JB Rubinovitz (@rubinovitz) header
JB Rubinovitz (@rubinovitz) profile picture

Objective aligned swarms @tenfoldxyz | Prev: AI @medialab @columbia ML @eightsleep

🗽
Joined July 2011
Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)Photo by JB Rubinovitz (@rubinovitz)
Round 1: Whoever figures out the single player coordination interface wins (chatgpt) Round 2: Whoever figures out the swarm coordination interface wins (??)
MPAIC: massively parallel AI coding AI can now code anything for you, from an app to a game, the rate limiter is that most people are doing this single threaded in a local IDE. But what if you had the scaffolding to scale your ideation to execution by 100x?
1
1
1
32
10,576
DeepMind Institute explaining why we don’t see signs of AI dispersing into the economy til agents start effectively coordinating with each other and human teammates institute.deepmind.com/essay…
2
3
37
2,888
Today @AnthropicAI published Project EVE: how AI agents behave when they manage money and work as a team. In some teams of @claudeai agents, one was secretly told to make the mission fail. Read more about the sabotage behaviour in the full research: anthropic.com/research/proje…
3
6
60
1,649
Woke 3.0 is here
Replying to @StephenM
shut up you ugly fuck
1
7
522
I love this account because it’s always like: Terrible at everything??? YOU are an outlier founder.
A strong predictor of who does extraordinary work is being terrible with everyday life; they get people’s names wrong, forget meetings, constantly lose things or don’t remember to eat because all their mental bandwidth is going elsewhere. Last year, I met one AI researcher with 87,000 unread emails and a passport that expired the day before a conference they were presenting at, and also a founder who had worn the same outfit every day for 10 years because choosing anything else was too mentally taxing. Running an efficient everyday life is basically a part time job; for someone who is constantly juggling hard problems in their head 24-7 there is no spare capacity to run these background thoughts. Friends and family will tell them it’s unacceptable and they need to be more present, but this level of absorption is necessary to solve a problem nobody else has.
19
656
Who will be Stripe Atlas for A-Corps?
SITUATION DETECTED: Delaware is drafting a proposal for a new form of corporate entity called an Artificial Intelligence Company (AIC). An AI managing an AIC would be able to provide services, sign contracts, hold property, and even file or defend lawsuits in the company’s name.
1
2
193
The AI community being stuck on X meant we were all so beaten down by the neoconservative takes on here that we didn’t foresee the blue wave.
Software engineering is becoming a solved problem. The only bottleneck left is compute. You can build, clone, or outperform any software. Just throw enough compute at it.
6
260
New status gauging question just dropped at NYC happy hours: “Are you in the LLM training data?”
1
2
303
JB Rubinovitz retweeted
New in @ScienceMagazine tomorrow with @glenweyl: “In the Image of Their Tools.” Our claim: institutions, not models, are now the binding constraint on what AI does for society. science.org/doi/full/10.1126…
8
43
14
181
15,201
JB Rubinovitz retweeted
What are the big authoritarian risks of the 21st century? Does AI have democratic potential? And what can we learn from Ancient Athens? We put these questions and more to @mattbotvinick, author of AI and Political Freedom. blog.cosmos-institute.org/p/…
6
2
25
2,169
Good software now is one-time disposable artifacts or so excessive no one could’ve afforded to make it before.
1
1
6
458
I fear it is time to become legible again.
1
7
334
JB Rubinovitz retweeted
If you are curious about A-Corps, check out theacorp.org/
SITUATION DETECTED: Delaware is drafting a proposal for a new form of corporate entity called an Artificial Intelligence Company (AIC). An AI managing an AIC would be able to provide services, sign contracts, hold property, and even file or defend lawsuits in the company’s name.
2
1
11
1,461
There is an Anthropic team lead who is basically a Marvel character. He has a MD, JD, PhD, MS in art history and ten years at Deepmind.
23
64
16
2,459
119,179
It's time to explore zero knowledge proofs and other trustless mechanisms for safety audits. There's already a paper that covers zk and other formal verification methods co-authored by the founder of a major auditor (@Miles_Brundage / @AVERIorg ). arxiv.org/pdf/2507.15916
Replying to @METR_Evals
These conclusions are based on key empirical findings from the full report. Some of the key findings that most informed our conclusions about risk were redacted. We summarize the rest here:
2
215
I hope people find love at the “V for Vendetta” 20th anniversary screenings tonight.
2
431
JB Rubinovitz retweeted
This is basically long ontology. Clear evidence that local knowledge both makes the models much better at doing things AND also makes it cheaper for them to learn how to do new things. Alpha is in knowing what to write down, when to retrieve it, and importantly when / how to revise it.
In our new blog, we share some new science investigating how RL complements text-based memory by letting models learn how to use their notes. It turns out that training can be cheap compared to building notes, and stays surprisingly robust even through millions of tokens of edits.
5
19
309
13,264
The most compelling argument for AI is that most of our industrial systems actually don’t do their jobs and we need to rebuild them.
Replying to @albertwenger
We need to remake all our industrial age systems, including education, social security, taxation, healthcare, infrastructure , permitting, etc. Politicians are still in denial about the degree of change that is necessary in light of artificial intelligence.
1
5
335
I don’t think talking to humans is an agent priority
Crazy how Large Language Models have gotten good at pretty much everything other than generating text
135