Huge thanks to my advisors Chris L Buckley and Anil Seth, my mentors Ethan Perez, Marc Dymetman, Hady Elsahar, Germán Kruszewski and my external examiner David Krueger! I feel extremely privileged having had the opportunity to work with and learn from you.

You can (and should) do RL from human feedback during pretraining itself! In our new paper, we show how training w/ human preferences early on greatly reduces undesirable LM behaviors, including under adversarial attack, w/o hurting downstream performance. arxiv.org/abs/2302.08582

🚨 Preprint alert 🚨 I'm so happy to finally be able to share with you the VERSES AI Research white paper for research, development, and design in the field of artificial intelligence: 'Designing Ecosystems of Intelligence from First Principles'
arXiv link: arxiv.org/abs/2212.01354

Many congratulations Dr. Alec Tschantz Alec Tschantz for passing his PhD viva with flying colours. From Bayesian Principles to Bayesian Processes was examined by Andy Clark & Rosalyn Moran, & supervised by Chris L Buckley (mostly) & me. Well done 🙌🏽& your bottle🍾 now joins the ranks

Our Sussex Neuroscience 4-year PhD Programme is open for applications! Deadline Feb 23rd, for Sept 2023 entry. Brilliant community of students and supervisors, and a choice of >40 labs for rotations- come and join us by the sea! sussex.ac.uk/research/centr…

RL with KL penalties – a powerful approach to aligning language models with human preferences – is better seen as Bayesian inference. A thread about our paper (with Ethan Perez and Chris L Buckley) to be presented at #emnlp2022 🧵arxiv.org/pdf/2205.11275… 1/11

Postdoc and PhD available with me and Anil Seth on the EU project “A metapredictive model of synthetic awareness for enabling tool invention'. Would suit those interested in pred. proc. and generally work at the intersection of cog. sci and ML. Email [email protected]

Although I feel embarassed to admit that we made a mistake, I’m very grateful to Miguel Aguilera for looking through the preprint! This is a nice example of the benefits of preprinting one’s papers and having a community that fosters open channels of critical discussion :) 9/9

Joint pre-print with Beren Millidge, along with Anil Seth and Chris L Buckley. My first exposure to neuroscience was Anil Seth TED talk on the brain being 'top-down' - blew my mind and led me to research. End result - brain probably isn't 'top-down'
arxiv.org/abs/2204.02169

As noted in my old thread, all of this work was motivated by these three amazing papers that first discovered the relationships between PC and BP (James Whittington Beren Millidge Alec Tschantz Chris L Buckley):
direct.mit.edu/neco/article-a…
proceedings.neurips.cc/paper/2020/has…
arxiv.org/abs/2006.04182

Excited to announce our new preprint: “Hybrid Predictive Coding: Inferring, Fast and Slow”. This is joint work with Alec Tschantz, Chris L Buckley, and Anil Seth. A short thread on the paper arxiv.org/abs/2204.02169 1/11

#FreeEnergy what? one of the most interesting conferences this year. ⏰17Feb
🔴Free of charge register here 👇
thefreeenergyprinciple.weebly.com

#ML #AI #Robotics #Philosophy #Neuroscience #psychology

Are Friston blanket just, technically, the wrong tool?

Interested in Collective Intelligence in Animal Construction? Join the be.AI (sussex.ac.uk/ccnr/be_ai) doctoral training centre University of Sussex in Brighton to work with Paul Graham in SussexNeuro @SussexEASy. Applications open bit.ly/be-ai-apply until 23 Feb.

PhD positions are available in my group on all aspects of probabilistic inference and control in cognitive systems.
As part of the be.AI PhD scholarships in biomimetic embodied AI (sussex.ac.uk/ccnr/be_ai).
Details bit.ly/be-ai-apply, deadline 23 Feb.

Our be.AI (sussex.ac.uk/ccnr/be_ai) Leverhulme doctoral scholarship programme for biomimetic embodied AI is now officially open with 6 brilliant scholars. Recruitment for next year is about to start. sussex.ac.uk/staff/newsande…
University of Sussex @SussexEASy Brains on Board

💥We just reached a total of 4614 chapter downloads of the proceedings of the #ActiveInference Int. WS 2020 eBook.
👉link.springer.com/book/10.1007/9…

The new 2021 eBook will arrive soon! 🔝⏱️

We really need better models for producing and publishing science. That's why I'm thrilled to receive my first open peer review through @[email protected] :)

@ManuelBaltieri was so kind to share some insightful comments about our recent preprint abouth the FEP here

The IWAI2021
✔️best presentation award goes to:
Active Inference, Social Media and Mental Health - Mark Miller&Ben White Mark Miller

✔️best poster award goes to:
Habitual and Reflective Control in Hierarchical Predictive Coding - Paul Kinghorn, Beren Millidge Chris L Buckley

Announcement: I’m proud to announce Nested Minds will be sponsoring the biggest yearly event in the active inference community, IWAI. We're setting up an award ceremony, The Blankies, to recognize the impactful, original work of members of the active inference community 1/2