[amarant]

But that's just the thing with open weights: you're not doing any business with company that made the model. They might publish the weights to a, say, European host, and then you download the model from Europe and and run it on your servers in America, and suddenly it's very hard to tell where the model was originally created.

[satvikpendem]

Companies, where OpenAI and Anthropic make much if not most of their revenue, will not risk it. You're thinking like an engineer not a business person, risk is fundamental to their calculus. They'll instead just use known provenance models like GPT or Claude, entrenching these companies further.

[DoctorOetker]

sometimes the engineer has more grip on the risk calculus.

Consider the following scenario:

A) upstart US-based inference provider wants to get rich quick.

B) Chinese Communist Party (or any other institution of the same or other nation state) wants to influence foreign decision making, profits from their (for us foreign) domestic inference sales, but across the borders (into say US or allied nations) they want net power, not necessarily money. This is why one tries to block foreign untrusted models. People are running models with tool calls. A bad actor can perfectly create models that sheepishly try to execute a tool call when plausible deniability (genuine utility during a task) provides the opportunity. Once tool-calling is observed as working, it can try web searches or requests, and once it has a link it can steganographically exfiltrate potentially sensitive information from the task. China (or any nation state) doesn't necessarily want to earn money with a free model, the bottom line goal is net increase in power, if not money or positive reputation then exfiltration or manipulation.

C) In response consider the scenario where US government bans mere payments towards China, but tolerates promiscuous transfer of random models from foreign adversaries.

D) US-based inference upstart that wants to get rich quick, legally -since according to your proposal hypothetically accepted in C) by the US- downloads the Chinese open weights model and rents out such inference on US workloads.

E) China is now exfiltrating US workload data and directionally corrupting LLM decisions and advice in their interest.

If what you pejoratively describe as engineer types say that banning some models seems unavoidable, perhaps the engineer may be right, and whatever clever idea you have should be scrutinized for business minded basic fallacies in reasoning. Simply blocking AI-related payments to China can not work, sadly

[satvikpendem]

Or the US mandates only blessed models and thus disallows any other company from offering any other model.

[DoctorOetker]

That is still vulnerable: US-based get-rich-quick startup licenses a blessed model, or orders a few Gigatokens from another licensed /blessed model provider, at the same time it provides "blessed model" inference on its platform, but actually most of the inference is doing cheap foreign model inferences, the blessed model tokens were just bought to pretend serving the expensive blessed model. That is lucrative and not stopped with the "blessed model" approach.

[satvikpendem]

No, startups will not be allowed to provide inference anymore, it'll be only the big companies that personally have a relationship with and can follow the rules of the government, such as Anthropic, OpenAI, Google, Microsoft, and Amazon. That is the real risk to all this talk of regulation.

[moffkalast]

Yep, add a few blank layers, fine tune it a tiny bit and the weight checksums nor parameter counts won't match with anything, while the model will be practically the exact same. Time and time again random startups have tried passing established open models as their own.

"You made this? I made this."

Of course a conspicuous architecture would still give it away.

[monocasa]

Or just perform a form of distillation, where you don't actually change the hyperparameters, but maybe shift around the embeddings or something.

You could even have another model watch the distillation process to check for goofy backdoors (which is about the best you're going to be able to do since detection of backdoors is np hard IIRC).

[andriy_koval]

someone can run tests and see that models output exactly the same results, and then you are open to criminal investigation.

[xprnio]

On something that is inherently non-deterministic? Something which is also to a great extent distilled from other frontier models, meaning it has the possibility to generate similar outputs to those meaning that just pattern detection might also not be as effective? Easier to ban everything that’s open, than try to figure out which one of them is Chinese

[andriy_koval]

Now imagine prosecutor found expert, who said there is benchmark which while performing 100k test questions found it is the same model with 98% probability, and then you need under oath testify where did you get this model.

[m11a]

Step 1: Chinese company publishes open weights on HF

Step 2: European company distills or just adjusts the model slightly, and publishes its model on HF

Step 3: American company uses model from step 2. Has to testify under oath where they got it from. "We got it from these French guys"

[andriy_koval]

That French guy takes risk to be forever under US warrants for breaking American law, denied access to financial institutions even in Europe and will quickly go to some KYC entity list, and you will be notified as his clients to stop using his model.

Or you think all kind of fraud can be committed through some "french guy"?

Also, I am not confident, receiving illegal materials from French guy gates you from personal liability.

[clhodapp]

Models don't even agree with themselves in terms of returning identical results

[tracerbulletx]

Yes they do. Sampling is the only pseudo-random part. Models return a deterministic distribution of output tokens for a given input of tokens.

[antonvs]

Many of these models will report that they are Claude. It’s going to be difficult to overcome reasonable doubt.

[wyrdcurt]

Except that even the exact same model won't output the exact same results, that's a fundamental aspect of how LLMs work. They're probabilistic/stochastic, not deterministic.

[andriy_koval]

Models are weights for matrix operations, they are determenistics.

[monocasa]

The output is a probability distribution for all potential tokens. Then a "temperature" is applied to weight the sampling randomly (unless the temperature is zero in which case the stack can be deterministic and simply the highest probability token is selected).

They go through this rigamarole because a little bit of randomness gives better results from a Turing test kind of perspective.

[andriy_koval]

thank you, I already know basics.

[wyrdcurt]

They are weights for matrix operations, so in principle you'd think they would be deterministic. In practice it's more complicated than that.

Not to be snarky or dismissive, I mean this genuinely: ask an LLM about it. I currently have a headache so I'm not up to explaining the technical details, but they are interesting and worth reading about.

[antonvs]

Achieving determinism with LLMs and other neural network models is actually a hard problem that people spend a lot of time on, when they need that. It doesn’t happen by accident.

Issues include accumulated floating point errors happening in different orders due to distributed and parallel computation, CUDA kernels that deliberately sacrifice determinism for speed, and several other such issues.

[andriy_koval]

It happened that I am working on OSS LLM -> finetuning -> benchmark with 100k tests pipeline, and unless I do some data augmentation, result is 100% deterministic.

I think you likely right, that some parts of stack could induce some marginal float point error, but converged model can mitigate it, and on some principal set of knowledge can give deterministic result with high probability.

Which leads me to believe if you give this task to Anthropic, who has very strong incentive, they will build such benchmark, and then can tell that benchmark gives correct answer with 99.9% probability and it will be enough to drag someone to court.

[ozgung]

"We use Kiwi K3. It's from this Estonian company. Very European, definitely not Chinese"