Christopher Summerfield @summerfieldlab@qoto.org

I found this to be a very surprising result. great paper.

https://www.sciencedirect.com/science/article/abs/pii/S0010027723000112?dgcid=coauthor

our review article on continual learning in natural and artificial agents is now out in TINS! https://www.sciencedirect.com/science/article/pii/S0166223622002600.
with @SaxeLab and Timo Flesch

this is a great project: https://www.biorxiv.org/content/10.1101/2023.01.16.523429v1
congratulations
@KrisJensen

random thought...
it's nearly 4 years since Rich Sutton wrote his Bitter Lesson blogpost: http://www.incompleteideas.net/IncIdeas/BitterLesson.html
he wrote before the explosion of interest in large transformer models. His claim is that "maximally general methods are always better". Maximally general means two things: 1) avoid human priors about cognition; 2) avoid human training data. It's interesting to reflect that he was both very right and very wrong.
- very right, that simple algorithms massively scaled can give you (decent) systematicity, without the need for symbolic bells and whistles. That's not to say that there aren't still important ingredients of intelligence missing in large transfomer models. But the level of composition you get in large generative models is much more impressive than most of us predicted. So he was broadly right about (1)
- very wrong, because the really impressive stuff relies on volumes of human feedback. As RLHF comes to the fore, it's become clear that self-play is only going to work for a very narrow set of problems. You need human data in spades.

For anyone who might consider doing a PhD in Granada, Spain this is an excellent opportunity: https://websepex.com/2023/01/14/seis-ofertas-de-contratos-predoctorales-en-el-cimcyc-universidad-de-granada/
Join a great team in one of the most beautiful cities in the world :)

re: ChatGPT as a co-author...there is nothing new under the sun. When Newell and Simon wrote up their work on the Logic Theorist in 1955 (which found a novel proof for one of the theorems in Russell's Principia Mathematica), they listed the AI as a co-author.
Also nothing new under the sun: the paper was rejected for being insufficiently novel.

happy to share this new preprint - we asked how goals warp the representations of allocentric space in human BOLD signals
https://www.biorxiv.org/content/10.1101/2023.01.12.523762v1.full.pdf

with a great team included @hugospiers @nicoschuck and lead authors Paul Muhle-Karbe and Hannah Sheahan

I am grateful to Paul Middlebrooks (not here yet) for inviting me to talk about my book on Brain Inspired today! looking forward to the discussion

OK, finally giving up on twitter. looking forward to the discussion here!