Pinned
Here’s what I’ve been working on recently: @anthropicai. I’ll be spending a lot of my time on measurement and assessment of our AI systems, as well as thinking of ways govs/others can assess AI tech. There’s a lot to do!
Jack Clark
32.2K posts
Jack Clark
@jackclarkSF
@AnthropicAI, ONEAI OECD, co-chair @indexingai, writer @ importai.net Past: @openai, @business @theregister. Neural nets, distributed systems, weird futures
- People leaving regular companies: Time for a change! Excited for my next chapter! People leaving AI companies: I have gazed into the endless night and there are shapes out there. We must be kind to one another. I am moving on to study philosophy.
- Technological Optimism and Appropriate Fear - an essay where I grapple with how I feel about the continued steady march towards powerful AI systems. The world will bend around AI akin to how a black hole pulls and bends everything around itself.
- One like = one spicy take about AI policy.
- As someone who has spent easily half a decade staring at AI arXiv each week and trying to articulate rate of progress, I still don't think people understand how rapidly the field is advancing. Benchmarks are becoming saturated at ever increasing rates.
- AI skeptics: LLMs are copy-paste engines, incapable of original thought, basically worthless. Professionals who track AI progress: We've worked with 60 mathematicians to build a hard test that modern systems get 2% on. Hope this benchmark lasts more than a couple of years.
- A mental model I have of AI is it was roughly ~linear progress from 1960s-2010, then exponential 2010-2020s, then has started to display 'compounding exponential' properties in 2021/22 onwards. In other words, next few years will yield progress that intuitively feels nuts.
- Five years ago the frontier of LLM math/science capabilities was 3 digit multiplication for GPT-3. Now, frontier LLM math/science capabilities are evaluated through condensed matter physics questions. Anyone who thinks AI is slowing down is fatally miscalibrated.
- Just checking in on alignment of LLMs in China, it's going about how you'd expect.
00:00 - Anthropic will work with the Trump Administration and Congress to advance US leadership in AI, and discuss the benefits, capabilities and potential safety issues of frontier systems.
- An increasing number of people are gazing deep into the dark pool of AI timelines and are realizing that something is about to jump out of that pool upon society. It's interesting to see 'the awakening' ripple across people, reaching ever more distant disciplines.
- In the last decade: - figured out cut&paste for DNA (crispr) - reusable rockets (SpaceX) - crude but generally useful AI systems (llms, image/vid models, RL for inventory) - promising fusion approaches (helion, etc) This decade is going to be so wild. It's very exciting.
- Stable Diffusion: $600k to train. I'm impressed and somewhat surprised - I figured it'd have cost a bunch more. Also, AI is going to proliferate and change the world quite quickly if you can train decent generative models with less than $1m.Replying to @KennethCasselWe actually used 256 A100s for this per the model card, 150k hours in total so at market price $600k
