Ted Sumers

Come work with me! job-boards.greenhouse.io/anthropic/jobs…

@deanwball alas

the word of the week is “alas”

New paper out in TiCS: There's a ton of pressure on 'truth'. Is AI spitting out truths, or hallucinations? Are media/govs providing truths, or fake news? We argue that answering such qs requires understanding people's *intuitive theories of truth*: authors.elsevier.com/c/1mgHP_V1r-Mu…

ok in the past 48 hours i have learned twitter is not good for my mental health, marriage, or kids gonna go touch grass and hang with the family, then go back to work on monday will be back for the next major moral crisis. wishing everyone the best til then.

@ShakeelHashim survivorship bias at this point?

the most credulous employee base in the entire world

@GarrisonLovely yeah i’m honestly curious about the psychology there. somehow our views about the technology we’re building are just radically different.

If you work at OpenAI, Google (esp DeepMind), or xAI and you're not lobbying your bosses to draw the same line here, you are complicit in letting your work be used to build a totalitarian nightmare state (that you'll also have to live in!). You have more power than you realize.

DoW thing is esp scary because IMO we are frighteningly close to a surveillance dystopia. I build AI-based monitoring tools. We use them to identify misuse of our models (eg cyberattacks). The good news: they work really well. The bad news: you could use them on any data.

Anthropic was started when senior OpenAI researchers were concerned that the company was not doing enough around safety and alignment for the powerful tech they were building. So they started their own company, around principles that builders of AI models should do as much as possible to make sure these systems are not misused, are aligned, and ultimately lead to human flourishing (to the extent this is possible). You can see this everywhere from the model cards to the way these models "behave" in practice. Anthropic has also been one of the largest providers of public goods for the AI community, regularly releasing new research and things like the Economic Index which tracks the impact of AI on the economy. I don't know the extent to which Anthropic will be able to live up to these principles, but they clearly have been a constraint relative to companies like xAI which largely dgaf (e.g., each Grok release barely has a model card). Despite the constraints, and with fewer resources compared to the bigger labs, Anthropic has led frontier model development. Especially in the last 6 months, we have seen the other bigger labs play catch up (e.g., agentic AI tools like Claude Code). They have shown it is possible to lead technological development while maintaining principles. This is now being tested. The US govt, and Pete Hegseth in particular, are attempting an unprecedented destruction of a leading US company (that is what being classified as a supply chain risk is) because it refuses to comply with actions that the US govt itself declares as unlawful. I have deep respect for the many AI researchers and policy folks across the labs, as well as those who are working in the WH/govt. I hope that the community can unite against what will clearly be a huge handicap for US AI leadership, as well as the clear govt overreach which will have horrible consequences for economic uncertainty.

Sad that this is the world we live in, but proud to be part of this company. Developing AI is playing with fire. Let's do it transparently, in accordance with democratic values, and make sure everyone benfits.

I'd strongly recommend reading this. Incredibly proud to work at Anthropic and (still) optimistic we can find a path forward to develop and use AI responsibly in collaboration with USG. anthropic.com/news/statement…

@kushalx7 @Anthropic @AnthropicAI LOL good call thanks 😅

@tedsumers Ted, you might wanna replace @Anthropic with @AnthropicAI in your bio

Come work with me! job-boards.greenhouse.io/anthropic/jobs…

@johnjhorton @AlexTamkin `cat`'s out of the bag i guess

@tedsumers @AlexTamkin i think you mean `ssh -q`

Highly recommend applying to this role if you're at all interested! Extremely high ceiling for impact

@johnjhorton @AlexTamkin sssh 🤫

@AlexTamkin i didn't know ted was on twitter!

strong work @AlexTamkin 🦾

sonnet x.5 continues to outperform 😂

@jayelmnop enjoy your sweaters, imma go make paperclips

I recently moved to the Code RL team at Anthropic, and it’s been a wild and insanely fun ride. Join us! We are singularly focused on solving SWE. No 3000 elo leetcode, competition math, or smart devices. We want Claude n to build Claude n+1, so we can go home and knit sweaters.

cogsci on llms continues

@alex_y_ku omg love this

(1/11) Evolutionary biology offers powerful lens into Transformers learning dynamics! Two learning modes in Transformers (in-weights & in-context) mirror adaptive strategies in evolution. Crucially, environmental predictability shapes both systems similarly.

@karthik_r_n arguably we evolved to communicate so we could manipulate others' behaviors; and modern forms of communication (e.g. twitter) have just pushed that dynamic totally outside its ecological niche. commodification of attention and all that

Humans evolved to communicate so we could coordinate better. But these days, it feels like we communicate so much, yet coordinate so little.

what i've been up to!