Victor Sanh

TBA | Building autonomous AI engineers | Prev. early crew @huggingface 🤗 | AI x NYC

[Victor Sanh @VictorSanh]

https://t.co/GeZPNw7bPo

[Victor Sanh @VictorSanh]

@colorado_plays @etnshow Fascinating

[Victor Sanh @VictorSanh]

@andimarafioti wikipedia is a message/collaboration board for us, the peasant humans

[Victor Sanh @VictorSanh]

@anatolium @julien_c

[clem 🤗 @ClementDelangue]

Sorry to hear that! As you mentioned, we've been pretty clear on the commitment from NVIDIA to keep us open, neutral and silicon agnostic (cf that screenshot from the 8k filling). But we'll work hard with our actions in the coming years (like we did in the past 10) to show you that and hopefully regain your trust (in case this is not just a ragebait)!

[Victor Sanh @VictorSanh]

@salesforce @DarioAmodei @Benioff @AnthropicAI @Dreamforce Were there no more chairs left?

[rishi @RishiBommasani]

*Public* is critical here. A lot of times in frontier AI policy conversations, including on topics advanced by folks in the core AI safety community, I see a focus on transparency to governments due to various issues from broader disclosure. These issues are real and should be handled carefully in relation to the specific disclosures.

But I think the general trend needs to be public transparency, not just government-only transparency. The simplest argument for this is to contras the marginal utility gain from information being made available to a government vs. the public given the disparity in technical capacity. If governments have very little technical expertise (as most agree today, irrespective of the specific governmental body or jurisdiction), and a much larger community of independent experts exist outside governments, then there is a huge premium for getting the information to this broader collection of nonprofits/academics/think tank analysts/etc.

In general, I think a lot of the interest in more restricted disclosures lies in an uncanny valley. It seems like an elegant intermediate solution to a complex tradeoff but actually it might be closer to the worst-of-both-worlds rather than the best-of-both-worlds. On most issues in the frontier AI space today, I would say we should push fully to public transparency or tolerate none at all.

[Victor Sanh @VictorSanh]

@kroscoo Now imagine that people with zero ties to china/asia do the jump, similarly to how people sometimes move to the middle east

[Victor Sanh @VictorSanh]

imagine a future where incredibly talented ai engineers/researchers move to china for professional reasons

talk/philosophize less and do more

caricaturally simplistic but sounds appealing

[Victor Sanh @VictorSanh]

europeans proving once again how good they are at regulating technologies they are not leading the creation of

[etn. @etnshow]

BREAKING: 40 British MPs have signed a letter to Prime Minister Andy Burnham calling for a ban on the development of superintelligent AI.

[Victor Sanh @VictorSanh]

Not so long ago, the mainstream population would panic every time a prominent scientist gave their opinion loudly. We’ve been quick to forget the lessons.

"I do AI research at {insert_frontier_lab}" is the 2026 version of "I am an epidemiologist" in 2020. It converts a semi-educated forecast into an authority claim.

But being close to the AI frontier doesn't translate into expertise in forecasting societal outcomes, just as virology didn't translate into expertise in policy tradeoffs. Proximity to the science ≠ proximity to a holistic answer.

COVID discourse kept swinging between extremes. From "don't panic, masks are not necessary" to "masks are mandatory", from "it's just a flu" to "this is the plague". Sometimes within days, as we learned more. And often stated with full confidence by the same experts.

The AI safety discourse does the same. Scaling is over, scaling is accelerating (just a different form of scaling). Alignment is tractable, alignment is hopeless. The confidence tracks the news cycle more than it tracks the evidence. Any P(doom) looks more like an educated guess today than consensus evidence.

The only true thing at this point in time is that we don’t know.

Remember hydroxychloroquine. Plausible solution, loud and charismatic messenger, a tribe forming around the belief, and the belief becomes identity before any strong evidence can be gathered and verified. Once it’s identity, evidence can’t convince believers. And you end up with doctors who say “we don’t know yet” getting attacked in the street.

We are running the same sequence with AI. People outside the AI industry are making drastic decisions based on whoever sounded the most legit that week. And policy written for whichever camp lobbies the most.

[Victor Sanh @VictorSanh]

@dawoodnyc @Gradio congrats desk neighbor!!

[Victor Sanh @VictorSanh]

Still think it's one of the wildest pivot in the AI industry. From AI friend to the most important platform for open source and open science! We'll write about it in history books for decades to come.

Incredibly excited for @ClementDelangue @julien_c @Thom_Wolf & the @huggingface team to continue pursuing that mission with the goat @JensenHuang!

🤗🤗🤗
So many inimitable faces in this pic of the team from 4 years ago!

[Victor Sanh @VictorSanh]

@NielsRogge @huggingface @nvidia solid picture, congras!

[Victor Sanh @VictorSanh]

@victormustar congrats! next level🔥

[Victor Sanh @VictorSanh]

@mervenoyann @nvidia writing history, congrats!!

[Victor Sanh @VictorSanh]

@LysandreJik the goat! congrats!

[Victor Sanh @VictorSanh]

@MichaelBenayou1 @nvidia @JensenHuang incredible, bravo!

[Victor Sanh @VictorSanh]

@BrigitteTousi big day! congratulations :)

[Victor Sanh @VictorSanh]

@LucSGeorges happy birthday and big congrats!