Lightnews — Scholar-powered news

Reposted by Minsuk Chang

Chris Warren @chrisvvarren.bsky.social · 20d

Trying to keep my professional chill but I’m SO excited Carnegie Mellon is launching a cluster hire in computational humanities—MULTIPLE JOBS!

1. Asst Teaching Track Prof in Computational Humanities - apply.interfolio.com/173622
2. Asst Tenure Track Prof in CH - apply.interfolio.com/173626

3 42 84

Reposted by Minsuk Chang

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 20d

Our NeurIPS submission arxiv.org/abs/2502.08938 did not get in, but it's one of my favorite papers and I think one of the better papers we've ever put out so I want to highlight it

Reevaluating Policy Gradient Methods for Imperfect-Information Games

In the past decade, motivated by the putative failure of naive self-play deep reinforcement learning (DRL) in adversarial imperfect-information games, researchers have developed numerous DRL algorithm...

arxiv.org

5 8 74

Reposted by Minsuk Chang

Edward Grefenstette @egrefen.bsky.social · Jul 21

Do you have a PhD (or equivalent) or will have one in the coming months (i.e. 2-3 months away from graduating)? Do you want to help build open-ended agents that help humans do humans things better, rather than replace them? We're hiring 1-2 Research Scientists! Check the 🧵👇

3 6 19

Reposted by Minsuk Chang

Bret Victor @worrydream.com · Jun 29

Archive of archives.
worrydream.com/Links2025/

Links 2025

worrydream.com

6 44

Reposted by Minsuk Chang

luokai @luok.ai · Jun 26

Claude Code, meet your rival: Gemini CLI 🔥🚀

Google just unleashed Gemini 2.5 Pro at the command line—free, open-source (Apache 2.0) , 1-million-token brain baked right into your terminal. Chat, code, search, script—all with zero tab fatigue.

1 4 24

Reposted by Minsuk Chang

Markus Wulfmeier @mwulfmeier.bsky.social · Jun 24

Don't have a robot? Try our newest Gemini Robotics on-Device VLA in simulation!

Or become a trusted tester and tune and adapt the model yourself!

www.youtube.com/watch?v=nVMY...

Gemini Robotics on-device in simulation

YouTube video by Google DeepMind

www.youtube.com

1 1 5

Minsuk Chang @minsuk.bsky.social · Jun 24

I really think ACs can be completely replaced by LLMs. Finding the right reviewers, sending (reminder) emails, nudging them for discussions, summarizing reviews, etc... The only issue is that LLM calls are expensive, whereas human ACs work for free because we all volunteer. Sigh.

1

Reposted by Minsuk Chang

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · Jun 21

Hiring a postdoc to scale up and deploy RL-based planning onto some self-driving cars! We'll be building on arxiv.org/abs/2502.03349 and learn what the limits and challenges of RL planning are. Shoot me a message if interested and help spread the word please!

Full posting to come in a bit.

Robust Autonomy Emerges from Self-Play

Self-play has powered breakthroughs in two-player and multi-player games. Here we show that self-play is a surprisingly effective strategy in another domain. We show that robust and naturalistic drivi...

arxiv.org

4 24 60

Reposted by Minsuk Chang

Jeffrey P. Bigham @jeffreybigham.com · Jun 22

I work in the best university in the world, less than 3 hours from my rural hometown, yet superstar students travel here from all over the world… if we keep treating them badly, will centers of excellence move elsewhere?

"If my kids excel, will they move away?"

jeffreybigham.com/blog/2025/wh...

In Pittsburgh, I’m a professor at Carnegie Mellon University in the top computer science school in the world. I’ve also worked in various large technology companies, who have offices in Pittsburgh to connect with and employ Carnegie Mellon faculty and students.

jeffreybigham.com

9 43

Reposted by Minsuk Chang

Ethan Mollick @emollick.bsky.social · Jun 14

👀This is impressive: Using an autonomous agent based on o3-mini and GPT-4.1, a team from Harvard, MIT & other institutions reproduced and updated an entire issue of Cochrane Reviews in two days… saving 12 person-years of work.

The AI reviews captured more papers & were more accurate than humans.

