Tessera
Folgen

Benjamin Laufer

@laufer.bsky.social

PhD student at Cornell Tech. bendlaufer.github.io

67 Posts279 Follower200 Folgt
Benjamin Laufer@laufer.bsky.social · 7d ago

In a new @fastcompany.com op-ed, I connect recent developments involving Anthropic, Kimi K3, and the EU AI Act to a question from my research: how do rules aimed at one company change the behavior of others in the AI supply chain? www.fastcompany.com/91580189/sho...

Should AI companies be able to outsource safety?

Rules aimed only at downstream applications can make AI products less safe. Policymakers should hold both model makers and the companies building on them accountable.

fastcompany.com

Benjamin Laufer@laufer.bsky.social · 2w ago

One intuition behind many AI policy proposals is that downstream AI applications -- the companies deploying AI in healthcare, finance, education, customer service, etc. -- should bear responsibility for ensuring safety. Our paper asks: What incentives does that create for the firms building AI?

Illustrative example of our game-theoretic model. This numeric instance of the game consists of one general-purpose producer and three domain-specialists. Each player has a different utility in performance-safety space which dictates the path of development. Within this setting, the no-regulation game (Upper Left) reveals the players’ investment efforts when no floor is imposed on safety. Regulating the domain-specialist alone (Upper Right) exhibits backfiring for all three domains, meaning the regulated safety level is lower than it would be without regulation. In this particular example, the same floor is assumed for all three domain-specialists. Regulating the generalist alone (Lower Left) improves the safety level slightly across all three domains, compared to no-regulation. Finally, a regime that targets both generalists and specialists with regulation (Lower Right) is able to 1) retain the improved safety performance from regulating the generalist, 2) improve the safety level of least-safe domain-specialist, while 3) avoiding backfiring. The purpose of this figure is to visualize the model’s incentive mechanisms; none of these panels represent real regulations.
Rafael M Batista@rafmbatista.bsky.social · 10mo ago

Saw Ben presenting this today. It’s really neat work. Ben is finishing up his PhD at Cornell (advised by Jon Kleinberg) and is currently on the job market

Benjamin Laufer@laufer.bsky.social · 12mo ago

In a new paper with @didaoh and Jon Kleinberg, we mapped the family trees of 1.86 million AI models on Hugging Face — the largest open-model ecosystem in the world. AI evolution looks kind of like biology, but with some strange twists. 🧬🤖

The 2500th, 250th, 50th, and 25th largest model families on Hugging Face. They show varying numbers of generations (between 3 and 8) and different edge types, including adapters, finetunes, merges, and quantizations.
Marcelo Rinesi@marcelorinesi.bsky.social · 12mo ago

This is quite clever and useful (read the full thread + the paper). I think/hope it opens up the path to a parallel study of their evolution on the epistemic/semantic space (i.e. what things they get better/worse at over time, what the utility gradients... 1/ via @tedunderwood.me

Benjamin Laufer@laufer.bsky.social · 12mo ago

In a new paper with @didaoh and Jon Kleinberg, we mapped the family trees of 1.86 million AI models on Hugging Face — the largest open-model ecosystem in the world. AI evolution looks kind of like biology, but with some strange twists. 🧬🤖

The 2500th, 250th, 50th, and 25th largest model families on Hugging Face. They show varying numbers of generations (between 3 and 8) and different edge types, including adapters, finetunes, merges, and quantizations.
Benjamin Laufer@laufer.bsky.social · 12mo ago

In a new paper with @didaoh and Jon Kleinberg, we mapped the family trees of 1.86 million AI models on Hugging Face — the largest open-model ecosystem in the world. AI evolution looks kind of like biology, but with some strange twists. 🧬🤖

The 2500th, 250th, 50th, and 25th largest model families on Hugging Face. They show varying numbers of generations (between 3 and 8) and different edge types, including adapters, finetunes, merges, and quantizations.
Benjamin Laufer@laufer.bsky.social · last yr.

I am finding that AI chatbots and language models are rapidly changing my own personal research practices – and my own ethical judgments about the appropriateness of the use of AI.

Lily Xu@lilyxu.bsky.social · last yr.

I'm hiring for a machine learning data scientist & research assistant for summer 2025! Join me on a project on invasive species management with an innovative startup doing on-the-ground removal of environmentally destructive invasive animals. Paid, full-time w/ possibility to extend.

I'm hiring for a machine learning data scientist & research assistant for summer 2025! Join me in working on a project on invasive species management, by predicting species risk and planning capture strategies. This work will be in partnership with an innovative startup that is working directly to capture environmentally destructive invasive species in the US. This project will be working with a US-based conservation partner to build predictive machine learning models and design harvest strategies for the removal of invasive animal species. The primary goal is to develop usable ML models and optimization tools to inform practical environmental decision-making on the ground. There will also be opportunities to extend this work into a research publication for a top-tier AI venue. The project would be: Paid, full-time internship for summer 2025 Possibility of extension beyond the summer (part-time or full-time) Remote, but candidates should be authorized to work in the US Supervised by Lily Xu, assistant professor at Columbia University Ideal candidate background will include: Strong background in CS, data science, and/or applied math Experience developing machine learning models Excellent coding skills in Python Excellent writing and interpersonal communication skills Genuine interest in conservation/sustainability Nice to have: background in optimization methods Nice to have: experience with GIS and geospatial data Candidates would ideally have already completed an undergraduate degree, but exceptional undergrads will also be considered. How to apply: Please send a CV and 3–5 paragraphs of your background/interests via email to <lily.x@columbia.edu> with the subject line "Application: ML for invasive species management". Applications will be reviewed on a rolling basis.
Kenny Peng@kennypeng.bsky.social · last yr.

4) The “most common dog” in NYC is a Yorkshire Terrier named Bella. Jack Russel Terriers are often “Jack” and Charles Spaniels “Charlie.” Huskies are always named Luna—the reason for which is unclear (?).

Bild
Divya Shanmugam@dmshanmugam.bsky.social · 2y ago

We have a new review on generative AI in medicine, to appear in the Annual Review of Biomedical Data Science! We cover over 250 papers in the recent literature to provide an updated overview of use cases and challenges for generative AI in medicine.

Bild
Meredith Whittaker @meredithmeredith.bsky.social · 2y ago

📢NEW: 'Open' AI systems aren't open. The vague term, combined w frothy AI hype is (mis)shaping policy & practice, assuming 'open source' AI democratizes access & addresses power concentration. It doesn't. @smw.bsky.social, @davidthewid.bsky.social & I correct the record👇 nature.com/articles/s41...

Why ‘open’ AI systems are actually closed, and why this matters - Nature

A review of the literature on artificial intelligence systems to examine openness reveals that open AI systems are actually closed, as they are highly dependent on the resources of a few large corpora...

nature.com

Benjamin Laufer@laufer.bsky.social · 2y ago

I passed my “A Exam” yesterday meaning I am officially a “PhD Candidate” rather than a “PhD Student.” (Huge title change, I know.) Thanks to everybody who has supported me along the way!