Making the edge cases work, and lowering the cost of running it via good model choice, context management etc is in some cases really hard. That's valuable in a decent number of cases. I have a system that does something in financial markets - making sure it doesn't screw up, and doesn't cost a fortune to run, is the entire thing for me.
While I don't believe they are doing the following: you can calibrate by inspecting the reasoning traces. That is the relevant distribution. If you ask someone to explain how/why they are classifying something one way v another, you can get a reasonably good understanding of their confidence level.
This tells me the confidence of the LLM's belief about the response - which is different from the calibrated confidence score. The former also is useful (just not what I thought their advertisement sells - and from the article it seems like it tripped up others as well), and there are different techniques to extract such a value [1] [2], typically via "response sampling", i.e., interrogate the LLM slightly differently to see if it changes its answer.
I mean the model can learn from it during RL training. The confidence score is affected by the tokens prior to it it's output. I was using the word "you" loosely.
In my company, and I think in most companies that are using AI at all, one of the first ways it got integrated is as a classifier, to tag orders based on feeding all their data into a prompt and asking for a structured output.
I think demand for tools that are more tailored for this type of integration is high. I don't really understand why Jev is supposed to get my company's decisions right more than an LLM, but regardless of the tech I think people are just excited about the possibility of iterating faster, more explainability, higher-level tools that are specifically created to help hone classifiers etc.
I think the point of Jev is to thread the needle of the gap between non-LLM classifiers and LLMs.
Classifiers like classical NNs require:
- annotated data, potentially a lot of it
- training
- inference
#2 and #3 aren’t a big deal if you have an ML engineer, but #1 will always be a potential headache no matter who you are. The tradeoff is that they could be quite fast, cheap, and you can get probabilities, not just classes.
With LLMs you get:
- zero shot classification (no dataset or training required)
- potentially can use third party model providers like OpenAI off the shelf. Don’t even need to host your own model.
The downside to LLMs is that they are comparatively slow and expensive to traditional classifiers. Historically they also were prone to hallucination or malformed responses, though not as much these days. You also can technically get log-probs back, but these aren’t equivalent to the classifier probabilities.
Jev gets you the zero-shot, zero-infra benefits of LLMs, while being closer to the speed and cost of traditional ML classifiers, as well as both classification and probability responses.
Yeah but they weren't that great, you couldn't ask for arbitrary classifications after the model was trained. You are underestimating what they've done here, even if it does seem a little overhyped.
the algorithms technically, sure, however the outcomes definitely depend on data quality and coverage like any other training method, this is well known
Read the paper. They train RLCR on existing big math problems. They subtract a brier score penalty from the correctness reward. No new confidence labels are needed.
I did, in the first days Jev came out, when people were bringing it up. Another assumption. Please review the HN commenting guidelines, the one which starts with "Please don't comment on whether someone read an article." is relevant here.
Nothing in that paper changes that ML algorithms are dependent on the training data. We can step back from Jev and algos to consider Bayes Theorem. If your sample is not representative of the population, your resulting statistics will be off. The same is true here. If the data you train a model like Jev with is not representative, the probabilities and confidences it outputs will not be representative.
What makes Jev interesting is that it works well out of the box across domains. What people who are well known in the field believe is that this is the result of Typesafe having a really good training data set. People are saying similar of MiMo-2.6 today.
"Did you read the article" doesn't apply to a link someone put in a comment. If you are going to be a hall monitor, at least do it properly. You are just acting in bad faith at this point.
The relevant data is the reasoning trace. Doesn't need user data. You can learn from people's detailed reasoning steps how confident they are, even outside your domain.
We were talking about Jev and probability, now you're changing the problem, a rhetorical trick some people try to employ.
Another that uses dice rolling, coin flips, and an inventory level example to drive home the point that Jev's output are not real probabilities for outcomes.
I taught it (ML course; a day on RL, at a university), you should really stop making assumptions friend. Data quality and coverage matters in learning algorithms.
My initial comment and every one following is about RLCR and that paper. You don't appear to grasp the basics of that paper, it's reward function or how the optimizer is updating weights.
> You are out of your depth and grasping at straws.
Do you have any credentials or evidence that others can use to determine if this statement is not more accurately describing the author who wrote it?
Perhaps a PhD in ML, research output like published papers, or teaching/professional experience - all things I have
We could debate the merits of the paper contents, but I suspect you have intentionally moved on to personal attacks. Regardless, nothing you have said (nor can be found in this paper) has been a counter argument that learning algorithms are sensitive to training data, where the measured output difference is used by the optimization algorithm when updating the parameters. Garbage in, garbage out is a saying for a reason. No algorithm fixes non-representative data.
That's what I meant - we are currently use fp4 formats for training, and we cannot quite get away with that, despite dynamic quant and small block size - we still have to use quite a bit of higher precision (fp8 or even fp16) in various model components.
I might still be misunderstanding what you are saying, but bitnet also keeps high precision latent weights during training. The optimizer updates those, while the weights used in the forward pass are quantized to ternary values.
Doesn't really even need to look like it. If you can verify rewards, RLVR will optimize really really well. If you can't... it's a struggle. There are probably fewer fields where you can verify rewards than one might hope.
> There are probably fewer fields where you can verify rewards than one might hope.
2 tasks I've done today that I believe robots are nowhere near being able to do: Cleaning my wardrobe and draining bad fuel out of my generator. As in generic use cases.
Hard to verify that your wardrobe is clean. Also hard to verify that the bad fuel is out without physical sensors. Many, many tasks are quite difficult to verify beyond "you know it when you see it". That doesn't work so well for training a model.
