News for hackers. Stuff that matters.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:28PM (#50552360)
I always wonder what you lose when you start maxing out some specific benchmark. It is really hard to do this in a balanced way and I would love if everybody that did these kind of efforts would take the time to also list the before/afters of a number of other benchmarks that they did not use in their training.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by gardnr on Wednesday November 18, 2026 @04:42PM (#50552521)
Check the article. It’s the first graphic that has other benchmarks.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:52PM (#50552697)
Yes, they all go up. I am wondering what was lost.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by nblintao on Wednesday November 18, 2026 @04:46PM (#50552603)
Good point! First of all, none of the benchmark tasks were used in training, though the training data was designed to improve capabilities on tasks like TB4.1. If you consider TB4.1 the main target, the blog actually shows before/after results on several other benchmarks, with no regressions and even some improvements.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @04:51PM (#50552684)
Yes, I've seen that, but I wonder what is missing because I find it hard to believe that you can just do some RL and then end up with a model that is better across the board. I would find it far more believable if you outperformed on the benchmark that was your goal (assuming you selected training data to give you an edge there) and underperformed on things that you were not optimizing for. But since it all moves in one direction (better) I feel there must be some kind of unknown at play here. Even so, task specific AI is definitely the future, you don't hire a plumber to do your bookkeeping either...
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by reissbaker on Wednesday November 18, 2026 @04:58PM (#50552785)
Why do you find it hard to believe that you can do some RL and improve across the board (albeit less strongly on other semi-related tasks than your objective)? That's the idea behind generalization: it's not perfect but you get a little bit of cross-task generalization, so if you just keep spamming more RL environments you can hill-climb to AGI. That's what the labs are doing!
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @05:05PM (#50552875)
You can hill climb but you will (or at least, I think you will!) lose generality.

And no, that won't get you to AGI, that is a completely different goal.

Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by banghua13 on Wednesday November 18, 2026 @05:25PM (#50553176)
If you carefully curate the data, plus the original model isn't saturated for RL post-training, it's possible to improve targeted benchmarks without regression on other metrics. This experiment is done with terminal bench 4.1 as a target, and the regression on other benchmarks has been carefully controlled through data + infra design.
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @05:34PM (#50553282)
Ok, so the obvious question then is: this seems (I'm not saying it is, it just looks like a small thing compared to training up a model from scratch) a relatively trivial undertaking for the authors of the original model, if there was this much left in the can why didn't they do something similar? Or is there some step here that only you can perform? Or is it simply a matter of economics?
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by mik1998 on Wednesday November 18, 2026 @06:28PM (#50553826)
[delayed]
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by jacquesm on Wednesday November 18, 2026 @07:01PM (#50554114)
That mirrors my experience exactly. I keep a couple of older models (and much larger ones as well) for when I want to do stuff that is unrelated to computer programming because the models that are really good at that tend to fare less well at other stuff. I suspect that there is a boundary where maximizing benchmarks for coding at the expense of general knowledge is going to give you a model that is really fast and good but 'narrow' and that this may also lead to loss of performance when applied to coding tasks that are less well represented in the training set (such as, for instance, DSP code).
Re:DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12... by hrncn on Wednesday November 18, 2026 @04:49PM (#50552656)
At least so far I am not aware of any generally usable small models that were actually improved except for Thomson Reuters qwen, but even that improvement only applies to specific topics. Everything else was pretty much a waste of time due to benchmaxxing and all kinds of weird issues caused by too much RL.