Michael Nielsen
@michael_nielsen
Searching for the numinous 🇦🇺 🇨🇦, currently live in 🇺🇸 Research @AsteraInstitute http://michaelnotebook.com https://bsky.app/profile/michaelnielsen.bsky
I imagine links are being censored, but there is a lot more activity over at b-sky recently, I'm at "michaelnielsen.bsky.social" if you happen to be using it...
I'll bet adversarial attacks are possible - ones where you can add some innocuous text to a paper (or alter the existing text in seemingly minor ways) and significantly increase the likelihood of acceptance by some given model
ICML’s Statement about subversive hidden LLM prompts We live in a weird timeline…
On Ian Fleming’s collection of 1,100 books that “made the world”: chatgpt.com/share/687f143e…
This is really lovely:
A while back, I started making a map of meaningful places that I've been to and that I'd like to see Each spot comes with a note about why it made it on the map
A very useful thread on the OpenAI Gold Medal IMO performance:
Today, we at @OpenAI achieved a milestone that many considered years away: gold medal-level performance on the 2025 IMO with a general reasoning LLM—under the same time limits as humans, without tools. As remarkable as that sounds, it’s even more significant than the headline 🧵
Changing my email .sig to "Proudly overusing em dashes since 1995."
Remarkable chart. (Grok, fwiw, thinks it’s accurate)
people have no idea what life was like before they were born
Singularity is Near:
we are building a wall and we have zero experience and claude is telling us all our instructions