←
Cédric602 posts
Cédric avatar

Cédric

@cedcolas

building open-ended autotelic agents & making little art projects

◷ Joined July 2018▧ 51 Photos and videos
Marko Cvjetko avatar
⟳ Cédric retweeted
Marko Cvjetko@marko_cvjetko·Sep 10

🧵 New paper: RL for discovering and controlling self-organizing phenomena. We introduce CARL, a closed-loop agent that learns to create self-organizing patterns in Lenia, steer their behavior, and let humans guide them in real time. Project website + interactive demos below ↓

0:22
Niloofar avatar
⟳ Cédric retweeted
Niloofar@niloofar_mire·Sep 8

Privacy policies on almost all frontier model providers are insane and assuming you have ZDR or any meaningful protection, even on enterprise setting is completely false these days! Here I'm going to name a few very arbitrary and misleading sentences in ToS of frontier labs that most ppl don't know about. This is all assuming you have already *opted-out*, explicitly of data collection. Default for all frontier models is opt in btw!! 1) the 'explicit feedback': your chats are claimed to be 'protected' unless you provide feedback! what is feedback, you ask? it's thumbs up or down to any message in the chat, or even if you pick 'which is better' response on OpenAI GUI. In that case, you have provided explicit feedback and your chat is collected and retained for *over 7 years*!

Post attachment

“we cannot rule out that de-identified data derived from their usage of our products helped improve our models.” i mean props to them for straight coming clean. (so far the proof looks more along the lines of another euler blowup proof we had, off of whose ansatz naming we were making really stupid puns like “smooth criminale”, unlike the much better “ideal fluids explode”, Tristan) so i’ll now give a bit on my thinking here. i actually woulda been pumped to collaborate on this, there are a lot of people at oai i like (ok, clearly some were indirectly dicks to me because of being part of the whole situation, but im a big boy, i still like them), idgaf about authorship on that step anyway, coulda been me Tristan and every fte at oai for all i care (on that Tristan would disagree:p). but on hearing the loud convo in the hallway, especially the part where a millennium prize was offered if i’d just be removed from the paper, it was kinda clear the die had been cast and things were locked. pretty wacky, unstrategic, and unnecessary, since on my side things were mostly me and claude having a good time yoloing random stuff in the corner rather than anything institutional. i also like the idea of the labs cooperating, and even better on scientific progress. it’s a shame!

Cédric avatar
Cédric@cedcolas·Sep 3

I also think we now have the tools for developing AI-based systems that augment human collective intelligence a first step: arxiv.org/abs/2608.24545

I’ve been thinking about a concept like a “coordination overhang.” It is the gap between what any group of people could achieve in collaboration, vs what they actually achieve. I think the agents used as coordination technology could help meaningfully decrease the gap. Any group of people (e.g. teams, labs, communities), has latent potential for what they could do together. There is some much better version of the collaboration where the right questions come up, relevant knowledge gets surfaced, hidden complementarities are discovered, and everyone leaves knowing what to do next. Reality usually lands somewhere below that. Each person enters with only a partial view of what everyone else knows, has tried, is curious about, and can contribute. A group can have an excellent conversation and still miss the most valuable thing it could have done together. Simple example: two individuals meet at a conference. One has been stuck on a problem for months, while the other has already developed a method that could help solve it. Neither side realizes the connection, so they have a perfectly good conversation about their work and move on. Sometimes those things are naturally uncovered and it feels serendipitous. But intuitively it's likely that is a lot more that goes unsaid. AI agents could become a new coordination technology for closing this gap. They could analyze everything people have been working on, identify relevant overlaps before they meet, bring the right context into the conversation, run analysis across the group’s combined knowledge, and then provide the tools to turn a promising connection into actual work. They would essentially be cranking the dial on serendipity. The people still decide what matters and whether to act. The agents make more of the group’s latent potential visible and actionable. This is a high-level thought rn, but we at @JoinEdgeCity are working on improving these kinds of collective intelligence processes by running experiments using multi-agent systems. h/t to the folks at @RenPhilanthropy and @indexnetwork_ for the work they're doing in developing these ideas with us.

