-
Executable Code Actions Elicit Better LLM Agents
Paper • 2402.01030 • Published • 170 -
CodeAttack: Revealing Safety Generalization Challenges of Large Language Models via Code Completion
Paper • 2403.07865 • Published • 1 -
CodeAttack: Code-Based Adversarial Attacks for Pre-trained Programming Language Models
Paper • 2206.00052 • Published • 1 -
CodeACT: Code Adaptive Compute-efficient Tuning Framework for Code LLMs
Paper • 2408.02193 • Published • 1
David Barnes PRO
DataImaginations
AI & ML interests
I mostly waste my days away obsessing over AI's application to my lifelong love of psychology and NLP.
GPT came out when I was at LSE studying a professional Data Science course, and I instantly realised it was a waste of a year. I was about to put the LSE project on my portfolio before it goes live, when IBM called me flogging some course on advanced AI, which was basically the same as the LSE course with a teeny bit of AI added on. This gave me the idea to demonstrate how pointless most of it is in this day and age, by replicating all the LSE projects on ML, and automating it all with AI. That, and i've built (so far) reasonably unique way of using NLP within model architecture to enable video game characters to behave, talk and recall, exactly as if they are within the game world on a teeny it of RAM. I call it the 'Persona-Forge, and it has some automation i'd like to make... more automated before I demonstrate it to a software publisher who somehow became interested