My understanding is that the underlying tech is pretty old and well-established. So if you understand those older models, this is old news. A lot of the negativity is basically "yeah, this is old stuff."
But what Jev actually has done that's new and interesting is that it allows you to define a new semantic decision at runtime in natural language, without training a model for that task, and get a fast, cheap, bounded, probability-bearing result designed to be consumed directly by software.
In other words, you don't have to fine tune a model. You should if super high quality is really important to you -- but you should also do that with other things that people reach for LLMs for as well. But for the rest of us for whom good enough is, well, good enough, and who aren't interested in training or fine-tuning models? Jev unlocks a lot off-the-shelf.
You should be impressed. It's doing something more novel than the haters suggest.
A lot of the hate is in the same spirit of the infamous hacker news comment discounting Dropbox because you can just setup an FTP server to do the same thing[1]
In this case, its people who have built BERT classifiers for years, pointing out you can just fine-tune some model to classify. Of course that's true, and still have value, but its not exactly turnkey. And Jev is surprisingly high quality at what its built for.
I frequently have found myself saying, "People pay for that?! But xyz already exists," in the past, but overtime I've learned that (1) marketing is important and (2) if something reduces friction enough to get the wheels turning on something you're trying to achieve, there's always someone (actually a lot of people) out there willing to pay for it even if it's measurably worse than a self-scaffolded solution or even a really good solution that requires a lot of tinkering to get things rolling.
I'm normally happy to jump onboard the HN hate train, but I think this time it's more that the launch was full of marketing slop about AGI and positioning Jev as an LLM competitor, when as you note its basically just a general purpose classifier.
This isn't to knock the team at all, but imo the most impressive thing about Jev is that no one did it before. The second most impressive thing is that there was such a large number of people that didn't realise that classifiers were useful.
As a programmer that uses coding agents a lot, and I mean all the time while I am awake, I really haven't seen the benefit of Jev. It can't code, or I don't know how to make it code. Don't tell me about how good it is about classifying stuff which I don't do, I build apps and need coding agents and LLMs for that, desktop, web, mobile, all kinds of apps
If anybody has pointers on how to make Jev code in any language I'd like to know
Jev does a useful thing, but to me it feels like it should be some middleware plugin available to many AI harnesses, not a standalone product. Under the hood, model tool calls were doing a similar classification to decide if a tool should be called amidst a set of tool descriptions - so for instance, it might be nice plugged into that use if it delivers faster, more accurate tool invocations.
But on HN, there are lots of much more qualified people saying Jev is a con, a step backwards, or bonkers that people are impressed.
As William Gibson said, the future is already here, it’s just not evenly distributed.
So people get excited about existing technologies because they are new to them.
Wrapping an existing technology in a new package is a common way to introduce it to new people…there are many wrappers around ffmpeg, html2pdf, etc. that are marketed as products.
And they seem like magic…hell cassette tapes feel like magic to me even though I had a cassette player in 1973.
Often but not always, the best tool is the one in your hand or the one you know of because then you can just get to the work you are trying to do.
BeRT, BART and FLAN-T5, for text classification. All of those models are ~half a decade old now, are based on the transformer model, and small enough to run locally. They're not perfectly SOTA, but perform quite well in embedded applications. Fine for low-stakes stuff, I think.
Image classifiers are a hugely competitive space. You've got BeiT, DeiT, MobileOne, ConvNext, FastViT and several more models that all support image classification pipelines.
Thanks for quick reply. I downloaded FLAN for comparison and you are quite right, gives similar results, with the benefit of privacy and self-sovereignty in serving the model.
So other than untestable claims of _how_ Jev generates responses, it's main feature right now is its API design?
My understanding is that the underlying tech is pretty old and well-established. So if you understand those older models, this is old news. A lot of the negativity is basically "yeah, this is old stuff."
But what Jev actually has done that's new and interesting is that it allows you to define a new semantic decision at runtime in natural language, without training a model for that task, and get a fast, cheap, bounded, probability-bearing result designed to be consumed directly by software.
