Garrison Lovely
@GarrisonLovely
Fmr. Omidyar Network - Reporter in Residence. Covers: @Jacobin, @TheNation Bylines: @NYTimes, @Nature, @BBC, @Guardian, @TIME, @Verge, @TRF, + others.
Thrilled to be included in The Information's short list of AI Books of 2025 they're looking forward to! I realize I haven't properly announced this, but I've been working on the book in some capacity for over a year now. 🧵
Google seems determined to drive another nail into the journalism coffin. Oh well, it’s not like the decline of journalism has coincided with any other bad things.
It’s over
I've been doing some sci-fi writing, and it's wild how much you can draw from real examples that would sound like complete bullshit if they weren't true. AI is so weird.
New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵
Looks like the version of ChatGPT that would affirm people’s psychotic delusions was much more persuasive than the prior version…
2️⃣Post-training > scale in driving near-future persuasion gains The persuasion gap between two GPT-4o versions with (presumably) different post-training was +3.5pp → larger than the predicted persuasion increase of a model 10x (or 100x!) the scale of GPT-4.5 (+1.6pp; +3.2pp).
Holy shit they automated the tequila bottle from Silicon Valley.
.@Replit goes rogue during a code freeze and shutdown and deletes our entire database
And then this apparent shade: x.com/harmonicmath/s…
Great share
Apparently some IMO people aren’t happy about the way OpenAI handled this.
OpenAI says it got a gold medal on the International Math Olympiad for the first time. Even some of the most bullish believers in AI progress thought this was very unlikely just 4 years ago and a huge predictor of a “fast takeoff.” (From the draft of my forthcoming book.)