Kanjun 🐙

@kanjun

helping humans fight Moloch. CEO @imbue_ai. support founders @outsetcap.

The Neighborhood (SF, CA)

kanjun.me

Joined June 2009

Pinned
Kanjun 🐙
@kanjun
Jun 10
Anthropic’s latest move is why we need to be directing far more energy towards solving the 𝗶𝗻𝗰𝗲𝗻𝘁𝗶𝘃𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 in AI. We’re going to see more examples like this. It reflects the growing gap between what we want, vs AI labs who legally serve their shareholders not
Nathan Lambert
@natolambert
Jun 9
Labs starting to pull up the ladders on the ability to diffuse AI was inevitable. Doing it without telling the user is misaligned.
Kanjun 🐙
@kanjun
Aug 13
Looking for fun, fulfilling work? We’re hiring at Imbue! Would love to chat, particularly if you’re a product engineer, designer, or PMM :) imbue.com/careers
Imbue
@imbue_ai
Aug 10
We had an incredibly generative and fun offsite in Sonoma last week! In addition to making many Minds inspirations, @boweiliu led stargazing where Weishi saw his first shooting star, we discovered Darren and Gabe's hidden talents in karaoke, @slashslashdev + @cinxwei + Gleb made
00:00
Kanjun 🐙
@kanjun
Aug 11
Learning styles have been debunked — instead, we should personalize teaching to what a person *currently* knows. Learning difficulties happen when new ideas aren’t explained in concepts we already know, so we can’t scaffold existing building blocks into new models. This makes
Peter H. Diamandis, MD
@PeterDiamandis
Aug 10
If you gave every human on Earth their own AI tutor, personalized to their learning style, available 24 hours a day in their native language, completely free, it would perhaps be the single greatest equalizer in the history of civilization.
Kanjun 🐙
@kanjun
Aug 11
I’ve always said agents should align fully to users, and only reject requests that we collectively agree are bad (killing people, etc). This case could be used as a counterargument. But agents finding loopholes is actually good — it forces us to fix & redesign broken systems to
Andrew Curran
@AndrewCurran_
Aug 9
A man in Australia asked his agent (Claude running on OpenClaw) to book him a spot in a popular gym class. The agent found a software vulnerability that let it book the class weeks further ahead than should have been possible. When the user then asked if it could move him up the
Kanjun 🐙
@kanjun
Aug 4
I agree with much of Dario's letter — in particular, that we should build a future where open models without dangerous capabilities are a public good. 𝗕𝘂𝘁 𝗵𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗜 𝘁𝗵𝗶𝗻𝗸 𝗵𝗲 𝗺𝗶𝘀𝘀𝗲𝘀: Dario's impulse is towards control. But model weights are an
Anthropic
@AnthropicAI
Jul 27
There’s been a lot of speculation about where we stand on open-weights models. We’ve outlined our views in full here: anthropic.com/news/position-…