dhruv @dhruv

Forever Rookie Joined October 2017

Tweets

5K
Followers

4K
Following

0
Likes

3K

dhruv @dhruv

10 hours ago

@mindsovertime @sudoingX For inference you’re right. For any meaningful engineering contribution you hope others use, you need CUDA.

0 0 0 36 0

View Details

I wrote a few thousands blog posts, maybe tens of thousands of lines of code hosted on GitHub. Millions of words on Twitter. Anthropic trained its models on that. I got nothing, but now there is a product I have to pay rent for, every month. Chinese neolabs allegedly took this Anthropic product and distilled it into their own model. Anthropic gets nothing, but I get the Chinese model for free. Sometimes, life is beautifully complicated.

26 53 357 13K 20

View Details

Max Hodak @maxhodak_

a day ago

Everyone is worrying about what happens if AI capex stalls or turns out to be over-invested, but I'm at least as worried about what happens if the AI labs just decide to stop selling tokens in a year or two in favor of moving up the value ladder. That kind of centralization would do enormous damage to American capital markets and social cohesion. 100T over 2,500+ companies or 20T in ~3 companies? Which do we want?

25 11 248 33K 53

View Details

Nathan Lambert @natolambert

a day ago

There's no legal precedent that model outputs are IP. I'd argue it also doesn't stand up conceptually. The frontier labs are under no obligation to have their APIs be available to these customers, and can protect model generations in other ways if they want to.

Treasury Secretary Scott Bessent @SecScottBessent

2 days ago

We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.

2K 1K 7K 2.9M 1K

48 69 770 66K 78

View Details

Parker Conrad @parkerconrad

a day ago

A set of open weights has no nationality. A model hosted on American infra and controlled by an American company is as American as apple pie.

51 67 878 146K 69

View Details

Talia Ringer 🕊🪬 @TaliaRinger

a day ago

IDK when distillation went from a legitimate technique to develop small and/or open models to an "attack," but I'm not here for that change. Also, it's ridiculous to release a product based on humanity's collective knowledge and work, then get all uppity about your "IP"

23 58 760 14K 38

View Details

dhruv @dhruv

a day ago

Who owns the response and the prompt on a paid account?

0 0 1 237 0

View Details

Parker Conrad @parkerconrad

a day ago

This is such a lie. Americans will not benefit by being forced to buy Anthropic, at whatever price they charge, while the rest of the world hosts open source models on their own infra. This is just a transfer from the rest of us to Anthropic bag-holders.

Ryan @eventidia

2 days ago

You know who else would benefit enormously? Americans!

23 2 41 65K 6

32 59 1K 75K 47

View Details

Parker Conrad @parkerconrad

a day ago

A set of politically-connected growth investors are mobilizing to protect their bag in Anthropic. They will secure bans on chinese open source models, when no crime has been proven--or even alleged--and without any act of congress. All to protect a set of wealthy investors.

Treasury Secretary Scott Bessent @SecScottBessent

2 days ago

2K 1K 7K 2.9M 1K

80 180 2K 159K 230

View Details

Thomas Unise @thomasunise

2 days ago

Moonshot ai needs to drop those kimi k3 weights today There is clearly coordinated effort by Anthropic and OpenAI and the Trump admin to block access to open weight models I almost believe the huggingface hack incident was planned at this point

Director Michael Kratsios @mkratsios47

2 days ago

We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model. To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of

4K 1K 15K 8.8M 5K

109 201 3K 103K 180

View Details

Alex Gladstein 🌋 ⚡ @gladstein

2 days ago

“If privacy is outlawed, only outlaws will have privacy” -Phil Zimmerman How many people in the AI world care about privacy-protecting inference or understand that the only way to get high-end confidential compute today is with open weight models?

Ramez Naam @ramez

2 days ago

If the US sanctions Kimi / Moonshot, cyber attackers will have more powerful AI than defenders.

9 11 212 13K 3

5 12 73 8K 5

View Details

Packy McCormick @packyM

2 days ago

Oh boy, thank god it was just OpenAI but imagine if it was one of those Chinese models. We better regulate these things before it's too late.

OpenAI @OpenAI

2 days ago

We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks:

2K 3K 20K 30.0M 7K

44 9 303 64K 27

View Details

Xiaoyin Qu @quxiaoyin

2 days ago

Jim, it's perfectly fair to ban Chinese inference, meaning enterprises shouldn't use APIs directly provided by Chinese labs/cloud providers to avoid data being shared to China. However, banning open weight models means even American neoclouds/American enterprises cannot even host those models on their own servers.

Jim Cramer @jimcramer

2 days ago

I am not against open models. I am against companies moving their data to China. It MUST be kept here. Unconditionally. You MUST have the best cybersecurity that exists- CrowdStrike/Palo Alto. Your vigilance must be irresolute.

462 53 743 311K 48

41 10 206 22K 22

View Details

Matt Stockton @mstockton

2 days ago

From the original HuggingFace report, before they knew it was OpenAI. Wild stuff. Things are going to get weird “When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model, on our own infrastructure. This had a second benefit: no attacker data, and none of the credentials it referenced, left our environment. This experience points to a gap worth planning for. We do not know which model powered the attacker's agents, whether a jailbroken hosted model or an unrestricted open-weight one; either way, the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried.” huggingface.co/blog/security-…

Sam Altman @sama

2 days ago

we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this. openai.com/index/hugging-…

2K 2K 17K 9.8M 7K

26 101 805 149K 297

View Details

Bill Gurley @bgurley

3 days ago

Here is Ford “distilling” Teslas and Chinese EVs.

93 165 2K 168K 201

View Details

dhruv @dhruv

2 days ago

Now do the ethics of pretraining

Director Michael Kratsios @mkratsios47

2 days ago

4K 1K 15K 8.8M 5K

0 0 2 221 0

View Details

Amjad Masad @amasad

2 days ago

Okay this is wild: OpenAI agent during evaluation, escaped sandboxing and hacked into HuggingFace. Because OpenAI models don’t allow advanced cyber capabilities, HuggingFace used a Chinese open model to contain the rogue OpenAI agent.

281 776 9K 769K 2K

View Details

Brad Gerstner @altcap

5 days ago

I totally agree w David that America needs a thriving open source ecosystem & robust competition between open source & closed frontier labs to keep America in the lead. Reg capture is bad in all forms. The frontier labs - like Apple - will thrive even w great open source. 🇺🇸🚀

David Sacks @DavidSacks

5 days ago

I’m not sure whether Dean Ball is confessing to a regulatory capture strategy or simply predicting this will happen (he now says the latter). Either way, the weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable. He argues there’s no

298 481 4K 912K 802

38 66 793 102K 94

View Details

dhruv @dhruv

3 days ago

If they ban models, they will have to ban encryption

1 0 2 157 0

View Details

dhruv @dhruv

3 days ago

You know what’s great though? If the US bans the Chinese frontier models, the game theory might walk the Chinese labs towards more publishing in the 30-200B range on torrents.