danielmarkbruce

I find this too. In fact, recently I've been pushing more and more to the latest and greatest model every time there is an update. It just saves me so much headache.

The gambling platforms are a new thing in the US, voted on at each state in the past few years. For example, in california it was a proposition back in 2022 (proposition 27 i think, which was defeated thankfully). Several states put it directly to vote.

We are responsible, and it doesn't appear you have any understanding of how it all came about.

Read a little on the history of gambling in the US. It was illegal in almost all the US until recently, and could have been contained.

If you are in the US, you are to blame for the US gambling situation. We make the rules as a society.

We are unethical for letting it get to the state it has. The voters define the phrase "money talks".

Society (i.e us) is to blame. We made this stuff legal. There was no need whatsoever. We don't get to blame the people working at these companies.

I'm saying something even worse - humans are bad at decision making, we usually don't even most of the context required to make good decisions so we use heuristics to make decisions, and proxies for "good decision making".

If the above appears to be bizarre thinking, take a couple extra laps around the block.

Now I'm not saying AI is better, in fact above I mention that it's worse - but don't hold onto some nostalgic idea that decisions were made by people with lots of context thinking rationally and incentivized properly.

They post train on big math and improve calibration across five different non math benchmarks so the new claim "can only give as accurate predictions and probabilities as the underlying data they are trained on represents" is also off base. RLCR generates its own calibration examples from ordinary questions and answer keys. RL usually isn't trying to represent a data set, it's closer to search.

It's more nuanced than that - I never said faking it, people usually believe what they write in those documents/ppts. And they do actually sound clever. Next time you see a decision of any magnitude, especially something which sounds "strategic" - trace the decision back as far as you can, consider the question from several angles, and really question it. You will see gaping holes, decisions made using "frameworks", anecdotes, hopes and dreams.

People get fired effective immediately.

In many (most perhaps, but hard to know personally) companies, most people don't really understand the market they are in, the competition, their own products, the customers etc well enough to actually make good decisions. They also aren't likely to be around (or held responsible) for decisions as they play out over years. So what has historically happened is that people use proxies for good decisions and understanding - which are clever sounding documents and presentations.

This is a long winded way of saying "people made up clever/sensible sounding stuff". Now it's easier to do it with AI so the problem is worse. However, I'm not sure what you were looking for was ever really there - the "inscrutable machines" and "inscrutable incentives" were always quite inscrutable.

The end: make lots of money. The means: systematically take existing reasoning models, do some more post-training of some sort to make them achieve the same outputs with less reasoning tokens (ie, cheaper). Same quality but cheaper is always valuable.

It's unclear if they can do this systematically and it's unclear if they can do it better than others. But, lots of things are unclear in AI at the moment, this doesn't seem outrageous on the surface. And, it could just be marketing. And it could be the first option with the backup of the second.

>you have to have training data with accurate probabilities

This was your claim. If you can't read and understand that paper in relation to your claim, you are out of your depth. You haven't made a single claim relevant to that paper - just hand wavy comments about data.

>> I think suggests to me a level of consistency in the contextual environment that would probably never exist

This is, basically, 100% of the thing. We will never get to the level of automation some folks think for this exact reason.

This is true, and it's strange that people don't think much about the fact that approximately zero % of people on planet earth are actually trying to cure diseases.

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.

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.

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.

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.

Take RL 101. This is a common pattern.

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.

"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.

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.

Existing datasets, different reward function.

I don't think you've ever done either of these training steps. You are just handwaving.

RLVR and RLCR really don't need a whole bunch of special data.

Sure, and most days it doesn't rain.

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.

For certain tasks, it seems much, much more efficient. That's not nothing. People have been using LLMs for various classification tasks.

No, you don't. You do RLCR, similar to that proposed here:

https://arxiv.org/pdf/2507.16806

The claim of how they are doing it is likely wrong.... if you had to bet, it's likely an encoder model of some sort.

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.

Never said equal, just that no one has much. It's just not that hard of a claim to understand. No need to twist oneself into a pretzel.