YHK - yet another hacker news clone

Anthropic blocks users for anything touching training larger models

95 points pretext 3 hours ago 21 comments

nubinetwork 1 hour ago | parent

Kindof old news, they've been doing this for months...

pkoird 1 hour ago | parent

Better block than falsify information. Last I heard they were deliberately botching it whenever someone asked semi-advanced questions on model training.

dominotw 1 hour ago | parent

so models can reveal "advanced" model building secrets ? how?

How is it a secret if everyone has a model that is exactly like the other. Surely there are no secrets

georgemcbay 1 hour ago | parent

> Surely there are no secrets

Not yet, especially not with all of the incestuous cross-hiring that all the labs do.

But perhaps the various L. Bob Rife variants running these labs are using their advanced unreleased models to work on their own Snow Crashes.

impossiblefork 59 minutes ago | parent

It could be simple training tips that are obvious, but which any particular builder might miss.

nostrebored 50 minutes ago | parent

I don’t think it is revelation, it feels like active incompetence. You can lay out every decision and watch Claude/OpenAI actively fuck up in inexplicable ways.

jbellis 36 minutes ago | parent

They walked that part back within days. Still made me permanently leery off relying on anthropic for anything important.

tristanj 1 hour ago | parent

This user is based in Hong Kong, where Claude is geo-blocked and not available. If you try visiting claude.ai from a Hong Kong IP address, you get an "unavailable in this region" message.

They are likely accessing Claude via a VPN. Doing model training work over a VPN is a highly suspicious signal, and likely triggers Anthropic's classifiers. It's not surprising that their account got flagged.

leobuskin 1 hour ago | parent

US/EU here (both, due to traveling), same story with anything related to modifying existing weights or SotA approaches in training/research, it's not a geolocation issue. The safeguards don't check your IP (but pretty soon will check someone's IP [pun intended])

alansaber 44 minutes ago | parent

I've read writeups about a user trust score and i'd be surprised if it wasn't true, there's no way the safeguard is just a cls in isolation

giancarlostoro 1 hour ago | parent

I have been using it to train a few BERT models.

alansaber 42 minutes ago | parent

I doubt BERT would trip safeguards

WhereIsTheTruth 1 hour ago | parent

US Security Chip act next, your GPU will soon enforce these restrictions too

skeledrew 35 minutes ago | parent

Nice. Push more people to use Chinese chips.

gdiamos 53 minutes ago | parent

Where I draw the line is asking Claude to label training data. I don’t do this.

I ask it questions about large models like big MoEs all the time. I’ve never had it refuse a question, but I would expect it to refuse sometimes.

For training data from existing models I just get open models to spill their guts. It’s fine

In industry we train employees to not reveal proprietary information.

I think it would be easier to do this for a model than a lot of employees

ktosobcy 23 minutes ago | parent

> You built your models on the world's data

most important bit…: stealing world's data…

mapontosevenths 21 minutes ago | parent

I'm using it to train models with a novel linear flash attention mechanism right now. No refusals, and it's been running for at least a week solid.

I suspect this person was actively using it to generate or classify training data directly.

JBits 11 minutes ago | parent

What sort of data are you getting Claude to generate?

clcaev 16 minutes ago | parent

Perhaps a manifest index/list of training data and learning method should be mandatory for “public” use of any model, sort of like food ingredient package labeling. The list of texts used and any weighting seems important to facilitate a public understanding / debate of model expectations and behavior.

peri-cl 6 minutes ago | parent

One day, soon, Anthropic will rest on the ash heap of history, and we will be able to laugh at this fleeting, unpleasant episode. It'll be the butt of jokes—like the crooks who tried to corner the onion market in 1955. Dario will be remembered as some name-forgotten crook who once-upon-a-time tried to corner the world AI market; the what's-his-name responsible for the passage of an antitrust "Onion Futures Act" equivalent of 2028—original purpose long-since obscure.

(I'm not dismissing the infamous onion racket. The onion scheme genuinely hurt people, even though today everyone seems to find it very funny. Future generations will think Anthropic was absolutely hilarious. Like some joke about onions or something).

This, too, will pass. Chins up!