Cambridge MLG (@CambridgeMLG)

MLG alumni 🤝 Twitter beef

as is tradition.

Bruno Mlodozeniec Verified account @brunorganised Nov 6

I’m usually a big fan of a @giffmana takedown, but “this is well known” misses the core new things in our work:

The 2022 paper is super relevant and we should have cited it. Agreed the E/α ambiguity in scaling laws is itself known, the paper doesn't claim to discover it. But the linked works don't give you a way to answer the questions: for a given ladder, is my fit actually near-optimal? How wide is the range of predictions from all near-optimal fits?

We contribute both:
1) Algorithms that provide provably guaranteed answers to those questions.
2) Using these algorithms, we quantify how suboptimal the fits of a standard approach people rely on are (Chinchilla Approach 3), and how much of a difference that can make when using the resulting scaling law for prediction.
View quoted post on X

Cambridge MLG @CambridgeMLG Sep 7

Rich Turner is hiring a Research Associate at Cambridge to work on AI for weather forecasting 🌥️

Strong deep learning background required, but no prior weather experience is fine. Especially relevant: spatiotemporal modelling, PDEs, time series

Cambridge MLG @CambridgeMLG Jul 27

Congratulations to @katie_m_collins on this amazing nature paper!

Katie Collins Verified account @katie_m_collins Jul 17

Games have long served as a testing ground for understanding problem-solving and strategic reasoning in humans and machines, across many fields – from AI, to cognitive science, economics, and more. Yet, in life people are regularly faced not just with repeated instances of the same problem or “system of rule and reward”, but entirely new problems, new systems, new games.

And when encountering new games, people not only need to decide what actions to take but whether a game is even worth playing at all!

How can people think reasonably about new games for the first time, so flexibly and quickly?

We take steps towards this question in our new work, out now in @Nature!

View quoted post on X

Cambridge MLG @CambridgeMLG Jun 7

We’re willing to believe it could produce what passes for a paper over at Oxford.

How To Prompt Verified account @HowToPrompt__ Jun 6

Claude Code can now run an entire PhD-level research pipeline by itself.

it runs a 10-stage workflow from blank page to publication-ready PDF, replacing the work of a PhD advisor, three peer reviewers, and a copy editor in one repo.

→ Deep research with 13 agents (PRISMA + systematic review)
→ 12 agents write the paper section by section
→ 5-person peer review (Editor + 3 Reviewers + Devil's Advocate)
→ Integrity agent catches fabricated citations + stat errors
→ Final output: LaTeX → PDF, ready to submit

After the paper is finalized, it runs a Collaboration Quality Evaluation that scores YOU, across 6 dimensions, 1–100. Direction setting, intellectual contribution, quality gatekeeping.

It tells you exactly where you were the bottleneck.

Drop it into .claude/skills/ and the whole pipeline auto-loads. Works in Claude Code, Cowork, and as a Claude Project.

100% open source. CC-BY-NC 4.0.

View quoted post on X

Cambridge MLG @CambridgeMLG Dec 20, 2025

Work by our Master student featured as a NeurIPS workshop spotlight

Steve Hong @steve_steve_hg Dec 20, 2025

Excited to share our work “Better Hessians Matter: Studying the Impact of Curvature Approximations in Influence Functions,” presented as a Spotlight at the NeurIPS 2025 Mechanistic Interpretability Workshop earlier this month :) ! 🧵👇 Paper: arxiv.org/abs/2509.23437
View quoted post on X

Cambridge MLG

Latest posts 5