In other words, you don't have to fine tune a model. You should if super high quality is really important to you -- but you should also do that with other things that people reach for LLMs for as well. But for the rest of us for whom good enough is, well, good enough, and who aren't interested in training or fine-tuning models? Jev unlocks a lot off-the-shelf.
You should be impressed. It's doing something more novel than the haters suggest.
> It's doing something more novel than the haters suggest
How is that different than https://www.nobodywho.ai/posts/jev-in-25-lines/
Couldn't you always get probabilities for a fixed set of tokens without training a new model?
A lot of the hate is in the same spirit of the infamous hacker news comment discounting Dropbox because you can just setup an FTP server to do the same thing[1]
In this case, its people who have built BERT classifiers for years, pointing out you can just fine-tune some model to classify. Of course that's true, and still have value, but its not exactly turnkey. And Jev is surprisingly high quality at what its built for.
1 - https://news.ycombinator.com/item?id=9224
I frequently have found myself saying, "People pay for that?! But xyz already exists," in the past, but overtime I've learned that (1) marketing is important and (2) if something reduces friction enough to get the wheels turning on something you're trying to achieve, there's always someone (actually a lot of people) out there willing to pay for it even if it's measurably worse than a self-scaffolded solution or even a really good solution that requires a lot of tinkering to get things rolling.
Beware of “just.”
I'm normally happy to jump onboard the HN hate train, but I think this time it's more that the launch was full of marketing slop about AGI and positioning Jev as an LLM competitor, when as you note its basically just a general purpose classifier.
This isn't to knock the team at all, but imo the most impressive thing about Jev is that no one did it before. The second most impressive thing is that there was such a large number of people that didn't realise that classifiers were useful.
JEV is rule-based programming with a better pattern matcher. Charles Forgy did this back in the 80s.
As a programmer that uses coding agents a lot, and I mean all the time while I am awake, I really haven't seen the benefit of Jev. It can't code, or I don't know how to make it code. Don't tell me about how good it is about classifying stuff which I don't do, I build apps and need coding agents and LLMs for that, desktop, web, mobile, all kinds of apps
If anybody has pointers on how to make Jev code in any language I'd like to know
Jev does a useful thing, but to me it feels like it should be some middleware plugin available to many AI harnesses, not a standalone product. Under the hood, model tool calls were doing a similar classification to decide if a tool should be called amidst a set of tool descriptions - so for instance, it might be nice plugged into that use if it delivers faster, more accurate tool invocations.
Hey Kuyawa, it's not a tool for agentic coding. Have a look at product page: https://typesafe.ai/blog/introducing-system-one-models-and-j.... Use cases: "AI-Powered Workflows / smart if-statements"
You're allowed to be excited about new technology.
What are you allowed to feel when it's old technology?
Anything you want. Do as you please, stop waiting for the internet to tell you how to feel about things.
Excited or not.
You don’t need permission to feel how you feel.
But on HN, there are lots of much more qualified people saying Jev is a con, a step backwards, or bonkers that people are impressed.
As William Gibson said, the future is already here, it’s just not evenly distributed.
So people get excited about existing technologies because they are new to them.
Wrapping an existing technology in a new package is a common way to introduce it to new people…there are many wrappers around ffmpeg, html2pdf, etc. that are marketed as products.
And they seem like magic…hell cassette tapes feel like magic to me even though I had a cassette player in 1973.
Often but not always, the best tool is the one in your hand or the one you know of because then you can just get to the work you are trying to do.
First make it work, then make it good. Good luck.
> what software was doing this prior to Jev
BeRT, BART and FLAN-T5, for text classification. All of those models are ~half a decade old now, are based on the transformer model, and small enough to run locally. They're not perfectly SOTA, but perform quite well in embedded applications. Fine for low-stakes stuff, I think.
Image classifiers are a hugely competitive space. You've got BeiT, DeiT, MobileOne, ConvNext, FastViT and several more models that all support image classification pipelines.
Thanks for quick reply. I downloaded FLAN for comparison and you are quite right, gives similar results, with the benefit of privacy and self-sovereignty in serving the model.
So other than untestable claims of _how_ Jev generates responses, it's main feature right now is its API design?