Phd @Mila_Quebec. Forecasts, discovery, sequential decisions (single and multi-agent).
Montreal, QC
Joined April 2021
- Tweets131
- Following733
- Followers486
- Likes933
Andrew Williams retweeted
1/7: 🚨 AI agents are predisposed to collude. 🤖📉
Our #ICML2026 paper shows that reasoning models (like DeepSeek-R1) tacitly collude on prices with competing firms, even when explicitly told NOT to.
And their chain-of-thought reveals nothing. 🧵
Andrew Williams retweeted
New paper:
We present a "Unified Neural Scaling Law" functional form that accurately models & extrapolates the multivariate scaling behaviors of artificial neural networks as the variables listed in this attached video are varied.
(1/N)
Andrew Williams retweeted
PipelineRL: Faster On-policy Reinforcement Learning for Long Sequence Generation
Alexandre Piché, Ehsan Kamalloo, Rafael Pardinas, Xiaoyin Chen, Dzmitry Bahdanau.
Action editor: Sebastian Tschiatschek.
openreview.net/forum?id=A35a…
#pipelinerl #accelerato
Andrew Williams retweeted
ever been here?
open overleaf → write a paragraph → "hmm...this needs a citation" → open 15 different tabs → skim 8 abstracts → find the 1 actually relevant paper → format bibtex → paste it back on overleaf
if so, i built a plugin just for you. meet openleaf:
→ reads your paper paragraph by paragraph
→ searches major academic databases
→ filters out irrelevant papers using ai
→ one click to add BibTeX to your .bib
you'll also find the 🤝 friendly and 🔥 fire reviewers there. i don't think i need to tell you what they do :)
free. open source. no account. no data collection.
works with ollama, openrouter, openai api and more.
github.com/demfier/openleaf
dear algorithm, please show this to my fellow researchers in need 🙏
#overleaf #latex #opensource #academictwitter
Andrew Williams retweeted
I gave a talk on LLM zero-sum learning dynamics last week at MSR Montreal. I went over a few things that were not in the paper but that I'm particularly excited about; one of those is the connection between generalization and zero-sum learning. youtu.be/UyK3DgWY7yw
I have found Mila to be a great place to do a PhD. Feel free to reach out if you have any questions!
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2026 admission! For more information, visit mila.quebec/en/prospective-s…
1
110
Great opportunity for those new to ML research!
New to ML research? Never published at ICML? Don't miss this!
Check out the New in ML workshop at ICML 2025 — no rejections, detailed feedback, awards, and ICML tickets for selected authors.
Deadline: June 10 (AoE)
Submit: openreview.net/group?id=ICML…
Info: newinml.github.io/
4
266
Andrew Williams retweeted
LLMs have complex joint beliefs about all sorts of quantities. And my postdoc @jamesrequeima visualized them! In this thread we show LLM predictive distributions conditioned on data and free-form text.
LLMs pick up on all kinds of subtle and unusual structure: 🧵
Andrew Williams retweeted
This is fun because LLMs can condition on free-form side information, and make predictions about anything. This turns qualitative knowledge into quantitative predictions.
Here we condition Llama 3 on two datapoints, plus text. Changing the text changes the meaning of the data.
Andrew Williams retweeted
🚀 We have released our paper on ReTreever! 🌳🔍
ReTreever organizes and represents documents in a binary tree across various granular levels, balancing cost & utility while enhancing retrieval transparency.
📜 Read it here: arxiv.org/pdf/2502.07971
#AI @ServiceNowRSRCH
🧵👇
Andrew Williams retweeted
Happy to announce AlignVLM📏: a novel approach to bridging vision and language latent spaces for multimodal understanding in VLMs! 🌍📄🖼️
🔗 Read the paper: arxiv.org/abs/2502.01341
🧵👇 Thread
Andrew Williams retweeted
this is paper is kinda wild. turns out that if you simply ask an LLM to straight out predict a timeseries like this:
```
<history>
(t1, v1) (t2, v2) (t3, v3)
</history>
<forecast>
(t4, v4) (t5, v5)
</forecast>
```
making sure to prepend the prompt like this:
```
Here is some context about the task. Make sure to factor in any background knowledge, satisfy any constraints, and respect any scenarios.
<context>
((context))
</context>
```
it will just… do it? beating SOTA timeseries forcasting?!
llama 3.1 405b directly prompted is more precise at forecasting real-world series than:
- stats-based timeseries models (ARIMA, ETS)
- foundation models specifically trained for time series (eg. chronos)
- multimodal forecasting models (eg, time-LLM)
peak 'bitter lesson' behavior lol
Andrew Williams retweeted
📢 New paper “Language model scaling laws and zero-sum learning” @scifordl #NeurIPS2024
ℹ️openreview.net/forum?id=yBq2…
TL;DR: scaling improves LMs by mitigating zero-sum learning, a mechanism that could be targeted directly and independent of scale.
W205 @ 4:30pm
(1/12)🧵
Andrew Williams retweeted
Starting the workshop on Time Series in the Age of Large Models (TSALM) at #NeurIPS2024 with @tomaspfister's invited talk on Multimodal Time Series Modeling!
Andrew Williams retweeted
🚀 Excited to present our work on VCR: Visual Caption Restoration – the 1st and unique VLM benchmark testing if VLMs can focus on tiny but crucial details!
📍 Join us at #NeurIPS24 on Sunday, Dec 15, West Ballroom B
🛠️ Dive into the details: github.com/tianyu-z/VCR
Andrew Williams retweeted
🚀A game-changer for open-access multimodal AI! BigDocs is paving the way for transparent, accountable, and innovative document reasoning and code generation. Check it out! 💡👏
🎉 Excited to introduce BigDocs!
An open, transparent multimodal dataset designed for:
📄 Documents
🌐 Web content
🖥️ GUI understanding
👨💻 Code generation from images
We’re also launching BigDocs-Bench, featuring 10 tasks to test models on:
➡️ Document, Web, GUI Visual reasoning
➡️ Converting images into JSON, Markdown, LaTeX, SVG, and more!
📜 Paper: arxiv.org/pdf/2412.04626 huggingface.co/papers/2412.0…
🌍 Website bigdocs.github.io/
Andrew Williams retweeted
🎉 Excited to introduce BigDocs!
An open, transparent multimodal dataset designed for:
📄 Documents
🌐 Web content
🖥️ GUI understanding
👨💻 Code generation from images
We’re also launching BigDocs-Bench, featuring 10 tasks to test models on:
➡️ Document, Web, GUI Visual reasoning
➡️ Converting images into JSON, Markdown, LaTeX, SVG, and more!
📜 Paper: arxiv.org/pdf/2412.04626 huggingface.co/papers/2412.0…
🌍 Website bigdocs.github.io/
Andrew Williams retweeted
Learned optimizers can’t generalize to large unseen tasks…. Until now! Excited to present μLO: Compute-Efficient Meta-Generalization of Learned Optimizers! Don’t miss my talk about it next Sunday at the OPT2024 Neurips Workshop :) 🧵arxiv.org/abs/2406.00153 1/N
Andrew Williams retweeted
Spent the weekend hacking together Exa embeddings over 4500 NeurIPS 2024 papers - neurips.exa.ai
Let's you:
- do otherwise impossible searches ("transformer architectures inspired by neuroscience")
- explore a 2D t-SNE plot
- chat with Claude about multiple papers
