Hacker News Reader

Comment by BoorishBears | Hacker News Reader

But zero-shot classifiers with this level of intelligence, world knowledge, ergonomics, cost profile, and ease of use are new.

I feel like good engineering doesn't just ignore those things, or at least it didn't before recently. Now I guess social media has added a pressure to reduce everything to a hot take.

View on HN

> Ergonomics, cost profile, and ease of use are new.

Following AI from the academic papers side; jev really feels silly. They one-pass the LLM tranformer stack and tune the output network for a probability value.

(some clever pararellization optimisations to make it viable to offer as an api, since the normal kv cashing no longer works if you oneshot the tranformer)

The largest change is the packaging; An api with a tolken based pricing, and a schema to define the output structure for quick setup.

Previous projects would probably involve installing pytorch, running a converter script on Qwen, and write a fair bit of matrix math to change the output shape.

I'm kinda amused that it took this long though.

An LLM is a zero-shot classifier with a large number of classes. All you need to do is establish what the output means and you can fine-tune an LLM final layer for this task if you like (and others have done). A student of mine did this as an exercise two years ago, and it was cool, but not publishable.

I agree with you on the "ease of use" business though. No one thought to make this sort of thing commercially available.

But there is no hot take here. Jev is not some new paradigm; engineering-wise, it is a trivial modification to the existing pipeline. That doesn't mean it isn't commercially viable.

No, all they had to do was come up with a quality post-training recipe, production inference stack that wouldn't fall over, GTM, documentation, schemas, etc. etc.

(also most signs point to this being LLaDA 2.0-adjacent so throw in solving some substantial mid-training)

I think it's 100% a hot take to call what they built trivial. Or at least it used to be.

There was a time when that kind of stuff was something between sour grapes and cluelessness about the gap between an idea and an actual commercial product deployed at scale, but now that's just weirdly normalized.

In fact, if anything I'm the weirdo for repeatedly taking issue with the way people are trivializing it ¯\_(ツ)_/¯

Someone else posted the jevbench site which compares jev to a bunch of other models. If you look only at the accuracy dimension:

https://benchmarkheaven.com/jev-models?w=100-0-0-0#jevc-weig...

Jev actually isn’t anywhere near the top. It even loses to open weight clones. This tells me that whatever their “calibration” dataset is, it doesn’t seem to be anything special.

... why didn't you link to the actual benchmark which does have Jev at the top?

https://benchmarkheaven.com/jev-models

You linked to some weird subtable that labeled: " Not the default — not the JevBench Score", that can only be reached after you see what I just linked... lmao are you really this hard up about things?

Also every single question (even in the hard set) is single dimensional?: https://github.com/fstandhartinger/jevbench/blob/main/datase...

Jeeze, this is getting sad. I guess after all the mass-psychoses where people thought pointless things are going to change the world, we were due for a mass-psychosis where something interesting just has to be pointless?

Because I was specifically responding to your claim that Jev’s training recipe would give it better accuracy than others. It doesn’t have better accuracy than others. You could do as well or better by distilling qwen for example.

Jev is ranked higher than others on the overall benchmark due to speed and/or cost, not accuracy.

Well yeah, that’s the whole idea. If speed and cost don’t matter, you could use Astra.

Obviously it’s the speed and cost that make it compelling. The tradeoff is accuracy.

Enough to matter? Maybe, maybe not. It’s not like it’s way down the chart. It’s probably good enough for a lot of tasks.

Architecturally, it is trivial. That's something the community would have consensus on, so not a hot take.

I see your point, but Jev doesn't exist in a vacuum. When one (like me) says "trivial", they mean it relative to other attempts and developments in the field, all of which require everything you've mentioned at minimum. Commercialising any product, and doing it well, is hard. But the R&D factor here is substantially more straightforward than almost any other product in its category, because there is no architectural breakthrough here.

"all of which require everything you've mentioned at minimum"

Sorry who else did everything I mentioned? I think the guy behind Laya tried after noticing Jev's traction... but the site's auth went down and has stayed down for a day now.

"substantially more straightforward than almost any other product in its category"

More straightforward than the spite projects based on constrained decoding? Or Laya with it's couple of days post-training ModernBERT?

I have no doubt other teams can build models like this and I've love for a frontier lab to give us an even smarter model with these ergonomics... but in the rush to show Jev what's up, we're mostly getting slop.

PS: I don't know anyone who's done anything of note who uses trivial like that. The commentariat do, and the "I could have done that" crowd do, but I don't pay much attention to them until they actually do the thing.

By category, I meant other language models in general. The point of others putting something up to beat Jev is to show that, to date, no one has bothered to produce something like Jev, because anyone with decent LLM experience can roll their own for purpose with little effort and have been doing so for years. And can beat it on any metric you choose.

Let me put it this way. OpenAI and Anthropic have a slight moat over the Chinese labs because they have strong training data and the most advanced RL strategies. It will take the Chinese labs significant R&D effort to bridge that, especially in math (and there is a good chance they will, provided they want to).

Jev has no moat other than the fact that no one else has bothered to package a model in this way. Another lab could build a strong competitor very quickly if they want to put the effort in. That's the point of this post. There is no uncertainty about what they have done, nothing to figure out. Someone just needs to do it. I'm not sure what to say if you can't see the difference between the two. Jev is worth celebrating because of the idea to package it in this way. But it is not a paradigm shift and that is likely a problem for them.

> All you need to do is establish what the output means and you can fine-tune an LLM final layer for this task

Yeah, that was the original idea with GPT, Generative Pre-trained Transformer, and earlier open pre-trained transformers.

Today people use AI via APIs rather then fine-tuning models by themselves and when someone provides this as an API they got excited.

No it wasn't. Those were models trained from scratch, required large scale data, and the nontrivial parts involved training at scale and the autoregressive task which no one expected to work as well as it does. It is the difference between developing a foundation model, and using one. I believe Jev falls in the latter category, because the task itself is no different, only the output.

They almost certainly would perform worse than more specialized classifiers trained with less data. It’s kind of a paradox of generalization. I think there’s an interesting space where you use generalized models to generate ad hoc specialized classifiers.

Depends what you man by "more specialised". You wont train very good language understanding without alot of data. It probably uses the core tranformer stack from an LLM.

Classic classifiers are regularly just tuned general models; Training a CCN on ImageNet and tune it for cats and dogs gives better results than just training it on cats and dogs.

There is likley a small network used to tranform model output vector to probabilities, but that wouldn't be massive. Retraining that small network for specific task may beat jev; but that's bairly considered training by modern standards.

Expecting a strong zero-shot performer to perform worse in a low data regime?

That only makes sense if you try to rope in data previously used to establish the model's priors, but that wouldn't make sense in this context. That same additional data is what enables things like...

> use generalized models to generate ad hoc specialized classifiers.

They almost certainly would perform worse than more specialized classifiers trained with less data. It’s kind of a paradox of generalization.

Isn't this exactly what the bitter lesson is about?

Reply on news.ycombinator.com