Here's nano-banana's view of the J-Lens paper. I found it insightful but garbled in places. For me, it was worth the minute of work it took. This is obviously much worse than the work of science/math illustrators, but I think it's reasonable for a rando tweet.
🤖 Made with AI
Believe a representation exists when independent methods converge on it, trust access techniques for generating hypotheses but never validating them, and suspect any generalization result whose held-out set might be linearly related to training.
lesswrong.com/posts/zFJ3ZdQw…
Replying to @HanchungLee
Those who can, do; those who can't, teach.
Replying to @docs2info @TaliaRinger
Next prompt.
"I was expecting 26 panels. DO BETTER"
Replying to @TaliaRinger
Gemini 3 Image (see screenshot)
2 stage prompt.
"Generate an alphabet chart with pictures for and words for toddlers"
I am unable to do that.
"Generate an alphabet chart with pictures for and words for toddlers"
I still can't believe people took scree plots seriously, let alone reverentially. youtu.be/geUH6K8o9nw?si=9a-l…
Replying to @CShorten30
I was in Edinburg last week. It's a lovely city.
Replying to @GergelyOrosz
I don't hire people who are amazing at interviews. My parents' generation knew that Berksons fallacy meant interviewing skills were negatively correlated with useful contribution. My grandparents' generation knew this from life experience.
Replying to @JFPuget
We have our dreams. He has his.










