TalkRL Podcast

TalkRL Podcast is All Reinforcement Learning, All the Time.
Follow for interviews with brilliant folks from across the world of RL.
Host @robinc. DMs open.

[TalkRL Podcast @TalkRLPodcast]

E73: @danijarh (ex-@GoogleDeepMind Research Scientist) in-depth on Dreamer v4.
Audiogram edition 📡

[TalkRL Podcast @TalkRLPodcast]

@kjaved_ @how_uhh @velocizapkar @danijarh Agree this seems to defeat the purpose of small sample benchmarks!

Still hoping there is another solution to this issue other than slow envs... seems a case of goodhearts law

[TalkRL Podcast @TalkRLPodcast]

@kjaved_ @how_uhh @velocizapkar @danijarh Purposely wanting slow envs due to wanting sample efficient algos seems a bit like throwing baby out with bathwater. Then those with most compute have more advantage.
Why not just pay attention to sample complexity/hp sensitivity, and also have fast envs?

[TalkRL Podcast @TalkRLPodcast]

@how_uhh @kjaved_ @velocizapkar Or maybe... Craftax?
(Accelerated variant of @danijarh 's Crafter)

[TalkRL Podcast @TalkRLPodcast]

Kai was our guest back on Episode 11.
https://t.co/v6Qvv9Kidi

[TalkRL Podcast @TalkRLPodcast]

Glad to meet up with @kaixhin , research scientist at @SakanaAILabs, for lunch in Tokyo!

[TalkRL Podcast @TalkRLPodcast]

@danijarh Always a pleasure Danijar! 🙏

[TalkRL Podcast @TalkRLPodcast]

Apple
https://t.co/aVfIauHuzu

[TalkRL Podcast @TalkRLPodcast]

E73: Danijar Hafner on Dreamer v4
@danijarh (ex-@GoogleDeepMind RS) on offline world models for safe robotics, Shortcut Forcing for fast diffusion video models, outperforming OpenAI’s VPT with 100× less data, his “APD” theory unifying exploration and empowerment, and more!

[TalkRL Podcast @TalkRLPodcast]

Spotify
https://t.co/fLd60Kn21D

[TalkRL Podcast @TalkRLPodcast]

@jesswhittles Always enjoy your writing!

Is there some tension between integrity and diplomacy (which is often less about integrity than interests)?

[TalkRL Podcast @TalkRLPodcast]

@jaseweston @jackjingyuzhang @ericsmithnyc @DanielKhashabi @ben_vandurme Congrats!
Would you be open to a short interview on this?

[TalkRL Podcast @TalkRLPodcast]

@sirbayes @alexinch_ai @karpathy @RichardSSutton @dwarkesh_sp And how/if it differs on this level, after pre-training vs after post-training.

[TalkRL Podcast @TalkRLPodcast]

@sirbayes @alexinch_ai @karpathy @RichardSSutton @dwarkesh_sp Its interesting how central this is to current paradigm, yet how non obvious to most (including me).
Has this formulation been more spelled out somewhere by you or others?

[TalkRL Podcast @TalkRLPodcast]

@snowclipsed Just to be safe, lets start with n↑↑4

[TalkRL Podcast @TalkRLPodcast]

@jackjingyuzhang Congrats!
Would you or another co-author be open to speak about it on TalkRL Pod?

[TalkRL Podcast @TalkRLPodcast]

@NielsRogge @tw_killian True
https://t.co/HqTaeHZriq

[TalkRL Podcast @TalkRLPodcast]

@CsabaSzepesvari @karpathy My personal hot take is very different:
1. RL as a family of conceptual frameworks, is timeless.
2. Frustrations with modern deep RL algo performance, are mostly due to limitations of deep learning function approx
tldr; Give RL FAs that generalize better (plus algos) :D

[TalkRL Podcast @TalkRLPodcast]

@pmddomingos @CsabaSzepesvari @karpathy @RichardSSutton https://t.co/g2m6UlzZgr

[TalkRL Podcast @TalkRLPodcast]

@CsabaSzepesvari @pmddomingos @karpathy @RichardSSutton To your point: Sutton and Barto 2nd Ed, opening statements Sec 1.1 pp 2

[TalkRL Podcast @TalkRLPodcast]

@CsabaSzepesvari @pmddomingos @karpathy @RichardSSutton To your point: Sutton and Barto 2nd Ed, opening statements Sec 1.1 pp 2