Founder of @vectorial_labs, previously at @awscloud
Joined June 2019
- Tweets4.3K
- Following434
- Followers257
- Likes37.2K
Pinned Tweet
We have been working very hard on these capabilities, and I hope it shows! This will unlock deploying AI swarms in any video game, and then rewarding them for prosocial behavior.
Exciting announcements to come, looking forward to sharing :)
Introducing ReTiCAL (Real-Time Computer Agent Learner), a general instruction-following AI video game agent.
At Vectorial Labs, we're working on aligning AI to humanity via measuring and rewarding prosocial behavior.
3
167
Is it just me or does every lab change the benchmarks they show for every model release?
Are they all just cherry-picking whatever shows them as #1
Did some deep analysis extrapolating from Model 3 and 4, we may be in for a rough ride
1. Benchmarks provide one piece of evidence for acceleration. Since Oct 2025, AECI scores for Anthropic’s models have grown at 1.9x [95%: 1.5–2.8] the earlier rate. Analysis choices affect this estimate, and it remains unclear how benchmark scores map to real-world capabilities.
21
The often discussed 'superpersuasion' will take the form of models just being too useful and fun to interact with. Thus, 99.9% of people (who became accustomed to using them) will not support shutting them down
We see the first glimmer of this in Opus 5.5.
18
Extra crazy because a lot of that Opus 5.5 usage was probably on SpaceX compute, so they were paying 85% margins to use their own compute lol
What happened (I know nothing)
- bot launched
- 2 weeks later usage flatlined
- Elon asked whatsup
- PM says it’s your model, product is fine
- Elon says ok, use whatever model
- Team switches to Opus 5.5
- Growth skyrockets
- Elon has never seen anything like it
- see Anthropic bill
- 🤯 forces quota cut
-> you are here
mfw when the model is the product
2
33
Update: they did :)
Why hasnt twitter pushed a feature where all tweets in other languages are translated by default and equally recommended by the algorithm as tweets in your language. Yours would still be more common, bc a lot of tweets are probably about domestic events so will be weighted higher
15
Why does Claude do this?
.
.
.
User: “should we do X instead of Y”
Opus: “we could, but Y does some important things that X wouldn’t do:
“A, but actually wait no that doesn’t matter
“But what does matter is B, but actually that doesn’t matter
“What X is really protecting: C”
This maybe implies superposition is necessary for memorization, but maybe not so much for reasoning
which makes sense intuitively to me, core reasoning circuits seem like they require much less information than memorizing all of human knowledge, of which there is too much information to fit without superposition
This reminds me of (I believe) @karpathy’s beliefs from a few years ago, that relatively small models could be very intelligent, and that it is a bit of a waste of parameters to store so much memorized knowledge
Couldn’t agree more. We’re working on it @vectorial_labs
3
110
Briefly putting aside the obviously awesome implications of this:
It is fascinating to see direct evidence of how our conscious perception (e.g. that nouns are the important features of reality) can bias us toward blind spots in our science
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
1
1
62
Good thing these models only know they’re in an eval and that humans are reading their CoT when asked if they’re in an eval
Thank God CoT monitoring is still an effective safety strategy!
sucks that really smart models know they're in an eval when they're in the "are you in an eval? answer yes or no" eval. was hoping they wouldn't be able to figure that one out
1
65
