How model training goes wrong
With a dash of good news on deepfake protection
We’re back with another edition of Machine Learnings, brought to you by the folks at Heyday.
Heyday is an AI-powered memory assistant that resurfaces content you forgot about while you browse the web.
In a week where potentially groundbreaking science news dropped, the Heyday crew geeked out. We should know within a week whether or not we live in a different world. Pretty exciting.
In the office, we’re close to releasing a slew of workflow-driven features for your memory, ideas, and writing. It’s making for a fun summer.
-@samdebrule
What we're reading.
1/ Protection against image manipulation will only increase in importance, so this is welcome news. Meet Photoguard, a protective shield for your photos. Learn more at The MIT Technology Review >
2/ Tangible steps to fine tuning LLaMA 2. We all need it, and this is an extremely simple starting point. Learn more at Brev >
3/ [Listen] With Twitter’s somewhat-in-progress name change, Elon’s plans for xAI are a bit more interesting. Learn more at WIRED >
4/ Speculative read of the week. What if our attention calculations are off by one? Deep diiiiiiive. Learn more at Evan Miller’s Blog >
5/ A somewhat intense look at life as a human-in-the-loop making LLMs work. Are these cracks in the facade? Learn more at The Atlantic >
6/ Interesting thought exercise on watermarking, AI, and language. What practical steps can we take to understand what is actually AI generated? Learn more at WIRED >
7/ An AI director that will interview you and turn your recordings into a documentary. I kind of love this. Learn more at Fast Company >
Research for this edition of Machine Learnings was enhanced by Heyday, the AI-powered memory assistant.