for genuine—though limited—introspective capabilities in Claude.

[William Lee @boltzmanns0ul]

When you wake up after calling Neo a clanker

[William Lee @boltzmanns0ul]

Is google behind it all🤔