Overview
COLM is an academic venue focused on the study of language modeling, broadly defined, with the goal of creating a community of researchers with expertise in different disciplines, focused on understanding, improving, and critiquing the development of LM technology.
Sponsorship Details
Booth schedule
Tuesday, Oct 7
October 7
11:00am - 12:00pm
Meet the team - Ads, AGI Foundations
11:30am - 12:00pm
Cooking Hallucinations: Tempered-Training for Polymer Composite QA (PolyCompQA) - Speaker: Sam Blouir
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Foundations
1:00pm - 2:00pm
Meet the team - Alexa Toronto
1:30pm - 2:00pm
Do Biased Models Have Biased Thoughts? - Speaker: Abdelrahman Zayed
3:30pm - 4:30pm
Meet the team - Ads, Customer Engagement Tech, Security Services
4:30pm - 5:30pm
Meet the team - Ads, AGI Foundations, Alexa
Meet the team - Ads, AGI Foundations
11:30am - 12:00pm
Cooking Hallucinations: Tempered-Training for Polymer Composite QA (PolyCompQA) - Speaker: Sam Blouir
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Foundations
1:00pm - 2:00pm
Meet the team - Alexa Toronto
1:30pm - 2:00pm
Do Biased Models Have Biased Thoughts? - Speaker: Abdelrahman Zayed
3:30pm - 4:30pm
Meet the team - Ads, Customer Engagement Tech, Security Services
4:30pm - 5:30pm
Meet the team - Ads, AGI Foundations, Alexa
Wednesday, Oct 8
October 8
11:00am - 12:00pm
Meet the team - Ads, AGI Foundations, Sponsored Products
11:30am - 12:00pm
Live Demo: Information Extraction from Diverse Charts in Materials Science - Speaker: Sam Blouir
12:30pm - 1:00pm
Live Demo: Quantifying fairness in LLMs beyond tokens: A semantic and statistical perspective - Speaker: Chandan Reddy
1:00pm - 1:30pm
Live Demo: Implicit In-Context Learning: Evidence from Artificial Language Experiments - Speaker: Amy Ma
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Nova Scaling-Foundations
1:00pm - 2:00pm
Meet the team - AGI Foundations
3:30pm - 4:30pm
Meet the team - Rufus, Security Services
4:30pm - 5:00pm
Live Demo: Nova Act - Speakers: Anirudh Chakravarthy, Mete Kemertas
4:30pm - 5:30pm
Meet the team - AGI Foundations, AGI Information
Meet the team - Ads, AGI Foundations, Sponsored Products
11:30am - 12:00pm
Live Demo: Information Extraction from Diverse Charts in Materials Science - Speaker: Sam Blouir
12:30pm - 1:00pm
Live Demo: Quantifying fairness in LLMs beyond tokens: A semantic and statistical perspective - Speaker: Chandan Reddy
1:00pm - 1:30pm
Live Demo: Implicit In-Context Learning: Evidence from Artificial Language Experiments - Speaker: Amy Ma
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Nova Scaling-Foundations
1:00pm - 2:00pm
Meet the team - AGI Foundations
3:30pm - 4:30pm
Meet the team - Rufus, Security Services
4:30pm - 5:00pm
Live Demo: Nova Act - Speakers: Anirudh Chakravarthy, Mete Kemertas
4:30pm - 5:30pm
Meet the team - AGI Foundations, AGI Information
Thursday, Oct 9
October 9
11:00am - 12:00pm
Meet the team - Ads, AGI Foundations, Alexa
11:30am - 12:00pm
Live Presentation: Q/A with University Recruiter - Speaker: Emily Barbero
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Foundations, AGI Information
1:00pm - 1:30pm
Live Demo: FalseReject: A resource for improving contextual safety and mitigating over-refusals in LLMs via structured reasoning - Speaker: Chandan Reddy
1:00pm - 2:00pm
Meet the team - AGI Foundations-Scaling
3:30pm - 4:30pm
Meet the team - AGI Foundations
Meet the team - Ads, AGI Foundations, Alexa
11:30am - 12:00pm
Live Presentation: Q/A with University Recruiter - Speaker: Emily Barbero
12:00pm - 1:00pm
Meet the team - SF Lab/Autonomy, AGI Foundations, AGI Information
1:00pm - 1:30pm
Live Demo: FalseReject: A resource for improving contextual safety and mitigating over-refusals in LLMs via structured reasoning - Speaker: Chandan Reddy
1:00pm - 2:00pm
Meet the team - AGI Foundations-Scaling
3:30pm - 4:30pm
Meet the team - AGI Foundations
Expo talk
"LLMs write better programs when they think functionally"
October 8
1:30pm - 2:15pm
Speaker: Dean Foster
Room: 524A
Abstract:
Large Language Models (LLMs) can be prompted to generate code, but ensuring its correctness and efficiency remains a challenge. We propose that the key to improving LLM-generated code lies in leveraging the diverse mental models expert programmers use, including unit tests, pseudocode, and formal verification. This talk demonstrates that compelling an LLM to engage with these paradigms enhances its coding abilities. Specifically, we show that using the Lean theorem prover as an intermediate step for formalizing code properties—like termination—results in more efficient and correct Python code generation compared to direct approaches. We argue that a multi-paradigm approach, forcing LLMs to reason about code through different lenses, is a crucial step toward developing more reliable AI programmers.
Speaker: Dean Foster
Room: 524A
Abstract:
Large Language Models (LLMs) can be prompted to generate code, but ensuring its correctness and efficiency remains a challenge. We propose that the key to improving LLM-generated code lies in leveraging the diverse mental models expert programmers use, including unit tests, pseudocode, and formal verification. This talk demonstrates that compelling an LLM to engage with these paradigms enhances its coding abilities. Specifically, we show that using the Lean theorem prover as an intermediate step for formalizing code properties—like termination—results in more efficient and correct Python code generation compared to direct approaches. We argue that a multi-paradigm approach, forcing LLMs to reason about code through different lenses, is a crucial step toward developing more reliable AI programmers.
