The technical accuracy of what ChatGPT generates with this prompt is honestly surprising and impressive to witness.
GENERATIVE AI
-
Flow Matching for Video Generation: Data Scale and Generalization Benefits
By
–
I loved this research paper on Flow Matching, the most popular approach for video gen. TLDR: More data means harder to fit any specific data (say image), better generalisation, greater coverage of learned concepts. Pretraining: Use Billions of data to learn many things and none
-
Authors’ Copyright Concerns in AI Training Data Usage
By
–
Doesn't help the millions of authors who's books were used for training data without being one of the 500,000 that were illegally downloaded
-
AI-Generated Content: Does Convenience Kill Creative Innovation?
By
–
A Content AI company boasted,
“With our AI, you’ll always have fresh ideas. You’ll never have to stare at a blank page again.” A voice inside my head replied,
“In a world drowning in sameness,
don’t the freshest ideas blossom
because someone dared to stare at a blank page?” -
AI Training Data and Book Economic Value Impact
By
–
I'm don't think that an AI company scanning a book for training data completely destroys the economic value of that book
-
Fair Use and Book Scanning for AI Training: Legal Gray Areas
By
–
To clarify: it appears it is legal, at least in the USA, to buy a used copy of a physical book (used = the author gets nothing), chop the spine off, scan the pages, discard the paper copy and then train on the scanned content The transformation from paper to scan is "fair use"
-
AI Training on Books: Legal Loopholes Harm Authors
By
–
A book is only worth $3000 if the AI company illegally downloads it and gets caught If the AI company buys a used copy for 50 cents and destructively scans it (which apparently is allowed now) the author gets nothing
-

AI Hallucinations: Why Benchmarks Don’t Measure Real Performance
By
–
A very interesting text from OpenAI about hallucinations: AI models shine on leaderboards—but often not in practice. The reason: benchmarks usually only measure accuracy. Anything that is not correct—whether it is a clear error or an honest “I don't know”—is scored with 0
-
User satisfaction matters more than AI evaluation benchmarks
By
–
Users don't give a shit how well you do on evals. They care how well it solves their problems. You can often get evals to *somewhat* correlate with user problems, but coverage is usually mid at best and it's a moving target. We've seen far better success w/ A/B tests against $.
-

Sonoma Sky Alpha and Sonoma Dusk Alpha Stealth Models Released
By
–
Sonoma Sky Alpha and Sonoma Dusk Alpha two new stealth models are now available in anycoder via @openrouter Context: 2 million tokens