Lightnews — Scholar-powered news

Seth Karten

@sethkarten.ai

160 followers 270 following 140 posts

Autonomous Agents | PhD @ Princeton | World Gen @ Waymo | Prev: CMU, Amazon | NSF GRFP Fellow

Posts Media Videos Starter Packs

Pinned

Seth Karten @sethkarten.ai · Jul 23

🚀 New preprint!
🤔 Can one agent “nudge” a synthetic civilization of Census‑grounded agents toward higher social welfare—all by optimizing utilities in‑context? Meet the LLM Economist ↓

Diagram of LLM Economist: left—grid of persona‑conditioned worker agents; center—planner LLM sends tax schedule; right—social‑welfare ‘hill‑climb’.

Seth Karten @sethkarten.ai · 6h

You probably aren’t reading enough papers.
You probably didn’t cite the 10 closest papers to your work
Thus, LLMs probably have a better understanding of where your paper sits in the literature ¯\_(ツ)_/¯

Seth Karten @sethkarten.ai · 11d

We should have the highest standards for the most influential research companies

Seth Karten @sethkarten.ai · 13d

In the future people will play games for the mind similar to going to the gym for the body

Seth Karten @sethkarten.ai · 14d

If you arent paying attention, we are in a rapidly shifting period of ML paper culture. ICLR/ICML/NeurIPS are being treated as random, out of touch processes with more and more unnecessary work to submit
Most people are saying TMLR is the only good alternative, but are skeptical

Seth Karten @sethkarten.ai · 14d

The most interesting papers arent being published at the “prestigious” venues anymore. Where are you publishing and what do you work on?

Seth Karten @sethkarten.ai · Sep 2

Join our Discord for more info: discord.gg/E2DuX5FWF7

Join the PokéAgent Challenge @ NeurIPS 2025 Discord Server!

https://pokeagent.github.io/ | 362 members

Seth Karten @sethkarten.ai · Sep 2

🚨 Hackathon Weekend! 🚨

Jumpstart your PokéAgent Challenge submission ahead of NeurIPS!

📅 Sept 13–14
✅ Leaderboards reset Sat 10AM EDT
🎙️ Lightning talks in LLMs, RL, and Pokemon
💬 Live Office hours
🏆 $2k in prizes

PokéAgent Challenge @ NeurIPS 2025 Hackathon Weekend Schedule. Saturday, Sept 13th: 10 AM leaderboards reset; 12–1:30 PM livestream talks (overview, Aaron Traylor on Pokémon as an AI Problem, Seth Karten on Pokéchamp, Jake Grigsby on Metamon, plus more). Sunday, Sept 14th: 1–3:30 PM organizer office hours; 11:59 PM top teams earn up to $2k in GCP credits. Sponsored by Google DeepMind and AIJ.

Reposted by Seth Karten

Seth Karten @sethkarten.ai · Aug 15

The NeurIPS 2025 PokéAgent Challenge is offering compute credits, courtesy of our sponsor Google DeepMind, to help you train bigger models & run more experiments.

📌 To apply:
1️⃣ Make a submission to Track 1 or 2 at pokeagent.github.io
2️⃣ Fill out the compute credit form on the site

PokéAgent Challenge - NeurIPS 2025

pokeagent.github.io

Seth Karten @sethkarten.ai · Sep 1

We need an AI sports league and US robotics olympics

Seth Karten @sethkarten.ai · Aug 28

Does this include survey papers?

Seth Karten @sethkarten.ai · Aug 20

Eventually you could train this, but currently context switching at this general level is far too difficult. And I think the modular agentic system approach makes the most sense as this could scale to parallel running components

Seth Karten @sethkarten.ai · Aug 20

A foundation agent could be equally as modular with a harness to facilitate context switching. You could even wrap the harness in an llm harness to for routing which module to use

Seth Karten @sethkarten.ai · Aug 20

If you were deploying a robot irl, you would have
- a perception harness for visual understanding
- a planning harness for long horizon reasoning
- a control harness to make sure the actions are executed properly

Seth Karten @sethkarten.ai · Aug 20

The solution would generalize to another two player partially observable turn-based text game. The most bespoke items are tools, but there has been work recently that shows that you can make these tools modular LLM calls, further increasing generality

Seth Karten @sethkarten.ai · Aug 20

Interesting study. Though, are you in the model purist camp? (Give raw state. Let model do all the work.) Alternatively scaffolding/harness can do wonders improving the planning capabilities of LLMs: arxiv.org/abs/2503.04094

Seth Karten @sethkarten.ai · Aug 20

We did this at scale in the LLM Economist

arxiv.org/abs/2507.15815

LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra

We present the LLM Economist, a novel framework that uses agent-based modeling to design and assess economic policies in strategic environments with hierarchical decision-making. At the lower level, b...

Seth Karten @sethkarten.ai · Aug 18

Mad about data centers? Call your reps to build more nuclear

Seth Karten @sethkarten.ai · Aug 15

Hey #academics

Why are neurips workshop deadlines due a month before main track acceptances? Seems counterintuitive to have the two tracks compete with each other

#machinelearning

Seth Karten @sethkarten.ai · Aug 15

The NeurIPS 2025 PokéAgent Challenge is offering compute credits, courtesy of our sponsor Google DeepMind, to help you train bigger models & run more experiments.

📌 To apply:
1️⃣ Make a submission to Track 1 or 2 at pokeagent.github.io
2️⃣ Fill out the compute credit form on the site

PokéAgent Challenge - NeurIPS 2025

pokeagent.github.io

Seth Karten @sethkarten.ai · Aug 13

While the foundation model was not explicitly trained on the task, it can use its budget to adapt to the task distribution — meta-learning

Seth Karten @sethkarten.ai · Aug 13

Ok this is helpful to understand where people may get confused. Rather than giving a foundation model examples, one can have to foundation model explore, exploit, and improve all in-context (eg ICRL). This continuous improvement allows the foundation model to act as meta-learners

Seth Karten @sethkarten.ai · Aug 12

Whats your definition of each?

Seth Karten @sethkarten.ai · Aug 12

If your final product doesnt reason in-context, how is it supposed to meta-learn and address distribution shifts and environment changes?

Seth Karten @sethkarten.ai · Aug 12

Seth Karten @sethkarten.ai · Aug 12

Papers are dead. Maybe it is time to start the youtube channel…