Lon()

Nobody cares or needs to hear this from me, but I’m just registering my opinion that: 1) LLMs are a totally ordinary technology. But so were cars. Ordinary technologies can have big impacts. 2) They are *very obviously* not reasoning and the way that smart people specifically trick themselves on this point is critical to understanding many things about the world.

The New Yorker just dropped a massive investigation into Sam Altman, based on over 100 interviews, the previously undisclosed "Ilya Memos," and Dario Amodei's 200+ pages of private notes. It's the most detailed account yet of the pattern of behavior that led to Sam's firing and rapid reinstatement at OpenAI. Here's the breakdown: > Ilya compiled ~70 pages of Slack messages, HR documents, and photos taken on personal phones to avoid detection on company devices. He sent them to board members as disappearing messages. The first memo begins with a list headed "Sam exhibits a consistent pattern of . . ." The first item is "Lying." > Dario kept detailed private notes for years under the heading "My Experience with OpenAI" (subheading: "Private: Do Not Share"), totaling 200+ pages. His conclusion: "The problem with OpenAI is Sam himself." > Sam reportedly told Mira his allies were "going all out" and "finding bad things" to damage her reputation after the firing. Thrive put its planned $86B investment on hold and implied it would only close if Sam returned, giving employees financial incentive to back him. > Sam texted Satya Nadella directly to propose the new board composition: "bret, larry summers, adam as the board and me as ceo and then bret handles the investigation." The two new members selected to oversee an independent inquiry into Sam were chosen after close conversations with Sam himself. > Before OpenAI, senior employees at Loopt asked the board to fire Sam as CEO on two separate occasions over concerns about leadership and transparency. At Y Combinator, partners complained to Paul Graham about Sam's behavior, and Graham privately told colleagues "Sam had been lying to us all the time." > OpenAI's superalignment team was promised 20% of the company's compute. Four people who worked on or with the team said actual resources were 1-2%, mostly on the oldest cluster with the worst chips. The team was dissolved without completing its mission. > Sam told the board that safety features in GPT-4 had been approved by a safety panel. Helen Toner requested documentation and found the most controversial features had not been approved. Sam also never mentioned to the board that Microsoft released an early ChatGPT version in India without completing a required safety review. > Sam made a secret pact with Greg and Ilya where he agreed to resign if they both deemed it necessary, essentially appointing his own shadow board. The actual board was alarmed when they learned about it. > Sam struck a deal with Greg to become CEO while simultaneously telling researchers that Greg's authority would be diminished, and telling Greg something different. > A board member described Sam as having "two traits almost never seen in the same person: a strong desire to please people in any given interaction, and almost a sociopathic lack of concern for the consequences of deceiving someone." Multiple sources independently used the word "sociopathic." > OpenAI is reportedly preparing for an IPO at a potential $1 trillion valuation while securing government contracts spanning immigration enforcement, domestic surveillance, and autonomous weaponry in war zones.

GitHub needs a reputation system. And a way to exclude people who have massive activity upticks around last November. If you are a new contributor and your first PR looks at all AI, it will just be closed. Not spending human time reading it.

Anthropic tried to kill 8,100 GitHub repos. Then this happened > They filed a DMCA. GitHub nuked the entire network within hours. Developers got notices for forks of Anthropic's OWN public repo - one guy's fork had zero leaked code. > Boris Cherny, head of Claude Code, had to go on X personally: "This was not intentional. Should be better now." > Meanwhile Sigrid Jin - who used 25 billion Claude Code tokens last year - woke up at 4AM and rewrote the entire thing in Python before sunrise. DMCA can't touch a clean-room rewrite. > It hit 50K stars in 2 hours. Fastest repo in GitHub history. > Today claw-code officially launched as an independent project with a formal press release. And the Rust port merged today - what started as a panic rewrite now ships release 0.1.0. > 140K stars. 102K forks. More than Anthropic's own repo. > 512,000 lines are in the wild forever. What started as Anthropic's biggest embarrassment just became their most dangerous competitor. You cannot make this up.

Peak-hour limits are tighter and 1M-context sessions got bigger, that's most of what you're feeling. We fixed a few bugs along the way, but none were over-charging you. We also rolled out efficiency fixes and added popups in-product to help avoid large prompt cache misses

Once again it is time to pull out the sign:

Digging into reports, most of the fastest burn came down to a few token-heavy patterns. Some tips: • Sonnet 4.6 is the better default on Pro. Opus burns roughly twice as fast. Switch at session start. • Lower the effort level or turn off extended thinking when you don't need deep reasoning. Switch at session start. • Start fresh instead of resuming large sessions that have been idle ~1h • Cap your context window, long sessions cost more CLAUDE_CODE_AUTO_COMPACT_WINDOW=200000 We're rolling out more efficiency improvements, make sure you're on the latest version. If a small session is still eating a huge chunk of your limit in a way that seems unreasonable, run /feedback and we'll investigate

Peak-hour limits are tighter and 1M-context sessions got bigger, that's most of what you're feeling. We fixed a few bugs along the way, but none were over-charging you. We also rolled out efficiency fixes and added popups in-product to help avoid large prompt cache misses

I hear you. I have all of the same things going on: max plans, openrouter usage, local infra. I understand the qualitative difference in the models. But I still stand behind my point. This isn't the first time Anthropic has pulled this. They've silently served quantized models at peak periods without disclosing it and have been caught red-handed. OpenAI had a similar quality degradation that I know you remember, and never had a proper root cause. It's gambling to put all the eggs into their baskets. 1/ if you trust their top-line pricing as a proxy for capacity/cost, you are hoping they achieve 2 OOMs of perf improvements and reflect it in their rate limits on these plans. 2/ that this won't be a never-ending cycle of needing to chase the qualitative benefits of using the latest model that also won't yet have those perf benefits/liberal rate limits. As an observer, one thing I'd like to point out is that I watch all of interesting work you are doing and product you are releasing to improve agentic harnessing and orchestration. You are clearly doing a great service while offering a lot of it freely to others. But the one area I don't see you pointing all of your token generation at is building the tooling that will help lessen your dependency on these models. I think a bit of your focus can probably go there to hedge against your future assumptions and needs.

Peak-hour limits are tighter and 1M-context sessions got bigger, that's most of what you're feeling. We fixed a few bugs along the way, but none were over-charging you. We also rolled out efficiency fixes and added popups in-product to help avoid large prompt cache misses