Dylan Cope

Researcher in multi-agent RL and Cooperative AI. Postdoc the University of Oxford @FLAIR_Ox. PhD from @safe_trusted_ai. ex intern @CHAI_Berkeley

[Shimon Whiteson @shimon8282]

I'm hiring a Research Engineer for my multi-agent learning team at DeepMind: https://t.co/Z9yMrm148s

[Dylan Cope @DylanRobertCope]

@BlackHC I can understand it at a basic level, but it is going to cause problems in academia in particular. It's not exactly clear how/when the safeguards are triggered so even a basic ML researcher or an alignment researcher at a university might be nerfed.

[Silvia Sapora @silviasapora]

After interviewing for Research Scientist roles at DeepMind, Isomorphic, Meta, Cohere and more, I wrote up everything I learned. Technical prep, logistics, negotiation, and emotional breakdowns. Check out my guide: https://t.co/eLh20ggMHW

[Clarisse Wibault @ClarisseWibault]

Mean Field Games provide a framework for modelling large populations.
ICML26 Spotlight: Introducing Recurrent Structural Policy Gradient for partially observable MFGs with common noise, benefitting from faster convergence than model-free RL, but remaining tractable, unlike DP.

[Michael Matthews @mitrma]

Hindsight Experience Replay has become the ubiquitous method for goal-conditioned reinforcement learning, but leaves open the question of which goal to relabel with.

In this work, accepted at ICML, we propose instead simply Learning Everything All at Once (LEO).
1/

[Clarisse Wibault @ClarisseWibault]

CV has CNNs, NLP has transformers - what inductive bias does RL have? How can policies generalise to regions of the dataset suffering from poor transitions?

We motivate hierarchy by enabling distinct state-representations at different levels of the hierarchy @FLAIR_Ox @j_foerst

[Uljad @uljadb99]

Natural evolution's open-endedness leads to beautiful, complex emergent structures and self-organizing behavior 🌱✨. Replicating this in silico is famously hard 💻. Our paper points to a promising direction by evolving populations of competing neural cellular automata with lifelike behavior 🧬🤖 #Isambard

⚠️⚠️flashing lights, rapid cuts, or strobe effects in this thread! 🚨🚨

1/n

[Alex Goldie @AlexDGoldie]

1/ 🪩 Automating the discovery of new algorithms could unlock significant breakthroughs in ML research. But optimising agents for this research has been limited by too few tasks to learn from!

Introducing DiscoGen, a procedural generator of algorithm discovery tasks 🧵

[Dylan Cope @DylanRobertCope]

What exactly is the problem here? They frankly overpay to subsidise our education system, which really should be funded more by the government. A system that the conservatives all-too-happily commercialised.

[Neil O'Brien @NeilDotObrien]

"Oxford, Cambridge, Imperial, Manchester and University College London — are collectively enrolling about five Chinese Stem postgrads for every four Brits.
In engineering, there are some 3,300 Chinese postgrads versus 1,900 Brits; in maths, 700 Chinese versus 500 Brits."
MAD

[Zilin Wang @zilinwang4ai]

1/ 🚗 🌏 What if an autonomous vehicle could move to a new city without collecting a single human demonstration in that city?

I am so excited to introduce our new work: Learning to Drive in New Cities Without Human Demonstrations.

[Dylan Cope @DylanRobertCope]

@LukeTryl Fair enough. There isn't really a "right" thing to show, but imo this data is dominated by the overall popularity of each party. I would be more interested with that removed to more clearly see the relationship between board game preference and party preference.

[Alonso Gurmendi @Alonso_GD]

“Genes determine skin colour”
Yes they also determine height, moles, and hair colour. But we dont build social hierarchies around that. That doesn’t mean races are genetic. It means you ring fenced some genes and attributed moral/social value to them. We call this “racism”

[Dylan Cope @DylanRobertCope]

I suspect most of the people excited about this don't write or review papers themselves. LLMs can be useful writing assistants, but to me writing is still a form of thinking and I don't really want AI so prevalent in my thinking process.

[Robert Youssef @rryssf]

Holy shit… this might be the most unreal academic-writing upgrade I’ve ever seen 🤯

A team from NUS just dropped PaperDebugger an in-editor, multi-agent system that lives inside Overleaf and rewrites your paper with you in real time.

Not copy-paste. Not a sidebar chatbot.

Actual agentic editing inside your LaTeX editor.

Here’s why this is insane 👇

→ You highlight a messy paragraph, and it launches a full critique + rewrite pipeline
→ Returns clean before–after diffs like Git, then patches your document instantly
→ Runs Reviewer, Enhancer, Scoring, and Researcher agents in parallel
→ Uses Kubernetes pods to scale multi-agent reasoning inside the editor
→ Taps an MCP toolchain for literature search, reference lookup, and section-level enhancement

Deep research mode is even crazier:

It pulls relevant arXiv papers, summarizes them, compares your method against them, and generates citation-ready tables… all inline while you're writing.

It’s basically a mini committee of reviewers embedded in your document rewriting, critiquing, sourcing, and polishing without ever breaking flow.

If this scales, Overleaf stops being an editor… and becomes a full AI-assisted research environment.

[Dylan Cope @DylanRobertCope]

@carolownsu @bloodylikeab0dy Most Americans don't use electric kettles because the voltage of their electricity means that water takes way longer to boil (can be 2x as long as the UK). So even though its super energy inefficient and unsafe microwaves are used instead.

[Bidipta Sarkar @bidiptas13]

Introducing 🥚EGGROLL 🥚(Evolution Guided General Optimization via Low-rank Learning)! 🚀 Scaling backprop-free Evolution Strategies (ES) for billion-parameter models at large population sizes

⚡100x Training Throughput
🎯Fast Convergence
🔢Pure Int8 Pretraining of RNN LLMs

[Dylan Cope @DylanRobertCope]

Oh no! My fiction series has unrepresentative statistics!

This is a great case study in manufacturing statistics to lend an aesthetic of scientific objectivity to obvious racism. Watching these people twist numbers for literal fiction highlights the absurdity of thinking.

[Stuttering Craig (Official) @StutteringCraig]

Update: I looked closer - it’s actually two Black men and a Black woman, not just two men.

So I re-ran the math.
With that correction, the odds of this exact group (Asian woman + 2 Black men + 1 Black woman + one-legged man) randomly walking together in 1880 Chicago?

About 1 in 2.4 million.

Netflix didn’t just bend history - they straight-up violated statistics.

[Stuttering Craig (Official) @StutteringCraig]

Update: I looked closer - it’s actually two Black men and a Black woman, not just two men.

So I re-ran the math.
With that correction, the odds of this exact group (Asian woman + 2 Black men + 1 Black woman + one-legged man) randomly walking together in 1880 Chicago?

About 1 in 2.4 million.

Netflix didn’t just bend history - they straight-up violated statistics.

[Dylan Cope @DylanRobertCope]

Why do EAs think that they have a monopoly on talking about certain problems. Thinking that "altruism" is the only lens through which to see global health problems is ridiculous.

[Matt Reardon @Mjreard]

We interrupt @hankgreen's AI Safety week to bring you Hank’s ITN analysis of global health problems, but don't worry guys, he's definitely "not an EA"(!)

[Dylan Cope @DylanRobertCope]

@YaBoiClear @BerntBornich Are you okay?