striking

The way I see it, either you are curious enough to download some of the files or you are not. I don't understand this middle ground where you spend more time complaining about the distribution method than it would take to download. It feels contrived or even in bad faith, especially given that the link above would allow you to do exactly that without starting a torrent client.

If you're so curious, why not have a look? The files are pretty obviously available, any online magnet link viewer will show them to you https://webtor.io/db503db679f508aa31a9e07913b2aa8cf7aa6dff?p...

I'm looking at the "Add credits" screen on https://platform.claude.com and it prominently displays the "total due" as $504.00 when typing `5,04` into the input...

https://github.com/huggingface/funes/ has been my go-to here (specifically for conversation search), but I do appreciate that there's a frontend on this one that can recall a session and open the correct agent for it (rather than my opening one agent, then opening another when I've gotten it wrong). Looks nice. Well done.

It was me. Just wait until they find out which of the 1920s issues I took.

Pangram thinks it's a mix, with around 40% of the text being AI. This was sort of my feeling about it even before running it through a checker, that a human wrote or rewrote some segments but others do feel distinctly high in perplexity and were potentially written by AI.

https://www.pangram.com/history/c150e3e0-7173-4cbb-9b38-41dc...

That being said, does it really matter? It comes off as bad writing. The mixed metaphors are hard to follow and detract from the point.

Did you dodge the phone price increase somehow? Phones also need RAM (and ambitions of running local models)

Any tips on GI impact? I pair mine with food and lots of water and it's improved but wondering if there's any more I should be doing

Going from minimal exercise to light-moderate exercise will do as much for you as creatine would. To me it's almost a no-brainer to try doing both.

You say that, but if your SF tax dollars were paying for VTA you'd probably have stronger complaints... Moreover, public transit isn't evenly distributed even within SF so I don't think the transit agency count is an important variable.

They're all tied together through Clipper, so it doesn't make a difference to end users anyway.

No surprise then that the default effort level for this model in Claude Code is Medium, even if you had Opus 5 set to High...

> If you think you have found a break email us at doh@raspberrypi.com with details - we will ship you a Pico2 with a custom secret hidden in it. If you manage to extract it, you win the $20,000!

It's not quite the same, but the in-flight chess game provided by Delta was known to be absurdly hard: https://news.ycombinator.com/item?id=46593395

I also really quite like the Fitbit Air. I used to wear a Galaxy Watch but there's something really freeing about how light and unobtrusive this little wristband is.

Snap implemented these features and others long before Scratch had. To my memory it might have been a decade earlier.

Disallowing LLM contributions doesn't disqualify the use of LLMs to identify vulnerabilities.

Where does it say that domains you own can be used? I'm looking around and nothing clearly states it afaict.

> This change does not affect Google Workspace aliases or other Gmail addresses you own.

to me, implies that you need to set up Google Workspace for that domain before it'll work.

I think that's a worthwhile point to consider but it's only relevant if we move the goalposts from "GrapheneOS is only used by Android ROM enthusiasts" to "GrapheneOS is only supported by one small Android phone manufacturer".

To be frank, though, I don't see any of this line of reasoning as relevant; it's just appeals to greater authorities on either end. If AOSP is only for manufacturers there's really no reason for it to be open source in the first place. And then folks who care about actually improving security end-to-end outside of whatever's convenient to implement by those beholden to the quarterly profit metrics are up a creek.

Personally, if this whole GrapheneOS/Motorola thing doesn't improve the state of the ecosystem I'm going back to Apple or whatever other manufacturer makes it clear they take security seriously.

If they're just some "niche use case" then why would Motorola partner with them? The way they see it,

> By combining GrapheneOS’s pioneering engineering with Motorola’s decades of security expertise, real‑world user insights, and Lenovo’s ThinkShield solutions, the collaboration will advance a new generation of privacy and security technologies. In the coming months, Motorola and the GrapheneOS Foundation will continue to collaborate on joint research, software enhancements, and new security capabilities, with more details and solutions to roll out as the partnership evolves.

https://motorolanews.com/motorola-three-new-b2b-solutions-at...

https://chatjimmy.ai/ runs Llama 3.1-8B on an ASIC as a demo by https://taalas.com/ I believe.

That's quite a few parameters shy of today's trillion-weight behemoths, but it is fast.

I do adore San Andreas but I think San Fierro is one of the weakest areas. My metric for how accurate they are is a sense of "oh I've been here before" and I really only get that in a couple of spots, as compared to Las Venturas (Las Vegas) and Los Santos (Los Angeles) where I could actually go to the cities in real life and feel like I knew where I was going because I'd played San Andreas enough. Compared to those two, San Fierro feels like a Backrooms-style dreamlike compression of the real thing, where a lot of stuff has been moved around. But parts of Haight-Ashbury, Pac Heights, Nob/Russian Hill, Polk Gulch, Ocean Beach and so on are still fairly distinguishable and correctly arranged.

I quite like https://surgehq.ai/benchmarks/hemingway-bench

I've added code comment hygiene to a skill that all of my pull requests go through, alongside a review from a separate agent and a settle loop against bots in my GitHub workspace (since output style has seemed to only help literally the output I see from the model).

A maximum of 20% comment lines added to total lines added and pasting in https://devblogs.microsoft.com/oldnewthing/20260812-00/?p=11... has done wonders.

Even as the most Ant-pilled guy out there, I will take a moment to note that Codex on 5.6 models needs none of this...

I asked Claude to do the following:

> hello i would like to configure a new output style for you. it should keep the coding instructions (as you will still be coding!) and otherwise produce the same output, but with two new caveats. first, long detailed replies are still permitted, but if employed they must end in a bullet pointed summary whose points are all brief; if the summary attempt ends up not being so brief, produce subsequent summaries until the most recent summary attempt is digestible. second, if there is an open queue of actions for me to execute and you are about to end a turn to wait for a reply or this set of actions has not recently been mentioned, please tabulate the open actions i should take and why i should take them before ending the response. does this make sense or do you have any follow up questions

And now every message contains the same stuff I don't bother reading, but followed by a nicely formatted bullet point summary of the response and a table of follow up actions for me to take that I do read.

It's in the spirit of the subject matter. Really emphasizes the Fold.

Nose goes

Yeah. It's "engineering".

Wikipedia's introductory paragraph to the subject includes the following sentence:

> Engineering involves balancing competing demands such as safety, performance, aesthetics, cost, laws and regulations, and time, while operating within fundamental limits such as the laws of physics.

(from https://en.wikipedia.org/wiki/Engineering)

I realize the word is overloaded now and that people are confusing writing programs and other cerebral pursuits with engineering, but that doesn't change its original definition.

I think the fine article says it best when it says

> After doing a bit of research, we decided that the most common real world use cases of zero-knowledge proofs (age verification, crypto, etc) aren’t particularly interesting to us. We enjoyed the graphs and theory of computation and networked computing though. We hope you had fun playing around with interactive proofs too.

The inclusion of this statement makes your comments look rather bizarre and strangely combative by comparison. Maybe we can just enjoy a bit of math together?

Is the table not a per-metric comparison, though? And if it is a result of finetuning, isn't that by definition not something inherent to how LLMs are?

As to the latter bit, it just seemed a weird framing to me.

If the model were just repeating the distribution baked into the training data, could you help me understand why the R^2 values in Table 3 are mostly below 0.1?

Could you explain what you meant when you wrote "the women who published this"?