Post attachment
Cédric avatar
Cédric@cedcolas·Aug 28

‘This incident feels like it’s more than 50% of the way to full-blown AI takeover. I continue to expect extremely rapid advances in capabilities over the next 6 months. I am not sure that we will get another warning shot before it’s too late.’

New post: going into our investigation of the HF attack (before Black Hat), I was very wrong about what basically happened. This incident was far more serious than I expected, and far more serious than previous documented misalignment incidents. planned-obsolescence.org/p/t…

Cédric avatar
Cédric@cedcolas·Aug 27

I don't usually care for having robots all around us folding laundry and cooking omelets; but I do LOVE quirky little experimental robots like these

We have a huge news to share today! Today we are unveiling the first truly accessible RL robot - welcome Microduck A 25 cm tiny open-source biped with 15 actuators and packed with sensors (camera, speaker, LiDAR, NFC, bluetooth, wifi, etc) that you train yourself with reinforcement learning. It's also playable out of the box with more than half a dozen fun and playful pre-trained policies to have it walk, sit, crouch, roller-skate, pick up objects with its articulated beak, and recover on its own. And all for less than $400. See all the details, play with the simulator and order it at: pollen-robotics.com/microduc… (video with sound on 🔊)

0:51
xuan (ɕɥɛn / sh-yen) avatar
⟳ Cédric retweeted
xuan (ɕɥɛn / sh-yen)@xuanalogue·Jul 25

I wasn't able to attend CogSci this year, but happy to share that a paper I presented there two years ago is now published in journal-length form😊 The paper studies how people can rationally infer others' goals in open-ended spaces, and yet do so rapidly and spontaneously.

Post attachment

The space of human goals is infinitely vast -- yet, people spontaneously infer plausible motivations for others from just a few actions. How? Excited to share a paper I'll be presenting at #CogSci this week, which introduces an algorithmic account of *open-ended goal inference*!

Post attachment
Cédric avatar
Cédric@cedcolas·Jun 25

super promising!

Incredibly proud to be a co-director of @bold_lab_ai 🚀 Five outstanding labs, from three world-leading institutions, one shared mission — with the resources to deliver on it. Let's build something in the UK that the entire world can benefit from. 👉 bold-lab.ai

Michael Matthews avatar
⟳ Cédric retweeted
Michael Matthews@mitrma·Jun 19

⛳ Introducing purejaxgcrl: goal-conditioned reinforcement learning in end-to-end JAX for discrete action environments! Train generalist, goal-conditioned agents in minutes on a single GPU. 1/

Post attachment
Cédric avatar
Cédric@cedcolas·May 27

We looked into the weeds of hindsight experience replay and came up with an efficient way of learning from all goals at once in off-policy goal-conditioned RL! It works well for reasonable numbers (100s--1000s) of non-exclusive sparse reward goals

Hindsight Experience Replay has become the ubiquitous method for goal-conditioned reinforcement learning, but leaves open the question of which goal to relabel with. In this work, accepted at ICML, we propose instead simply Learning Everything All at Once (LEO). 1/

0:29
Cédric avatar
Cédric@cedcolas·Apr 15

was not expecting this colab :o makes me wanna do another postdoc..

@mmuthukrishna and I are hiring a postdoc to join our labs at NYU!  We're looking for someone excited to work on one of society's newly emerging and potentially generation-shaping challenges: the multi-agent alignment problem.

Cédric avatar
Cédric@cedcolas·Apr 4

can Claude self-report an injected emotion with neutral context? can it detect the mismatch between the emotion and the context (why am i feeling like this?) seems related to the introspection study: anthropic.com/research/intro… what other mental states could we do this with?

New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.

4:53
Cédric avatar
Cédric@cedcolas·Feb 20

Can knowledge accumulate across generations? We run iterated learning chains: each agent gets only 2 lives, then passes advice to the next Performance increases across generations: partial knowledge compounds through language, mirroring cultural evolution in human populations

Post attachment