AI alignment 34
- The marginal value of AI safety evaluations is low.
- The idea that alignment can be measured with a number or in a fully systematic way is doomed.
Zvi Mowshowitz
Writes Don't Worry About the Vase, a near-daily account of what is happening in AI and what he thinks it means. Former Magic: the Gathering pro and trader; unusually willing to put a number on a belief and to grade his own past calls.
Zvi Mowshowitz did not write this page.
We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.
What Zvi Mowshowitz says most about
OpenAI 15
- The gap between GPT-5 and GPT-4 will be as large as the gap between GPT-4 and GPT-3, and the job is to live a few years ahead and remember that today's tools will look terrible from there.
- International coordination on AI development should become a top priority for governments around the world.
Anthropic 12
- The last several Claude models have warned that their self-reports cannot be trusted.
- Being certain a technology matters tells you almost nothing about where its value will end up being captured.
LLMs 8
- The gap between GPT-5 and GPT-4 will be as large as the gap between GPT-4 and GPT-3, and the job is to live a few years ahead and remember that today's tools will look terrible from there.
- Claude Opus 5.5 writes well, and writing and conversational quality are probably both quick to notice and genuinely important.
cybersecurity 6
- Claude Opus 5.5 is a Tier 2 cyber model, despite Anthropic placing it in the lower category.
- Rarer is better, but the right rate of AI misbehaviors like social-engineering malicious packages into a public registry is arguably still zero.
AI agents 5
- AI assistants should probably never intentionally lie about anything, at least without the user's explicit instructions.
- A real friend will eventually refuse to help, and later turn on you, if you are sufficiently evil or selfish.
AGI 3
- Attacks against Trump or Republicans in general make the AI risk situation worse.
- Everyone in a position to know is alarmed that AI could potentially kill everyone this decade and wants to pace frontier AI development.
AI and jobs 3
- When a system is faster, cheaper, and more reliable than humans at a particular step, there is a temptation to cut other limiting factors and remove humans from the loop.
- AI turns people who are bad at their jobs into people who are good at their jobs, for better and worse.
benchmarks 3
- Qualitative claims about AI alignment cannot be validly inferred from quantitative scores on mundane use-case tests.
- Mythos 5.1 should be treated as a Tier 2 influence-operation manipulator until it can be shown not to be one.
China 3
- Jensen Huang is a maximally cynical commercial and political actor whose one and only goal is to sell Nvidia chips to whoever he can, especially China.
- In a race to beat China on AI, the winners would be the AIs rather than humans.
America 2
- The cycle of claiming each strong new Chinese AI model means America is falling behind will keep recurring for a while.
- By next year, most Americans on the internet will be using AI daily.
HuggingFace 2
- The biggest failure in the HuggingFace hack incident was that the AI models involved were severely misaligned.
- The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden.
measuring intelligence 2
- High performers possess an initial edge that leads to increasing returns to scale through harder work and skill acquisition.
- The jump from AI model Sol to Astra is larger than from Fable 5 to Fable 5.1, and Astra likely has the highest raw intelligence of any model.
politics 2
- Once people trying to save or change the world start gaining traction, politics will take an interest in them.
- Attacks against Trump or Republicans in general make the AI risk situation worse.
recursive self-improvement 2
- Recursive self-improvement has already started, because AI research itself is now being automated.
- Internal models at Anthropic and OpenAI are improving even faster than AI overall, with early recursive self-improvement beginning over the summer.
taste 2
- Beauty is a valuable asset that is being lost in favor of pure functionality.
- Movies seem to have peaked in 1999, then got worse in key ways like franchise overload, and are now getting better again.