One way: separately embed sender, recipients, subject, body - then use the embedding vectors as input to a logistic classifier
With that setup, I get 95% accuracy on email classification, training on 50-100 base examples. The model trains on CPU in under 1min, and it does inference in under 20ms (most of it is running the embeddings, so you can make it faster if you train your own embeddings model)
That code applies the embeddings + classifier setup on the Banking77 dataset. It gets 93-94% accuracy depending on the embeddings you use (SOTA for this is ~95%, with much bigger and slower models)
I can’t see your gist but spam classification is a textbook example of something you shouldn’t measure with accuracy. If 95% of your samples are not spam you can get 95% accuracy by always guessing not spam.
You should use precision (when your model says “spam” how often is it spam?), recall (how many of the spam emails did it catch), or f1 (balanced between those two).
That's a great point. My case is not for spam, the classes are more balanced, but you are correct that precision, recall and f1 would be better measures for some of these tasks
For a moment I thought this was going to be a metaphor — maybe an ancient Chinese proverb about how paint brushes are made from horsehair and how you can't hold the horse to paint before you've turned the hair into a brush.
There are two parts in the data you supply to Jev for classification - the prompt describing your classification and the data. The data can be quite small - a simple chat message. And prompt part could be considerable since you need to describe your rubrics well.
With Jev you each time pay for your prompt, you can't cache it.
I mean, it sounds like it's only ideal for cases with significant system prompt overhead. I don't think Jev was built to have a large well described prompt setup. To me its more like a happy go lucky small label classification tool with important decisions left to stronger agentic models or yk humans.
I just did a run with a benchmark I just used to test other models against. (It's about detecting irony in german soccer tweets). On my M5 Pro with 48GB it took over 30min to decide on just 100 tweets, the thinking definitely takes long.
It performed quite below Jev, but above other open decision models I tested (68 correct vs 79 correct for Jev - see [1]). I'm running it for the moderation benchmark as well, but that will probably take a few hours on my machine.
Update: Jeeves took about 2 hours to moderate 394 data points and performed really well. It’s not as good as Jev, but it’s super close! In general, it’s super cool that you can tune how strict you want content moderation to be with these models.
Ask Jeeves hired hundreds of cheap liberal arts majors to classify data, some users thought Jeeves was real, the stock spiked when big companies hired Jeeves to automate support thinking it was a silver bullet, and the whole thing collapsed when a better model came along, and it degenerated into ripping off rubes with bottom of the barrel ads.
My bots are all named Jeeves lol. I have a CLI tool I use that connects up to a LLM I made and I call it Jeeves too ...so funny. I really didn't use Jeeves all that much I tended to use...I think it was called Web crawler pre-google era
Cool engineering, but 17s p90 latency kind of defeats the point of a Jev-class model, which is supposed to be fast and cheap. Losing 10 points on MMLU along the way doesn't help.
just get an LLM to think and then force it to output a specific json with prefill post-think.
make sure to include good conditioning text in the prompt with examples of exactly what the output should be like. you don't want dissonance in the probabilities on the prefill.
A classifier is a subset of generative text, so I think responses like this miss the point. Jev is cheap enough and fast enough to sprinkle across your app in ways that LLM would be infuriatingly laggy and unnecessarily expensive, and it's never going to be injected to provide a sorting algorithm in python.
The point isn't that new type of problem has been unlocked, rather a new approach that can unlock new use cases.
when you want a machine to reason about the prompt and generate a structured output not using an actual LLM makes no sense. I have been doing it since the first chain of thought open models became available.
perhaps there may be a way to get a Jev-type model to think for a very specific number of steps to gain control over its latency, if so that would be the next step. truncating LLM thinking like this does not work well, and its thinking isn't efficient anyway.
Just curious, where has this term 'noul' come from for yes/no ansers?
/a bit more digging and..
A Noul performs a Bernoulli trial—an experiment with exactly two outcomes (yes or no)—but instead of picking one, it returns the calibrated probability (ranging from 0.0 to 1.0) that the statement is true.
Not quite.. that boolean is about whether the voltage exceeds some threshold. It's not about how close the voltage is to the circuit's maximum possible threshold, or how much it exceeds the threshold.
Both the voltage and the threshold are probabilistic. Within a tight envelope sure. This is true even of the physical fields that comprise the transistor itself never mind the signal it’s designed to discretize. A more extreme example is a transistor hit with a stray gamma ray. When you dig deep enough there isn’t technically a Boolean anywhere, that’s a platonic construct that makes discussion of engineering convenient.
You specifically said "digital" in GGP comment. I took that to mean that you were choosing the abstraction above which those details matter. Yes, at the physical level all circuitry is effectively continuous and stochastic. Analog circuitry is the manipulation of these to give a more continuous state, digital circuitry uses voltage ranges and a forbidden middle to enforce a kind of noise immunity. When you dig deep enough that the noise is a point of concern, you are no longer looking at a digital Boolean.
Jev noob here. I'm seeing all this jev talk and I understand the difference between this and normal models, but what are some actual use cases for jev?
Moderation or prompt injection prevention. In place of using a LLM to classify a message you use a decision model. It should be way less susceptible to injection. Or at least the injection is less impactful than getting full bash access.
Seems like this is the way, a hybrid approach where some of the pipeline will be jev like and some traditional LLM depending on the nature of the work.
Doesn't the fact that it's general purpose warrant a new term? It's partly that it doesn't need to be trained, but it's also able to play games based on game state, I'd imagine it would be hard to train a classifier to do something like this because you'd need to represent a good distribution of all the states. The general purpose llm world understanding underneath it allows for this.
I've used it to do web research where it follows the most appropriate links, decides what to record in state, etc. I struggle to see how you could implement something with a classifier. That said, I have no idea how deep the technology is and it might be replaced with open source pretty quickly since its drafting of the frontier models and the open source models seem almost as good.
I like the term decision model and I think it's warranted.
outside of ML, for normies i mean, classifier models are supposed to be general purpose. Like I don't think we say segmentation model to do X. In ML, we know of their constraints so we usually say classifier models trained to do X.
Not complaining that decision models are a better term though.
it's a generalistic classifier w/training needed tho. as far as I'm concerned that is somehwat novel and very practical. you get an ok classifier without working out training data and whatnot. sure, you can do classic ML experiments, but the closest that comes to mind is auto ml. not sure how far we got there, it's been a minute for me on this front.
so I don't agree on the premise that it's "just a classifier". it's not something earth shattering, but practical nonetheless
This isn't eating Jev's lunch. This is someone who doesn't understand the entire use case of Jev replacing it with something that doesn't handle it at all.
The number of people who feel the need to try and argue that you don't need Jev can only be astroturfing by those with something to lose - Anthropic and OpenAI employees.
Like it or not, companies are going to use Jev unless you can offer something just as cheap and fast.
I wonder just how much of the business automation market, previously held by LLMs, is at risk here?
interesting bench list, what about benchmark against smaller or bigger models? 9B looks too huge for small like laya, and too small for llm-level decisions.
If the model does autoregressive reasoning before the decision, doesn't that give up much of what a Jev-style model buys you (a single forward pass, cheap calibrated probabilities)? Or is the point mainly to keep the typed output and probability interface while getting better accuracy on harder cases?
reply