Tickets for UKNC25 Imperial College London close tomorrow 5pm BST! Register: neuralcomputation.uk/registration.h…

Microsoft claims their new AI framework diagnoses 4x better than doctors. I'm a medical doctor and I actually read the paper. Here's my perspective on why this is both impressive AND misleading ... 🧵

New preprint! Why are long-range connectomic interactions in the cortex dominant in shaping dynamics in some experiments but apparently negligible in others? We (w/ Rishi M, Eli Muller) address this by studying a new hybrid model of cortical dynamics. arxiv.org/abs/2506.19800

Excited to share Energy-Based Transformers (EBTs), which allows you to implement system 2 thinking in any modality! EBTs formulate reasoning as an energy optimization problem, allowing models to internally think without complexities like CoT or multiple recurrent latents.

I’m excited about this: 12 month fellowship for AI engineers to work on some of the biggest challenges in government👇

Can transformers analyze code efficiently? ✅ Yes. We prove transformers efficiently handle real compiler tasks (AST construction, symbol resolution, type infer) using only log size—while RNNs require linear size (in input length). Paper: arxiv.org/abs/2410.14706 #COLM2025

this was a blast — great job Danyal Mark Humphries Dan Goodman (and yes, london was legitimately hot enough that I wore shorts while giving a talk)

UK stand up tall!

I'm hiring a @ARIA_Research funded position at Imperial College London @ImperialEEE. Broad in scope + flexible on topics! Exciting projects on intersection of neural networks & new AI accelerators, from a HW/SW co-design perspective. Join us 🚀: imperial.ac.uk/jobs/search-jo…

thread on the new paper: The Serial Scaling Hypothesis joint work with: @phizaz, Yutong Bai, Kananart

Towards Greater Leverage: Scaling Laws for Efficient Mixture-of-Experts Language Models Impressive results. This resolves various problems around MoE all at once. First it reconfirms that MoE has a higher ratio of optimal data size relative to computational cost compared to

*New Paper* How genes for IQ shape brain organisation. Amazing work led by Alicja Monaghan MRC Cognition and Brain Sciences Unit reveals 2 ways. First, they’re associated w/ the ‘costs’ of network formation. Second, they define ‘high value’ areas within the network. How she did it…🧵 doi.org/10.1162/IMAG.a…

I have a new example why Multi-Agent / Zero-shot coordination matters in the real world for my slides! Interestingly, this problem is going to get worse rather than better as we deploy more autonomous vehicles: Currently humans act as a robust regularizer of the system and AVs

Genie 3 is here - it can generate an entire world simulation that you can interact with in real-time, just from a text prompt! It's pretty mind-blowing really when you stop to think about it, and it's rapidly improving - one day we will be able to build the Holodeck for real!

Wow, reviewing ARIA CEO applications and it’s a ludicrously strong field. Ilan Gur and the team have built a truly special organisation - a culture and team as ambitious and relentless as anything I’ve seen in startups - and the future looks very bright 👏👏

Good piece this. Always interesting to me that, 1/ UK does so well at attracting data centre investment despite incredibly high industrial electricity price. 2/ London and surroundings is so attractive despite high land price and planning barriers. bbc.co.uk/news/articles/…