AI + FM papers
@ai-fm-papers.bsky.social
A feed of interesting AI / math / formal methods papers. Posts by @m-dodds.bsky.social
Dataset here: huggingface.co/datasets/qui...
quinn-dougherty/fvapps · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
February 12, 2025 at 1:44 AM
Dataset here: huggingface.co/datasets/qui...
Sounds like essentially the LLM has a set of operations it can perform, and the policy constrains these operations to the ones that are reasonable. Eg “user can’t buy a ticket for less than <minimum price>”
December 9, 2024 at 6:11 PM
Sounds like essentially the LLM has a set of operations it can perform, and the policy constrains these operations to the ones that are reasonable. Eg “user can’t buy a ticket for less than <minimum price>”
Looks like the process is (1) free text input of developer’s policy docs, (2) translation (by an LLM?) into a candidate set of SAT-checkable policy constraints, (3) audit of policies by developers, (4) auto-enforcement of policies on incoming operations generated by LLM from user interaction
December 9, 2024 at 6:07 PM
Looks like the process is (1) free text input of developer’s policy docs, (2) translation (by an LLM?) into a candidate set of SAT-checkable policy constraints, (3) audit of policies by developers, (4) auto-enforcement of policies on incoming operations generated by LLM from user interaction
Previously #2, another SQLite3 bug discovered by Team Atlanta on DARPA AIxCC: team-atlanta.github.io/blog/post-as...
Autonomously Uncovering and Fixing a Hidden Vulnerability in SQLite3 with an LLM-Based System
SQLite3 in ASC
team-atlanta.github.io
November 30, 2024 at 9:22 PM
Previously #2, another SQLite3 bug discovered by Team Atlanta on DARPA AIxCC: team-atlanta.github.io/blog/post-as...
Previously #1, Project Naptime: googleprojectzero.blogspot.com/2024/06/proj...
Project Naptime: Evaluating Offensive Security Capabilities of Large Language Models
Posted by Sergei Glazunov and Mark Brand, Google Project Zero Introduction At Project Zero, we constantly seek to expand the scope and e...
googleprojectzero.blogspot.com
November 30, 2024 at 9:22 PM
Previously #1, Project Naptime: googleprojectzero.blogspot.com/2024/06/proj...
They target variants of known bugs: “By providing a starting point […] we remove a lot of ambiguity from vulnerability research, and start from a concrete, well-founded theory: "This was a previous bug; there is probably another similar one somewhere"
November 30, 2024 at 9:22 PM
They target variants of known bugs: “By providing a starting point […] we remove a lot of ambiguity from vulnerability research, and start from a concrete, well-founded theory: "This was a previous bug; there is probably another similar one somewhere"
Recent NSF award with several of the same authors: "Aligning Code-Generating Models with Formal Specifications" www.nsf.gov/awardsearch/...
NSF Award Search: Award # 2422214 - FMitF : Track I: Aligning Code-Generating Models with Formal Specifications Lock
www.nsf.gov
November 25, 2024 at 11:46 PM
Recent NSF award with several of the same authors: "Aligning Code-Generating Models with Formal Specifications" www.nsf.gov/awardsearch/...