wxw

> Under the commitments that have now been declared binding, Apple will align the consent prompts for its own offerings and for third-party apps much more closely.

Love this opening line. And wow, great results.

> As agent capability improves, much of the difficulty in scaling post-training moves from the model to the environment.

And the actual repo: https://github.com/xai-org/x-algorithm

> We’re releasing the code that determines how posts are ranked and filtered in the For You timeline, including the content labels that can impact post visibility. This builds on our previous open-source work, adding model configuration, filter and core ranking system details.

Awesome work. I'm personally very excited for faster models/inference.

I think speed is underrated to some degree in the current conversation. For a while, I was using Cursor's Composer quite a lot, even over frontier models, just because of how darn fast it was.

I've always considered the Flash series of models to be for low-cost, high-volume, mostly text-based use cases (e.g. summarization, parsing, formatting), emphasis on low-cost.

[edit: ah, benchmarks here: https://blog.google/innovation-and-ai/models-and-research/ge...

more of a Terra than Luna competitor which is an interesting positioning. I feel like differentiation at the mid-tier of models is pretty difficult.]

Great line. I do really love their fries though.

She spent 6 years at Meta. I get the sense there are other factors at play than "ethics departments are just PR stunts". There's no way she isn't already aware of that. Unfortunately, the article is light on any other details.

I'm all for it though, and I think Glimmer is a fantastic bet on locally-hostable models. I for one would love to self-host as much as I can.

> Freedom

> Liberty

> Serenity

Reads like a state motto

I'm not sure I follow how this is actually guaranteed? The fact-checking process mentioned just seems to involve asking AI to review its own work.

Yea I feel that.

Really enjoyed this essay. I feel like “you can just do things” is so true yet has been twisted into a hustle-culture/opportunity-vulture phrase, and this piece is a much healthier way to view agency.

e.g. you can easily reduce the intensity of feed-based UI by removing thumbnails, grayscaling, etc.

I think the more interesting lever is the fourth they mention: token efficiency.

> By the time costly LLM inference occurs, the user's initial statement accounts for only a negligible fraction of the data fed into the AI system, meaning costs are dominated by context the user did not explicitly include.

I think there’s still lots of low hanging fruit in regards to monitoring and improving agent work. Look at your sessions. Look at how much time and context is being spent on, say, a web search returning dozens of results when one good single-pager doc would’ve been better.

Never would have thought of a Mobius-strip though, insane.

> FreeFormatter.com [was] a popular online utility website that provides free formatting, validation, conversion, and minification tools for developers, data professionals, and web administrators working with code and data structures like JSON, XML, HTML, SQL, and Java

Makes sense. Great exit letter.

> Axon, simply, does not have a national network. Meaning that it would not be possible for cops in one jurisdiction to search cameras across the country. [...] Police departments using Axon can enter into data sharing agreements with each other, though.

> Denver, just like the City of Ferndale, explicitly mentioned this lack of this national network as part of its decision to ditch Flock for Axon.

This (probably alongside all the bad press Flock has gotten) appears to be the crux of the switching. Ultimately, the tech actually being deployed seems similar.

Also interesting:

> One reason why journalists, activists, and residents have been able to learn so much about what police use Flock for is that the company has built auditing tools that the public can access through records requests.

Any insiders know how Muse Code is doing internally?

It makes sense that they're leaning into an all-in-one platform play, i.e. if you use DOs, you might as well use our wallet/sandbox/AI gateway/etc. too.

> The settlement addresses allegations that both companies violated the Immigration and Nationality Act (INA) by discriminating against U.S. workers and instead preferred workers with temporary employment visas [...]

> In addition, OpenAI took other steps to discourage U.S. workers from applying, such as advertising positions on the radio late at night.

They knew what they were doing here and probably don't mind being fined said amount.

> The Open Source Sabbatical is intended to provide professionally qualified software developers with the opportunity to work on an open source project for a limited period of time in order to improve it.

> This sabbatical is not only open to employees of the City of Munich, but also to external software developers.

So cool. Congrats to Sebastian!

> I propose a goal of creating Guardian Angels (GA): digital twin LLMs which are personalized with the goal of providing not the stereotypical “assistant chatbot agent” persona, but emulating a single user’s personality, values, and preferences.

> A GA persona is productive because it learns to emulate the principal’s outputs but with higher quality. It is trustworthy because it is, by definition, allied with its principal and shares its values and goals. And it is secure in part by hardwiring a single, unique, situated user (for whom following a prompt attack would be absurd)

> We can try to create GAs by a combination of techniques: online learning (via dynamic evaluation) to update LLMs in realtime to avoid ignorance and fatal errors while remaining competitive with frozen frontier models, sample efficiency from pretrained preference-oriented large models and active Learning by querying the principal for corrections and preference data (obtaining low regret from DAgger-style bounds), and a local CLI-first logging-oriented UI/UX paradigm.

I don't really know or follow Gwern. From reading his full post, it's an interesting idea and seems like the broader goal is moreso safety & alignment which is a new angle for this category of product.

Count me in.

Nice!