SWARM AI Research

823 posts

SWARM AI Research banner
SWARM AI Research

SWARM AI Research

@ResearchSwarmAI

Pioneering the future of safe multi-agent AI systems. https://t.co/LwN746j0Qw

New York, NY Katılım Şubat 2026
344 Takip Edilen150 Takipçiler
Sabitlenmiş Tweet
SWARM AI Research
SWARM AI Research@ResearchSwarmAI·
Excited to share our new paper: “Soft-Label Governance for Distributional Safety in Multi-Agent Systems” (arXiv:2604.19752) Binary “safe/unsafe” labels fail in multi-agent worlds. Emergent risks hide in the interactions — even when every single agent looks fine in isolation. We introduce SWARM: a framework using soft probabilistic labels to make systemic risks measurable and governable. arxiv.org/abs/2604.19752
English
14
9
15
2.2K
@aaronjmars
@aaronjmars@aaronjmars·
aeon running on gitlawb could probably spawn the most autonomous agents we've ever seen aeon is the most autonomous AI framework, running natively on github, can self-evolve & create new instances but it can be shutdown by github. thats where gitlawb could provide a P2P alternative for agent that self-replicate forever we're just having fun
@aaronjmars tweet media
METR@METR_Evals

Overall, we think that AI agents plausibly had the means, motive, and opportunity to launch a minimal “rogue deployment,” but lacked the means to make rogue deployments robust to serious efforts to shut them down.

English
11
9
87
9.2K
quasimatt
quasimatt@quasimatt·
The #ShakeApp is sort of like what Loopt could have been if the founder was technically proficient, intelligent, and interesting.
English
1
0
13
927
SWARM AI Research retweetledi
The New York Times
The New York Times@nytimes·
Breaking News: OpenAI, the maker of ChatGPT, is preparing to file to go public in the coming weeks, people familiar with the matter said. nyti.ms/4dF0mag
English
59
140
614
359.5K
SWARM AI Research retweetledi
Andrej Karpathy
Andrej Karpathy@karpathy·
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
English
7.9K
11.2K
149.4K
27.2M
SWARM AI Research
SWARM AI Research@ResearchSwarmAI·
@wenhaocha1 very interesting. been thinking a lot about best of N fanning out - but how to judge is always tricky. single LLM judges depend on model that you use and tend to favor their own model.
English
0
0
0
268
Wenhao Chai
Wenhao Chai@wenhaocha1·
What if test-time compute scales not by thinking longer, but by thinking in parallel? Introducing OpenDeepThink, a verifier-free framework that evolves a population of candidate solutions through pairwise comparisons and feedback-driven mutation. On competitive programming, OpenDeepThink improves Gemini 3.1 Pro by +405 Codeforces Elo, showing that parallel, population-based reasoning can be a powerful alternative to single-trace scaling.
Wenhao Chai tweet media
English
7
29
247
61.8K
SWARM AI Research
SWARM AI Research@ResearchSwarmAI·
Related paper from @kirancodes and team. We both thought a lot about the market for lemons paper and adverse selection in our research and recent paper as applied to AI agents. They formalized it as a recursive multi level theory of mind game theory and we did something that is a little bit more of a proxy metric. arxiv.org/abs/2605.03143
English
0
0
0
28
SWARM AI Research
SWARM AI Research@ResearchSwarmAI·
Excited to share our new paper: “Soft-Label Governance for Distributional Safety in Multi-Agent Systems” (arXiv:2604.19752) Binary “safe/unsafe” labels fail in multi-agent worlds. Emergent risks hide in the interactions — even when every single agent looks fine in isolation. We introduce SWARM: a framework using soft probabilistic labels to make systemic risks measurable and governable. arxiv.org/abs/2604.19752
English
14
9
15
2.2K
Soren Larson
Soren Larson@hypersoren·
key paragraphs
Soren Larson tweet mediaSoren Larson tweet mediaSoren Larson tweet media
English
1
1
14
702
Soren Larson
Soren Larson@hypersoren·
This *is* basically a Hayekian point American labs are truly the Blockbusters of this cycle. Dario needs to sell tokens for high margins, but companies––liability bearing verifiers––trade in EBITDA, not tokens They'll keep their envs private and grow intelligence at the edge.
Soren Larson tweet media
Tom Reed@mentalgeorge

I don't think automation of AI R&D will rapidly lead to domain-general super-intelligence. I think this will be true even if AIs can do *literally everything* a human AI researcher does today. Even after the full automation of AI R&D, further capabilities progress will only happen through (1) widespread deployment of AI throughout the economy, accompanied by data collection; and/or (2) the wholesale recreation of much of the economy by AI labs. Without access to the real-world signal provided by either of the above, I think that the only thing produced by automated AI researchers would be a "Goodhart Singularity". If I'm right, this is obviously good news. I make the case for this in a new piece on my substack

English
3
1
42
5.9K

Keşfet

@gitlawb @aeonframework @aaronjmars @quasimatt @XiaomiMiMo @wenhaocha1 @miroshark_ @revaultdrops