meric_

But it of course is large and heavy like the quest pro and expensive most likely. Things that when you market to consumers don't pan out.

So they are optimizing on the other dimension (weight) which of course trades of many other dimensions.

(Except for of course Mythos and whatnot when they want to push the whole "safety" thing)

https://openai.com/index/where-the-goblins-came-from/

Small quirks can quickly add up in posttraining if not caught. Although TBH with how obvious Claude language is, I do feel like this is something Anthropic probably noticed and just assumed people would not care about. Now that people have obviously cared, they're probably actively looking to alleviate it

Will be interesting to see how people's opinions of it line up IRL, but so far I've loved Fable so hopefully will love this one too

But also effectively this is a classification model. It excels at specific certain types of workloads, and obviously will fail at others. Not really sure how one benchmarks this tbf. I can see their argument on why this requires a novel specific eval for whatever your usecase is. A consistent "global" benchmark might be hard to do

Some quick googling shows the ARPU of a meta user is approx 57$ a year, and ~65$ in NA. Even if the quest provided useful advertising data (it doesn't) it would never come close to 600$ over the lifetime of a user.

I don't even understand what you mean by "biometric data" for advertising purposes. What biometric data is the quest collecting? How in the world is biometric data useful for advertising?

There's a reason why Meta is tens of billions of dollars in the hole for reality labs. It's not to subsidize for advertising that's for sure.

Maybe a little distasteful since I suppose since these WordPress sites are not owned by WordPress, but I wouldn't really say it's unique or necessarily poaching

Muse code: https://developer.meta.com/ai/resources/blog/build-with-muse...

> Co-trained with the harness. Muse Code was in the training loop from day one, so tool calls succeed and plans execute cleanly. Crucially, we trained across multiple harnesses, so while the model is at its best in Muse Code, it still generalizes to other coding agents you already use.

Are these not the same apps

Similarly stores with loss prevention guards are probably losing money, but I'd much rather that than having to click a button to unlock items, etc.

Now is that enough revenue to earn back 7B.... I'm not too sure. I guess they're betting realy big that these sorts of model routers will gain explosive token usage in the future

The only downside (and a big one) is that by not being natively offered by OpenAI, Anthropic, etc. you're paying @ API billing and not plan billing, which makes it less appealing. (Of course Manus has a wrapper around this w some token system, but it ends up being expensive)

This truly is a case where AI is the solution. There's no way to automate this at scale with humans. As AI gets better hopefully detection and moderation will catch up

The title makes it sound like they actively seeked to silence a journalist, but nothing in the actual article correlates with that. In fact it's not even the title of the article nor claimed at all in the article.

There's no fancy AI needed to OCR out a 10. It's quite believable at least from Flocks perspective that it's the way the police want it set up. I mean the fact that officers typo'd in the wrong initial value seems to support that.

This is a problem with the cops. Not flock.

Sure for old tasks you could argue that now its not required to boost because infra errors are alleviated with better default limits. My point more so is that its a strange thing to index on because if you wanted to cheat on the benchmark, it does not particularly seem like something that shifts results? Once the API is out maybe I'll eat my words, but I don't really believe that if you manually tried to reproduce the results with lower limits you'd see significantly different results

https://www.anthropic.com/engineering/infrastructure-noise

Is anthropic benchmark maxxing and cheating on terminal bench too? They don't follow the strict resource "limits" either