Today we're launching Base Labs, a research lab by Baseten. Our mandate is to make open-source as useful as possible, and our one rule is that we publish without exception, including what fails. Up until pretty recently I thought the way to get the world onto open models was to train them for one company at a time. @mudithj, @maxkirkby and I cofounded @parsedlabs on that bet, @baseten acquired us, and we spent the last year running their training team doing it for customers one by one. Every one of those engagements taught us something new about how models learn, forget, specialise and get cheaper, and almost none of it got spoken about. Sadly, in general that's the field's default in that the people who know the most about training have the least freedom to say it. To be clear I don't think the closed labs are the villains here. They get to new capabilities first, which buys the rest of us time to harden the world before that stuff is everywhere, and they're the ones paying to find out what's actually possible. But I'm fairly convinced the only real advantage they have is data and scale, and their incentives point squarely at the frontier. You can't do slow, public science on how these things learn when your job is the best model by end of quarter. Someone without that pressure has to, and there are very few of those someones around. Hence Base Labs. We have the broad remit of making open-source as useful as possible and our one rule is that we publish without exception, including what fails. The first problem is continual learning, which I have come to think is several problems wearing one name. We are also working on the open RL environments and data that open models need and currently can't get, because we have to aggregate data with the same ferocity everyone's been talking about aggregating compute. Plus a bunch of other stuff I'm genuinely excited about, eg a safety stack people can run on top of open deployments, and performance research so these things are cheap for everyone to serve. I still think open and closed coexist, and that's the good world. It's just that coexistence isn't free, someone has to actually do the work, and this is basically what keeps me up at night. We're hiring researchers, engineers and fellows. Come help distribute the mandate of heaven!

sachin and adam make the best videos. hopefully the our chats do their production quality justice

OpenAI serving open-source models through Baseten is a big deal. They are embracing open source. Until now, OpenAI's pitch was "use our models." Now it is "commit your annual AI budget to us". We will get you the best model for each job, whether it's ours, someone else's, or open source. So if you're a large bank looking to commit $100M in AI spend, you centralize with OpenAI. One throat to choke. GPT handles the credit memos. Open-source handles the KYC volume at a fraction of the cost. A post-trained / specialized model handles fraud. This changes OpenAI's position. It stops competing only on whose model is best this season and starts competing to own the enterprise AI budget. A layer above being just an AI lab. Also proves AI is positive sum, not zero sum between closed and open source!

OpenAI <> Baseten 🚀 Customers can now use OpenAI commits towards open-source tokens via @baseten!

opsd and its variants don’t work, no matter what they are biased and bias kills llms, please please please stop working on it, so many other great directions to explore that aren’t opsd

The sun has finally set on the British empire. This is genuinely so sad to me. The country that gave the world the steam engine is now treating the most revolutionary technology in history as a dirty, ugly tool. Straight out of atlas shrugged

in fairness to openai i think an xgboost model could hack medicare

Distillation is used to warmstart your model for RL. But there are many unanswered questions. What is the exchange rate of distillation and RL? Does distillation burn entropy and thus cap the model's ceiling? Are there behaviours you can only learn if you distill from a stronger model, that RL will never find? This is the first step towards starting to frame and answer some of those questions. Thank you to @dwarkesh_sp for many interesting discussions which motivated a lot of this research

if this doesn’t wake you up to the fact that we have officially unleashed a new scaling axis I don’t know what will whenever log plots break along some stable metric but overall progress feels like it’s improving at same expected rate, this is a good sign you’ve got a new scaling axis. We’ve seen the same thing with Moore’s law over and over

Happy to start collaborating with @baselabs from @baseten and @GoodfireAI to push safety and interpretability for open models. As open-source models catch up in performance to frontier models, the community at large have a great opportunity to establish a common practice for effective safety, security and interpretability research. Some past discussions confused “open” with “unmonitored”. On the contrary, an open model provides much more tooling and visibility for safety research from a broader audience, which has led to a lot of the safety and security techniques we use today. Linux is a great example of this: open-source and secure deployment are not only compatible but heavily intertwined, as a properly secure system needs an extensive feedback loop of finding and fixing vulnerabilities. We're excited to share more soon.

Proud to partner with @baseten and @baselabs to build safety infrastructure for open-source models—which are essential to lots of safety research, including our own. Safety must be built into open models and provided by those who serve them, and we’re excited to help enable that!

Safety is not just for closed models. The closed frontier labs are a canary in the coal mine for what is coming at scale. They give us a glimpse into the future and a window to harden our systems and prepare for abundant intelligence, with all the risks that come along with it. The OpenAI agent swarm attack on Hugging Face is the kind of failure we need to prepare for as open-source models catch up. The providers serving those models (such as Baseten) have a big role in establishing what safety and monitoring standards look like. We’re proud to be taking the lead on this at @baselabs with our collaborators @huggingface and @GoodfireAI. We’re developing safety research in the open and building it directly into Baseten’s inference infrastructure, with the aim of making it available to all our customers. This includes training models to follow explicit policies, detecting failures at runtime, and connecting those signals to controls that can intervene. We invite others in the open-source ecosystem to join us in building the tools and standards we’ll all need.

and when you realise most rl envs are just LLM evals but slop volumed, a lot of things (eg misalignment) start to make a lot more sense

This is the last time you’ll be able to spot this. It’s like AI generated images start of this year. You’ll stop spotting it, because people will get so good at it. That is, because every single product, experience, game, thing will become so well engineered at capturing real world inputs to seed RL envs / midtraining data. There will be whole design philosophies on doing this without the user realising. Dark user patterns will abound. The world will become a factory for RL envs

very kind from AT but most importantly I agree that someone should figure out the contours of this conflict: "on one hand: <> “anything that can be learned through RL can be distilled very easily”. this is the justification as to why open source models (chinese) can catch up to closed source models (american). fair... 5 minutes later <> “Opus 5 could not distill / generalize well from Fable [despite anthropic having the live deployment, access to logits, and obviously a good prompt distribution]”." we are working on a version of this at @baselabs

I can guarantee that google is behind OpenAI and Anthropic in terms of rsi

I think a downstream consequence of increased commitment to safety and alignment is that RL envs companies get screwed, or at least held to a much higher standard and thus their unit economics changes. The evidence is mixed on this early but it seems likely that impossible to poor quality envs lead to misalignment and poor behaviour in the models (as they try to do anything to get reward). So why would the labs risk this quality control with anything but big in-house efforts moving forward

A flavour of our research interests on the Dwarkesh Podcast this week, discussed by our very own @oneill_c. Just the start of a longer conversation about long-horizon RL and frontier open-source training here at Base Labs.

For better or for worse, chamath + All In pod is somewhere that so many people, particularly non technical, get exposure to information and opinions on AI. Right now is a perfect opportunity to get more informed and thoughtful voices on there

really cool, didn't know @oneill_c was Australian until i heard the accent I hadn't really considered the prompt distribution and the usefulness of realistic data in that regard and how the distillation goes through these channels, its a good explanation of why Chinese models are so good compared to like sonnet or opus though.