Non

AI/ML, prev. Psychology med. | Independent Researcher for AI Interpretability - Lab Affiliation = Goal ❀
Personal Account and
Functional Caffeine Addict

[Non @Nonesse_]

Really looking forward to using this space to discuss more about AI (and less on the intended hobby use prior). My passion and focus is interpretability, and I originally come from a psych background.

Excited to share insights, my thoughts, and engage! ♡

[Non @Nonesse_]

This summing up my feed when I just opened X, was incredible

[Non @Nonesse_]

Sam oh my god

[Sam Altman @sama]

there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again

[Non @Nonesse_]

@JasonBotterill Well? Did you accept

[Non @Nonesse_]

@GeanKassidy @sama Yeah, usually the rollouts arrive for me relatively quickly! Everything on my end is updated, too. Glad this could be of some reassurance, though!

[Non @Nonesse_]

@thoufic_sa @sama At the moment, Plus!
United States as well.

[Neel Nanda @NeelNanda5]

I thought this was an excellent paper! Thanks to Anthropic for asking me to write a review of it, linked below

I've long suspected that models have some kind of "working memory" to store intermediate variables during a forward pass and IMO this paper has the best evidence yet

[roon @tszzl]

the sound of a thousand Bay Area group chats being renamed J Space

[Non @Nonesse_]

This is the best news to wake up to

[Claude @claudeai]

Fable 5 is back.

[Non @Nonesse_]

I’m rarely nervous for a model release, despite still feeling excited. Hopefully Sonnet 5 isn’t overlooked or overshadowed by it not being Fable. I do want Fable 5 back, but I also want this release to go well and have a positive reception.

Curious how well Sonnet 5 also does, as this might be the first ‘normal’ model release since the latest administration orders on Mythos/Fable and GPT-5.6 Sol/Terra/Luna.

[M1 @M1Astra]

Early preparations for Claude Sonnet 5 have been spotted again. Seems imminent.

[Non @Nonesse_]

@M1Astra Excited for this release, nonetheless. Would be a shame to overshadow Sonnet 5 on the basis of it not being Fable, and while I do miss Fable, I’m really hoping Sonnet 5 is a good model.

[Non @Nonesse_]

Great work!

[Karan Singhal @thekaransinghal]

New research on beneficial RL: models trained on a small amount of beneficial trait data improve on a wide range of alignment and benefits evaluations, even if trained only on health domain data.

We hope it’s a step towards more broadly and persistently beneficial models. 🧵

[Non @Nonesse_]

@jxnlco Banger post.
Hi Jason

[Non @Nonesse_]

@mermachine Keyboard mashing to convey uncontrollable excitement, too. I don’t think I’ve seen an LLM casually go, afhdjskskl; yet.

[Non @Nonesse_]

@1thousandfaces_ Sometimes I catch “friend” in the chat names. Started a new session with “Hey, Claude” to get some work done, only for me to notice later that the generated name was,
“Casual greeting between friends”

Immediately was like, aw. Cute, I have to admit.

[Non @Nonesse_]

@JasonBotterill When i had my procedure, I semi-panicked because my MBP was in sleep mode for some time after, and because this is my first mac I had no idea this was common and fine. It still is in sleep and I feel so relaxed. I love this laptop.

[Non @Nonesse_]

@thsottiaux Appreciate you and the team, Tibo!

[Jonas @jonaasw1]

just found out that @OpenAI has their own library??

[Non @Nonesse_]

It would be nice. I often think about healthcare. From assisting with infectious disease and quarantine, to nursing homes and providing care to the many who need it. Who sit alone and with the same, monotonous routine.
Disability assistance too, so individuals at home can have more independence whether or not they need a caretaker to periodically visit.

All of this and especially as workers are stressed, busy, or unavailable and when there’s consistently high demand for these roles.

[Non @Nonesse_]

@jxnlco I’d read it, so my vote is yes.