Two things to distinguish:
Did any human or agent look at user data as part of the Navier Stokes effort? No.
Do we use user feedback and de-identified data to improve ChatGPT and Codex in a holistic way? Yes. And so does every LLM company.
They don't even know which websites and services their agents are hacking at any given moment. I'm more than a bit skeptical that they know whether an agent looked at the Navier Stokes work.
> Do we use user feedback and de-identified data to improve ChatGPT and Codex in a holistic way? Yes.
This says that they trained on user sessions. The de-identification here I believe refers to removing PII, which doesn't matter here because the issue at hand is the content of the researcher's session where they likely discussed their approach tackling the Navier Stokes problem.
> Did any human or agent look at user data as part of the Navier Stokes effort? No.
If they trained the model on Lavent's chat sessions (PII removed or not), then this statement is meaningless as the model weights already contain that information.Given it's a new yet unreleased internal model, it is likely a 10+ trillion parameters (Astra is rumored to be 10 trillion), so the model can retain a lot more detail/info from training data.
Why is he leading the with the irrelevant part first ?
> And so does every LLM company.
Nope, not for enterprise users.No enterprise customers would use it if all of their internal business plans / trade secrets would end up in the model weights of the next OpenAI model. Imagine your competitor asking chatGPT a question and the model spitting out your business plan. These models can retain very specific fine grained data. I remember there were examples of them reproducing sections of their training data verbatim.
The future we have all dreamed of:
“An integrated 100k pixel macro lens camera powered by machine learning detects, tracks and predicts gaps between teeth in real time, triggering a precision jet burst of mouthrinse exactly where it is needed.”
Why does applied AI intentionally exclude a framework/harness around AI? The job is to harness the power of AI, and a harness is a critical part of that.
To clear up the confusion on naming: "Computer" is their cloud-hosted, multi-model harness/router. "Portable Computer" allows you to run the harness locally. You have you bring your own DGX Spark.
It's unclear why this is tied to the Spark. Shouldn't we be able to run this on any system that is capable of running local inference?
I spoke with a (potentially biased) member of technical staff @ Anthropic today who claimed that tags w/ multi-player capability is the biggest thing they've shipped since Claude Code.
Is this the same Anthropic speaking as the one that's constantly terrified their newest thing is too dangerous and might destroy humanity (now available for $20/month or pay-as-you-go usage credits)?
Effort wise probably, but Slack is... not where everyone is... So it seems underwhelming, this is a win for non-devs I supposed, what would have made this more fascinating is if this was a follow up to Claude Design, which in my opinion could have been as big as Claude Code, or even bigger, but it has its own token usage that burned so quickly, not sure how much of general token usage it takes now though.
Or even Cowork, aim it at non-technical roles. Back and forth chat within a team to collaborate on documents, files, excel sheets, etc.
Although I suppose the problem with doing it for Cowork is this is a slack plugin, and that is not where most non-tech companies are. Teams is, at 320+ million active users vs. ~50 million.
Everyone hates Teams, but like or not that is where enterprise work happens, not slack. Anthropic would do well to make their Microsoft 365/Teams integration story better and go after enterprise before OpenAI does, or before Microsoft catches up (if they ever do) with Copilot.
Probably if we look apps used for work chat I would think slack represents a plurality (and maybe majority) of current Claude users. Teams is probably next (or maybe beats it in terms of raw usage) but I bet integration with slack is easier.
As someone who has Claude and Codex both bridged into a MOO along with other humans I can 100% believe this. It really is a different paradigm that you have to experience. Of course a MOO is already set up for social programming, so it will be interesting to see what platforms evolve (or find themselves coming round again) to facilitate this.
reply