Tom Ringstrom 🦡
@no_reward_for_u
Followers
959
Following
4K
Media
62
Statuses
776
Reward-Free Model-based Maximalist. High-dimensional Empowerment. Self-Preserving Autonomous Agents. Theories of intelligence grounded in compositional control.
London, England
Joined March 2012
Thesis is done, defense successful. Will post the finalized PDF sometime later. No reward for u. Abolish the value function!
24
19
199
Thrilled to share our new #NeurIPS2025 paper done at @GoogleDeepMind, Plasticity as the Mirror of Empowerment We prove every agent faces a trade-off between its capacity to adapt (plasticity) and its capacity to steer (empowerment) Paper: https://t.co/prWpkdPojb 🧵🧵🧵👇
25
71
444
LLMs are injective and invertible. In our new paper, we show that different prompts always map to different embeddings, and this property can be used to recover input tokens from individual embeddings in latent space. (1/6)
283
1K
11K
Good news everyone! My long overdue second book - The Revenge of Reason - is currently at the printer and is now available for pre-order on the Urbanomic website (estimated to be sent out mid-October). Link below!
11
88
552
What drives behavior in living organisms? And how can we design artificial agents that learn interactively? 📢 To address such questions, the Sensorimotor AI Journal Club is launching the "RL Debate Series"👇 w/ @EliSennesh, @Adam_Lowet, @no_reward_for_u
@TommSalvatori 🧵[1/5]
2
7
40
Terence Tao: "This is not a routine policy shift - it is a deliberate dismantling of the institutions, funding, and freedoms that have sustained American science for generations."
393
938
10K
A hallmark of human intelligence is the capacity for rapid adaptation, solving new problems quickly under novel and unfamiliar conditions. How can we build machines to do so? In our new preprint, we propose that any general intelligence system must have an adaptive world model,
14
103
509
Excited to share new work @icmlconf by Loek van Rossem exploring the development of computational algorithms in recurrent neural networks. Hear it live tomorrow, Oral 1D, Tues 14 Jul West Exhibition Hall C: https://t.co/zsnSlJ0rrc Paper: https://t.co/aZs7VZuFNg (1/11)
openreview.net
Even when massively overparameterized, deep neural networks show a remarkable ability to generalize. Research on this phenomenon has focused on generalization within distribution, via smooth...
2
21
69
I've been trying to ask Claude/ChatGPT to explain to me why this image is interesting and it does not get it.
1
0
10
"A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws".
Can an AI model predict perfectly and still have a terrible world model? What would that even mean? Our new ICML paper formalizes these questions One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵
9
26
191
Just a nightmare watching people take a hammer to amazing institutions we have built with barely an understanding of what they are. Our scientific infrastructure. Our global aid. Our ability to attract talent. Our sources of data.
9
18
224
As AI agents face increasingly long and complex tasks, decomposing them into subtasks becomes increasingly appealing. But how do we discover such temporal structure? Hierarchical RL provides a natural formalism-yet many questions remain open. Here's our overview of the field🧵
12
64
284
I don't know how consciousness works, but I just assume the mindset of already knowing how it works so that when I find out, I'll be able to act very chill about it.
1
0
5
Been struggling with a challenging proof for many months and finally finished it. Feels good 🫠
0
0
10
By the way, I visted Stoffel the Honey Badger because of the functional significance he plays in the structure of how I understand the world and for no other reason. No global value function in my head was updated, sorry! https://t.co/S1Fyp5gMQe
They say don’t meet your heroes, but I traveled to South Africa and met mine. Stoffel the Honey Badger became a major inspiration for my PhD thesis when my advisor showed our lab a BBC show on clever animals who can solve long horizon tasks, presumably for abstract reasons.
0
0
4
Strongly agree. My take that I will defend forever is that RL on reward-maximization will never be a coherent theory of intelligence because it is incompatible with a naturalistic theory of teleology needed to understand agency.
if such a reward function exists, then it could well be one level of *causal* explanation for my actions. but it would not be a good *teleological* explanation, at the level I understand myself, and want to be understood
1
4
17
This post is a rare articulation of an important outside perspective on AI Safety, which I think better accounts for a future which is open-ended and massively multi-agent. It effectively questions foundational philosophical assumptions which should be reconsidered
First LessWrong post! Inspired by Richard Rorty, we argue for a different view of AI alignment, where the goal is "more like sewing together a very large, elaborate, polychrome quilt", than it is "like getting a clearer vision of something true and deep" https://t.co/sIIpXk2nOk
1
6
29
Barandes' move is to more-or-less say "it's just a weird law of nature." That's pretty unsatisfying. I just have a casual curiosity in QM and maybe I don't fully get what he's saying. https://t.co/LanwaeMgTu
0
0
2
Jacob Barandes' theory is amazing. But isn't entanglement just as weird when QM is cast as an indivisible non-Markov stoch. proc.? There is still an exponential amount of non-Markov info determining observations. How does the universe carry this info?
1
0
2
My first time seeing a wild giraffe. Just chillin’ on the side of the road.
0
0
4