←
Ana Trisovic440 posts
Ana Trisovic avatar

Ana Trisovic

@atrisovic

Research Scientist at MIT | previously with @HarvardBiostats & @IQSS, @UChicago, @Cambridge_Uni and @LHCbExperiment @CERN

◷ Joined May 2009▧ 69 Photos and videos
Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 12

🌍 AI adoption is converging across countries, but dependence on proprietary infrastructure is diverging. The gap isn’t adoption. It’s control.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 12

💳 Industry compute credits are associated with less open-weight use. More compute access doesn’t automatically mean more open science.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 12

🏛️ Public funding is associated with more open-weight model use. Private funding is associated with more model extension: fine-tuning and building on models.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 12

Funding decides more than whether scientists use AI. We linked funding acknowledgments in 104,226 papers (2018–2025) to how each paper uses foundation models. New #AIES2026 paper 🧵

Post attachment
Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 7

🎯 No feature of the model itself predicts its lifespan. Only release date does. A model's shelf life is set by the models released after it.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 7

⏳ Lifespans are shrinking. Each release year compresses the lifecycle by about 23–27%.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 7

📈 Adoption follows an inverted U: models rise to a peak, then decline. I measure this with two quantities: time to peak and lifespan, a model's scientific shelf life.

Ana Trisovic avatar
Ana Trisovic@atrisovic·Oct 7

The LLM in your methods section has an expiration date, and it's getting earlier. I tracked 62 LLMs across 108,000+ scientific papers to measure how long models stay in use. New #COLM2026 paper 🧵

Post attachment
Neil Thompson avatar
⟳ Ana Trisovic retweeted
Neil Thompson@ProfNeilT·1 Dec 2025

Our new paper breaks down algorithm progress in AI (which is faster than GPU progress). Surprisingly, much of the gains have only gone to the largest models, which is bad news for AI efficiency improvement. @MITFutureTech arxiv.org/abs/2511.21622

Ana Trisovic avatar
Ana Trisovic@atrisovic·25 Apr 2025

Excited to be at Dzmitry Bahdanau’s #ICLR2025 talk on the origin story of the attention mechanism, a foundational idea (and Test of Time runner-up) that shaped how we build and understand large language models.

Post attachment