If you are a #dataworker who has been affected by this, please reach out to us and let us know #OutlierAI

Mapping LLM (agent) performance with a Pareto curve, with cost or compute on the x-axis is definitely a good idea! – something we also did earlier with Kevin Clark’s ELECTRA model. (And Sayash Kapoor’s agent evaluation paper uses DSPy, ColBERT, and HotPotQA, oh my!)

When you look carefully, so much of the "innovation" of the last decades has consisted of highly clever ways to make people spend more for the same things, or to avoid a degraded version of the product aeaweb.org/articles?id=10…

NBER Summer Institute starts today! Live streams: youtube.com/nbervideos Programs: nber.org/conferences/su… - Monetary Economics - Corporate Finance - Impulse & Propagation Mechanisms - Capital Markets & the Economy - International Trade & Investment - Development of the

🎧 NEW: Trust & Trade episode #32 "The Non-Competes Ban." Sandeep Vaheesan (Open Markets Institute) & former acting FTC Chair Maureen Ohlhausen (Wilson Sonsini) discuss the FTC's powers & legal challenges surrounding noncompete clauses. trustandtradepodcast.com Listen now: 🔗

A great video from Stanford NLP Group faculty Christopher Potts on: ‣ The Stanford NLP Group 😊 ‣ Recent unprecedented progress in AI ‣ The evolution of language understanding & world models ‣ LLM fabrications & hallucinations ‣ Teaching in the era of AI youtu.be/WQzItGyoVGY?si…

An excellent paper by two influential IO economists, and not technical. Should be particularly valuable to non-economist lawyers interested in merger policy.

For some products and services, the most pressing question is whether they should be legal, not whether enough firms compete in the relevant markets. See prohibitions on asbestos and DDT.

arXiv -> alphaXiv Students at Stanford have built alphaXiv, an open discussion forum for arXiv papers. alphaXiv You can post questions and comments directly on top of any arXiv paper by changing arXiv to alphaXiv in any URL!

“Companies showed this is the way to repeal what the Legislature does," said SLS Professor William Gould for CalMatters in "Uber, Lyft, DoorDash workers remain contractors due to California Supreme Court ruling." stanford.io/3LMfRiZ

Lots of important, cutting-edge consumer protection and antitrust enforcement is being done by local and state officials. Here is just one example.

Happy to share my keynote at ACCC conference “Move fast and fix markets” youtube.com/watch?v=6WDHPa… With 5 lessons on the experience with Big Tech: #1 Economics: theory ≠ practice #2 Ex post competition policy not enough #3 Ex ante regulation is a must, but (probably) ...

Price fixing is price fixing regardless of the tools used. The lawsuit against RealPage brought by the DOJ and 8 states is a signal to businesses that they cannot hide behind algorithms or AI when colluding with rivals. openmarketsinstitute.org/publications/d…

NEW Discussion Paper: "The Economics of Large Language Models: Token Allocation, Fine-Tuning, and Optimal Pricing" By Dirk Bergemann, Alessandro Bonatti, and Alex Smoli: cowles.yale.edu/research/cfdp-…

NEXT WEEK: @YCombinator hosts the Little Tech Competition Summit in Washington, D.C. Confirmed speakers: - Chair Andrew Ferguson, FTC - EVP Teresa Ribera, European Commission - AAG Abigail Slater, Antitrust Division, DOJ Many more to be announced.

Reflect on 40 years of IO advances and emerging research (IO+) at the Industrial Organization in Retrospect conference, May 29-30 on campus. Learn more and register here: econincubator.uchicago.edu/events/industr… #IndustrialOrganization #EconomicsConference #EmpiricalIO #AcademicResearch

.Percy Liang & Tatsunori Hashimoto start the 2nd offering of CS336 Language Modeling from Scratch at Stanford NLP Group. The class philosophy is Understanding by Building. We need many people who understand the detailed design of modern LLMs, not just a few at “frontier” 🤭 AI companies.

.Stanford NLP Group will be in @Singapore with lots of ICLR 2026 papers. Tristan Thrush, Christopher Potts & Tatsunori Hashimoto will show how to select good pretraining data: LLM losses on texts correlate with downstream benchmarks, so select high-correlation docs. arxiv.org/abs/2409.05816