2 27 160

Reposted by Minsuk Chang

Thomas Dietterich @tdietterich.bsky.social · Jun 3

Yes!! Originally, generative ML modeled a joint distribution, e.g., P(X,Y), whereas discriminative ML modeled a conditional distribution, e.g., P(Y|X). A famous paper by Ng and Jordan (papers.nips.cc/paper_files/...) compared naive Bayes (generative) to logistic regression (discriminative). 1/

On Discriminative vs. Generative Classifiers: A comparison of logistic regression and naive Bayes

papers.nips.cc

3 3 14

Reposted by Minsuk Chang

Christopher Barrie @cbarrie.bsky.social · Jun 2

🤖💡 Are LLMs really developing social behaviors—or just reproducing familiar patterns from training? In our new paper, @pettertornberg.com and I take a closer look at recent claims of emergent conventions in AI and offer a different perspective... 🧵

6 26 92

Reposted by Minsuk Chang

Chris Paxton @cpaxton.bsky.social · May 29

Weave household robot cleaning up. This is teleop, but it's a really cool first step towards a commercial product.

11 7 60

Reposted by Minsuk Chang

Pierre-Yves Oudeyer @pyoudeyer.bsky.social · May 28

Humans' ability to invent their own games & goals is at the core of open-ended learning.

Understanding and modeling computationally how they do it would be enlightening to understand better human cognition and build open-ended AI

Great step in this direction in new paper by Guy Davidson et al.

1 4 18

Minsuk Chang @minsuk.bsky.social · May 24

TPUs are amazing.

3

Minsuk Chang @minsuk.bsky.social · May 23

TIL Canada has its own professional men's basketball league and Vancouver has a team.

1

Reposted by Minsuk Chang

Matej Balog @matejbalog.bsky.social · May 14

Announcing AlphaEvolve, our new LLM coding agent that has
- made new scientific discoveries
- discovered algorithms that are now deployed at Google (in Gemini, Transformers, TPU hardware design & data centers)

Blog: deepmind.google/discover/blo...
White paper:
storage.googleapis.com/deepmind-med...

AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms

New AI agent evolves algorithms for math and practical applications in computing by combining the creativity of large language models with automated evaluators

deepmind.google

5 40 120

Reposted by Minsuk Chang

David Pfau @davidpfau.com · May 9

The idea of "AI alignment" grew out of a community that thought you could solve morals like it was a CS problem set. Nice to see a more nuanced take.

Joel Z Leibo @jzleibo.bsky.social · May 9

Announcing our latest arxiv paper:

Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt
arxiv.org/abs/2505.05197

We argue for a view of AI safety centered on preventing disagreement from spiraling into conflict.

Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt

Artificial Intelligence (AI) systems are increasingly placed in positions where their decisions have real consequences, e.g., moderating online spaces, conducting research, and advising on policy. Ens...

arxiv.org

3 25

Reposted by Minsuk Chang

Jun-Yan Zhu @junyanz.bsky.social · May 10

[1/2] We've released the code for LegoGPT. Our autoregressive model generates physically stable and buildable designs from text prompts by integrating physics laws and assembly constraints into LLM training and inference.

Code: github.com/AvaLovelace1...
Website: avalovelace1.github.io/LegoGPT/

4 24 70

Minsuk Chang @minsuk.bsky.social · May 6

Not exactly the same, but something similar www.hmart.com/seafood-coin...

Seafood Coin Stock 3.17oz(90g) - The Best of Asia in America

www.hmart.com

1 2

Minsuk Chang @minsuk.bsky.social · May 3

Coach Pop, please write a book.

1

Reposted by Minsuk Chang

Stan Bileschi @bileschi.com · May 3

:thumbs_up:
robhorning.substack.com/p/contentment

Contentment

"social media" has always been oxymoronic

robhorning.substack.com

1 2

Reposted by Minsuk Chang

Andrew Lampinen @lampinen.bsky.social · May 2

How do language models generalize from information they learn in-context vs. via finetuning? In arxiv.org/abs/2505.00661 we show that in-context learning can generalize more flexibly, illustrating key differences in the inductive biases of these modes of learning — and ways to improve finetuning. 1/

arxiv.org

4 21 78

Reposted by Minsuk Chang

Mark Riedl @markriedl.bsky.social · May 1

This is a fantastic oral history of the last 10 years of NLP and AI. www.quantamagazine.org/when-chatgpt...

When ChatGPT Broke an Entire Field: An Oral History | Quanta Magazine

Researchers in “natural language processing” tried to tame human language. Then came the transformer.

www.quantamagazine.org

2 31 95

Minsuk Chang @minsuk.bsky.social · Apr 22

Should I get an Amazon Astro?

1 1