News for hackers. Stuff that matters.
Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by janalsncm on Tuesday September 15, 2026 @04:56PM (#49718807)
Technically speaking when you send the prefix “The capital of France is “ into an LLM it will also produce probabilities across its whole vocabulary.
Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by sothatsit on Tuesday September 15, 2026 @07:10PM (#49720113)
The probability values don’t really represent confidence in modern LLMs though, especially after RLHF and RLVR.

System One says they use RLCD, Reinforcement Learning for Calibrated Decisions, which presumably has accurate probabilities as an explicit optimisation goal.

Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by nkozyra on Tuesday September 15, 2026 @11:05PM (#49721644)
How is that different from RLVR?
Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by sothatsit on Tuesday September 15, 2026 @11:42PM (#49721843)
RLVR generally upweights tokens along the whole thinking trace that led to a correct answer, whether each token was "correct" or not. RLVR doesn't train a model to output an 80% likelihood, it just trains it to produce correct answers, and not to produce incorrect ones.

System One hasn't said how RLCD works, but they do say it is explicitly training models to output "calibrated" probabilities, which makes it distinct from RLVR. This is how they describe it:

> System One models are trained for calibrated decisions: their probabilities are optimized against outcomes to reflect uncertainty.

Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by porridgeraisin on Thursday September 17, 2026 @05:47AM (#49738448)
In RLCD (which is now an RL acronym that has 3 different unrelated expansions!), you basically massively negatively reward a distribution that is {yes: 0.9, no: 0.1} if the answer was no, and less negatively reward a {yes: 0.6, no: 0.4}. Many nuances when designing the details, but that is the rough idea.

It is a known existing thing variously called "calibrated RL" or such.

Implementing it on top of LLMs was difficult to get it to work, they seem to have done it up so its good enough for a polished product that works in a wide variety of usecases at the same time. I got accepted from the waitlist and it's really neat. Edit: it is now on vercel gateway.

One thing to note, the out of distribution behaviour will be different from what we are used to with regular LLMs. Theoretically, it should be worse, but practically, it depends on their method.

Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by jiggawatts on Tuesday September 15, 2026 @06:52PM (#49719949)
… which they could provide in their APIs but are vehemently opposed to because it makes distillation much easier, and faster.
Re:Jev: New frontier model 40-400x cheaper and 20-200x faster by bigglebear on Tuesday September 15, 2026 @08:39PM (#49720742)
Yeah. Yet another reason why open-weight models are better. If I want to use the logits, I can.