it also seems like this should scale and be practical if you just pick any sensible coding language that can be used to access the internet fairly easily this is the coolest paper in a long time imo every greenfield continual learning neolab should be taking note

if i had to steelman the general usage of OPSD, my recommendation would be: - use judges to identify behaviors in rollouts which expressly violate guidance which is *already in context* (e.g. system prompt rules, tool schemas) - insert a verbatim reminder of that guidance right before the failure occurs - train only on the tokens immediately following the reminder this avoids hint leakage, and combats the legitimate problem of rule adherence in long-context settings. however, you can also just do this with RL, and it's not clear why you should expect OPSD to be any better than folding those same failures into reward penalties. it's also training the model to take actions which might directly conflict with preceding reasoning, which could be an issue for CoT faithfulness if that's something you're concerned about. tool schema failures are the clear win, but if your model is regularly messing up tool calls, there are probably more bigger problems in your post-training to solve. i think it's pretty unlikely that any serious frontier lab uses it in a real way other than as a last-minute band-aid on weird formatting bugs which show up in final testing states. it's not bitter lesson pilled. some people like to think that *specialization* isn't bitter lesson pilled, and i very much disagree here. the most intelligent and successful humans are usually exceptionally specialized and spiky. somehow we convinced ourselves that being superhuman at *everything* ought to be a free lunch, but there's not really any historical precedent or theoretical argument for this being true in the limit. the success of frontier foundation models isn't really evidence here, they're *way* more expensive to train than specialized counterparts, but final-run training costs are dwarfed by research and inference, and it's much easier to ship a one-size-fits-all product. jev is a hit because nobody ever made BERT finetuning easy enough for non-expert developers. real continual learning has never been tried. we don't even have the cognitive core nailed yet.

what did you think open superintelligence stack meant? vibes? essays? we’ve had this as the north star for a long time, we’ve come a long way, and there’s so much more to build and research we’re hiring btw primeintellect.ai/careers

it reminds me of across the sea by weezer

one theory of mine re: new opus is that anthropic finally managed to properly scale "taste RL" which is noisier and more subtle than normal RLVR and just needs a crazy amount of rollouts, hence doing it on a "smaller" model + after getting more compute + inference gains