Humans wear a lot of hats when the do work. They don't even realize how many. My experience with building real systems using LLMs is that you have to be very explicit about such hats and you don't realize how many are worn until you see edge case after edge case after edge case. Check this. Check that. Check this. Check that. Check check check.
As a class, they steer things through emergent behavior and overlapping self interest in certain areas. As individuals, they can’t afford to trust each other without leverage.
As a class, they have a set of similar incentives (grow/retain resources and power networks) that lead to similar actions (avoid taxation, buy influence, weaken regulators, etc). They may cooperate at times, but they are not a coherent organization with a roster, votes, and leadership.
As individuals, they are dangerous competitors, and they can’t afford to trust each other too much without some kind of mutual leverage. This could be in the form of mutual equity holdings, blackmail, complex arrangements where each party has exposure, whatever. But even then, trust has limits.
There’s no cabal that runs things with rules and elections at this level. It’s all individuals who are immersed in the systems that made them wealthy. Those systems tend to lead to certain behaviors. That’s it.
> You could make this statement about any group at all.
Yes, that's the point
> And then once you realize people are in multiple groups, the whole thing devolves into chaos.
You have bad priors. That would only happen if groups could only have singular and simple goals. It requires those goals to be mutually exclusive. Because as soon as they overlap, even just a little, cooperation becomes an effective strategy.
Hell, history shows us enemies often cooperate. Mostly to defeat a mutual enemy, so they can go back to fighting each other, but that's still cooperation
Nobody is suggesting nobody or group cooperates on anything ever. This comment could go in the definition section of "strawman". My comment was in response to the idea that there is "a conspiracy of the elite". It's just an absurd idea to anyone who has spent time in the world working with people, both "elite" and non "elite".
In the bay area you'll find thousands and thousands of people who have worked at a company run by a billionaire founder. They will tell you said billionaire founder can't even get everyone in the company to conspire to row in the same direction.
But the comment above wasn't about an explicit conspiracy, with billionaires meeting in dark rooms colluding. It was about implicit, uncoordinated conspiracy, where objectives align. My comment is defending the latter
Exactly, there is no conspiracy, there’s just multiple actors moving predictably in the same direction, and sometimes coordinating weakly. I don’t know how my comments could have been read otherwise.
True, but these are generally limited agreements among a few individuals (at most a small handful at the top). Pulling off a big conspiracy between many powerful competing actors carries too much risk of defection. It may be possible but it’s certainly rare.
I do think that these limited agreements may be copied by other people in similar contexts, for instance there were other tech firms who made similar agreements with their own competitors. And the big tech anti-poaching agreements may have themselves been copied from similar agreements made in other industries, such as finance and fast food. The illegal conduct probably spread organically through word of mouth, as execs and board members informally exchange this sort of “best practice” with each other.
It’s likely that big shareholders or board members could hold private meetings to advise the CEOs of their companies on how to pull off this sort of trick, if they think it would help the stock price. Given how many people sit on multiple boards, this could be seen as a sort of conspiracy. It’s still very informal and of course it’s nearly impossible to prove.
Maybe I'm overindexing on "they steer" and "effective strategy" - it still reads that this group has way more control and deliberate planning than they do.
And, "implicit, uncoordinated conspiracy" is an oxymoron. The word "conspiracy" has a definition.
If all you are saying is "sometimes billionaires are aligned" - sure. Sometimes politicians are aligned. Sometimes teammates are aligned. If one wants a platitude with some predictive power "people are self interested" is going to work more often.
“People are self interested” works fine as long as you extend the analysis to include the class dimension. People in a specific class are generally going to advocate on behalf of that class, because it benefits them to do so. No conspiracy needed.
To a degree, yes, it’s why looking at people on the level of groups is not that useful for predicting individual behavior and beliefs, only aggregate behavior and beliefs.
However, some groups do have identifiable leadership, rules, membership rosters, etc. Political parties for instance, or governmental bodies. Decisions from groups like this often have an identifiable lineage (from proposal to vote) and are considered binding. This contrasts with the sort of emergent decisions I was discussing in the context of the very wealthy. There was no vote, there was no formal coordination, and the “decision” is not binding, yet everyone can see its effects. Opposing something like this is like boxing with a shadow.
Or, someone who knows the books and hot takes on "systems thinking" are pop-science, and the actual good ideas have been around for a long time and taught in operations research, control theory etc.
It doesn't need any intelligence at all. There's mutually beneficial things to any group of people. So it doesn't require explicit coordination. Even fairly unintelligent animals have symbiotic relationships with other species. So your argument is what, that billionaires are less intelligent than an ant?
In the real world it's hard enough to get 5 people on the same page about anything. The idea that dozens/hundreds of people with billions of dollars who are in business against each other in many cases are secretly colluding with any effectiveness is a joke. If they were, zuckerberg, larry and sergey would have been squashed in the 90's/00's by the elite of the time. Sam Altman would have been crushed years ago. Obama or Trump would have been put in their place, probably both. Look at all the billionaires all of a sudden kowtowing to Trump. The billionaires just got beat real bad in NYC.
Billionaires just don't have much power. This isn't a movie.
This seems to contradict itself. You first claim that there's no organization but then you claim billionaires have no power because of the more powerful above them. Hierarchy is organization, so which is it
It really sounds like you're trying hard to define a square circle. Your argument isn't even self consistent because what you described sure sounds like someone has power...
Nope. It's perfectly consistent. No one has much power at all. No one is in charge, there is no greater power sitting above us all calling all (or even most) the shots. To use someone else's term, there is no cabal.
You seem to conflate "no one is powerful" with "no one has any power in any situation at all".
I never said nobody hoped to be powerful, or increase the level of power they have. They just don't have much, and they can't/don't coordinate like some conspiracy theorists believe.
reply