Studying multi-agent teams with @CoCoSci_Lab & @Velez_CoLab. PhD Candidate @PrincetonCS. Prev: @CornellCIS, @MITBrainandCog.
Joined October 2019
- Tweets159
- Following363
- Followers449
- Likes550
Pinned Tweet
🚨 What if your LLM team could dynamically reorganize itself mid-task? Introducing LATTE: a framework that gives teams a ~caffeine~ boost via an evolving task graph they build and adapt in real time, achieving SOTA performance with fewer tokens and conflicts. New paper 🧵👇
Elizabeth Mieczkowski retweeted
Excited to share a new preprint where we conduct a resource-rational analysis of planning in a large-scale data set of online chess games! From a great team including @evanrussek @miouantoinette @SagivYotam @weijima01 and @cocosci_lab.
More in 🧵
osf.io/preprints/psyarxiv/cj…
Elizabeth Mieczkowski retweeted
How long does a finding about LLMs stay true?
A new paper re-tested 50 widely cited findings about LLM behavior (biases, hallucination, sycophancy, etc).
Within 15 months, most no longer held or had substantially reduced effects. The shelf life of LLM findings is short.
Almost half (44%) of the tested claims are now expired in today’s models, 34% are weakener, 16% persisted, and 6% have strengthened relative to the original claim
The only effects that got stronger were about sycophancy, LLM ingroup bias, and creativity.
This means we should be cautious when interpreting the long term durability of research on LLMs.
zou-group.github.io/llm-clai…
Elizabeth Mieczkowski retweeted
Last month, 10,000 OpenAI agents cracked a $1M Millennium Prize problem together.
We found OpenAI models are also the best overall teammates.
Introducing 🐟 the Collaborative Intelligence Benchmark, measuring how well AI models work as teammates. 🧵
1/ Governments and organizations often cut staff and overlapping roles in the name of efficiency. But what looks unnecessary on a normal day may be essential when something goes wrong. Our new preprint tests how redundancy protects human teams 🧵👇
8/ The cost of removing redundancy is easy to miss. ⚠️ An organization can look lean and efficient on every ordinary day and still be one absence from failure.
9/ Thanks to my wonderful collaborators: Rachit Dubey, @natvelali, & @cocosci_lab! Preprint, code, and data: osf.io/pdfyk
8/ The cost of removing redundancy is easy to miss. ⚠️ An organization can look lean and efficient on every ordinary day and still be one absence from failure.
9/ Thanks to my wonderful collaborators: Rachit Dubey, @natvelali, & @cocosci_lab! Preprint, code, and data: osf.io/preprints/psyarxiv/pd…
Elizabeth Mieczkowski retweeted
My lab is partnering with EdgeCity to study the ways in which AI agents can help humans coordinate and cooperate in a pop-up village of real people! We're piloting a cogsci-inspired system that is verifiable, inspectable, and steerable. Stay tuned for more!
