Diego Gomez

AI researcher @rlai_lab

[Diego Gomez @Sisyfuzz]

@yoavgo I don't see the contradiction in quantum field physicists describing astronomical behavior while modelling the world as isotropic and full of perfect oscillators. Simple rules lead to complex behavior. Identifying sets of simple rules is the whole point.

[Diego Gomez @Sisyfuzz]

@MartinKlissarov @GoogleDeepMind @egrefen Congrats! Hope you have fun.

[Marlos C. Machado @MarlosCMachado]

The recording of the talk I gave at the @CIFAR_News DLRL Summer School this year at @VectorInst in Toronto is now publicly available:

I discussed the interaction between RL and games (with many historical notes).

RL and Games: https://t.co/tZZ9QLDoe5

[Marlos C. Machado @MarlosCMachado]

For those interested, the keynotes of the @RL_Conference 2024 are now available online: https://t.co/PvFBsamvoI

Unfortunately, Doina Precup's talk was not recorded, but we have: Andy Barto, @EmmaBrunskill, @FinaleDoshi, @svlevine, David Silver, and @PeterStone_TX.

[Marlos C. Machado @MarlosCMachado]

Now that @RL_Conference is over (and what a success it was!), it's time to start thinking about the next one.

If you haven't heard yet, RLC is going north. We'll be organizing RLC'25 in Edmonton! I'm very excited about that and look forward to seeing everyone there.

[Dylan Foster 🐢 @canondetortugas]

Now that I have started using twitter somewhat regularly, let me take a minute to advertise the RL theory lecture notes I have been developing with Sasha Rakhlin: https://t.co/x16aGvE4tr

[Diego Gomez @Sisyfuzz]

@desariky Well, that depends on the representation learning abilities of the agent. RNNs can encode this type of states. But I get your point. It is interesting that you can trade 'state representation complexity' with 'reward complexity'!

[Diego Gomez @Sisyfuzz]

@sarahookr @tilmanbayer @RichardSSutton Rich couldn't care less about neural networks. It's not about making things big. It is about making things that won't become obsolete once there is access to more computation.

[Michael Levin @drmichaellevin]

Final version is out, on the topic of mechanisms by which small cognitive subunits can scale up into larger collective intelligences:
"Stress sharing as cognitive glue for collective intelligences: A computational model of stress as a coordinator for morphogenesis"
https://t.co/l6er8boe0F

@lakshwin_

[Diego Gomez @Sisyfuzz]

@davidbessis @estictry Biologists have trouble defining what is life and biology is the study of life. So...

[Finding The Frame Workshop @RLFrameWorkshop]

🚨@RL_conference's Finding the Frame Workshop submission deadline is coming up on May 15! 🚨
Help us re-think the conceptual foundations of RL, that often go unchallenged. Tell us how YOU think problems should be framed. Create the next paradigm shift✨🚀
https://t.co/bYbIW3QKB7

[Diego Gomez @Sisyfuzz]

@AravSrinivas @karpathy This is precisely what intelligence is...

[Diego Gomez @Sisyfuzz]

@nanjiang_cs @EugeneVinitsky Hardly control if your model is a generative NN

[Diego Gomez @Sisyfuzz]

@filippie509 @benjaminjriley This is assuming there is sth special about language as a state representation of the world. Yet, similar machine learning models can model sequences of words, abstract functions, or signals from physical systems. Language encodes universal mathematical properties of "reality".

[Diego Gomez @Sisyfuzz]

@filippie509 @Philosimplicity So, what I'm hearing is that an LLM with the ability to interact with the world (an RL agent, essentially) can result in an AGI.

[Diego Gomez @Sisyfuzz]

@s_batzoglou @Plinz It's more of an ontological axiom. There is no solution to the problem of induction. I think in any case is more productive to think about the predictive and control power of models, as opposed to be searching for some ultimate parametric model that explains everything.

[Diego Gomez @Sisyfuzz]

@s_batzoglou @Plinz That's because there is no truth. Just observations and models.

[Diego Gomez @Sisyfuzz]

@denny_zhou @ZhangLunjun Everything is a game

[Diego Gomez @Sisyfuzz]

@pmddomingos @tunguz Symmetry has little to nothing to do with concentration in Markov chains