In machine learning, each bug is a gift. Once discovered, each idea that "didn't work" before has a chance of working; behavior that was difficult to understand may become obvious. Overall, a bugfix is a moment to re-examine what we thought we knew, for the better.
@gdb
-
Discovering existing performance config after extensive profiling work
By
–
that feeling when you've been painstakingly profiling your codebase only to end up discovering an existing "go fast" config option that already solves your problem
-

$1M Grant Program for AI in Cybersecurity Launched
By
–
We're launching a $1M grant program for AI in cybersecurity. Looking to support projects like the following:
-
Process Supervision: Training AI Through Correct Reasoning Steps
By
–
Initial results with process supervision — training an AI by rewarding it for coming up with a correct thought process, not just outputting the right answer. Great results in mathematical reasoning:
-

GPT-4 Typo: Examining Language Model Error Patterns
By
–
https://
reddit.com/r/ChatGPT/comm
ents/13t216j/anyone_ever_seen_gpt4_make_a_typo_before/
… -
Comprehensive Review of All 134 ChatGPT Plugins
By
–
Someone tried out all 134 ChatGPT plugins: https://t.co/3cE00fRs3V
— Greg Brockman (@gdb) 27 mai 2023Someone tried out all 134 ChatGPT plugins:
-
ChatGPT Drives E-Commerce Referral Traffic Growth
By
–
ChatGPT starting to drive referrals to e-commerce sites:
-
ChatGPT Adoption: 58% Awareness, 14% Usage Among US Adults
By
–
58% of US adults have heard of ChatGPT; 14% have tried it:
-
Parallel experiments in democratic AI decision-making governance
By
–
Parallel experiments in democratic decision-making for how an AI should behave: