NitpickLawyer
today at 6:42 AM
> Labs have already used up internet-scale data
Despite this being the topic du jour of 2025, it was never true. Most of the "we've hit a wall with data" came from communicators / media and not researchers. It got popular because negativity sells. It's a false premise for a number of reasons:
a) Data curation is as important, if not more important than bulk data. Models becoming better at classification leads to better curation leads to cleaner data. Throwing common crawl and pray is so 2023. We've known this since llama3 days, it worked then, there's no reason to think this will not continue to work as the models imrpove.
b) Models are today good enough that you can augment / multiply your data easily with enough compute. You can now have a model take "authoritative content" and create more data from that + scenarios. Say you take a book on computer architecture. You ask models to break it down. Then you ask models to find examples for each topic. Then you ask models to ask questions and offer answers from several viewpoints. Then you take each of those and ask other models to flag inconsistencies. And so on. But you can whateverX your data from one authoritative source + bulk data into 5x - 10x "scenarios".
c) RL is really really really powerful. It's hard to do right (reward hacking, instabilities, etc) but once it works it "keeps" on working. Again, we knew this to be true a few years ago, ever since models really started to do well on math (highly verifiable). It only follows they're getting better on cybersec and other verifiable tasks. But now, with models improving, you get the same data augmentation pipelines as above, just better because they're also verifiable. For example, the way cursor augments their data: take a repo, ask an agent to identify a feature (it can be a large multi-file feature). Remove all code relating to that feature, but keep the original tests in the repo. While training, that becomes a RL scenario: implement this feature in this repo. Verify it with the original (hidden for training) tests. Reward appropriately. Now you can get 1 repo -> 20-50-100 scenarios. Instead of "feed everything into the pretraining", you're now creating scenarios, verify them w/ existing tools, and get your scoring function for the rewards. And, importantly, as the models become better in general, they also become better at this pipeline building exercise. So the next iteration gets trained on more scenarios, better scenarios, and so on.
> how will models continue to get better?
Probably the same. No one can know for sure, but at the moment, despite all the "walls this, slowdown that, plateauing" and so on, there are no signs of slowing down. And, as you noted, this works across the field of model sizes. There are, of course, theoretical information-based limits on size, but smaller models also improve, once "bigger" models can be used as training data generators, oracles for verification, rubric verifiers for open ended questions, and so on.
And smaller models (i.e. cheaper to serve) get to generate more traces during RL, and more rollouts give you better training, and so on. Next up - hardware optimised inferencing (ASICs basically). Once you have that, we can expect another wave of improvements. And so on.
CuriouslyC
today at 2:53 PM
Model output is pretty mid at augmenting, it can lead to distribution collapse. It's useful for smaller models because nobody wants manually to curate a specialized corpus and those models can't represent the diversity anyhow, but if the plan for infinite scaling was just to keep feeding the biggest model more of its predecessor's slop, that's not going to work out so well. It might work as a supplement for "thin" areas that have outsize importance for the amount of training data available for them though.
Big models are going to "tap out" on non verifiable fields within ~2 years, just because the pool of experts able to reinforce the models is going to get very small, and as the nuances get finer, the signal from reinforcement is going to get progressively less aligned with the intent. Math and code will be mostly tapped out in that time frame as well, even though we can technically scale them "infinitely," just because the cost benefit won't line up. At that point, most RL will be "gyms" with games that are designed to model designated valuable economic activity.
In the next few years, we'll get small domain specific distillates that are ridiculously smart in their domain (imagine if Qwen 3.X 27B went super saiyan), and even frontier labs will be routing to experts/orchestrating because the cost to serve/TPS difference is huge. They'll still train the god models for PR/marketing, c-suite use and distillation, but using them for day to day work would be like making houseware out of solid gold.
Refreshing to see someone actually understand training rather than treat it like dragging and dropping "internet.zip" into the LLM "knowledge" folder.
Gecko4072
today at 6:59 AM
Thank you for your response. Part c was especially insightful. Quite a smart way to do it and makes the possibilities of post training seem almost endless. Makes sense that you just need more time and compute.
A positive feedback loop then. RL->better model->better RL pipeline -> better modelâŠ
And weâve only recently started getting into the much better RL pipelines