Muskwalker

For a creator posting a video, Google gives specific guidelines on whether to tag it as AI:

"Was AI used to generate or edit your content in any of the following ways?

- Makes a real person appear to say or do something they didn’t say or do

- Alters footage of a real event or place

- Generates a realistic-looking scene that didn’t actually occur

To follow YouTube’s policy, you’re required to tell us about your use of AI. This includes realistic sounds or visuals made or edited with AI. [View examples](https://support.google.com/youtube/answer/14328491)"

So oddly an intro graphic only has to be tagged as AI if it shows a real/realistic person, place, or event, but (per the examples page) apparently any AI-generated music counts.

> Since at least macOS 15 the context menu key on typical keyboards is recognized and functions

Apple's adding it to its own keyboards as well, at least on the third-generation Magic Keyboard with number pad (between F13 and forward delete).

A conventional way to navigate GUIs is by using the Tab key to jump from control to control.

A common way this can fail is if the Tab key traverses through the UI elements in an unexpected order.

Yeah, Shieldfont's Github shows that it is being made with Claude as well, so it'll be a non-starter for those who are dogmatically anti-AI.

Their stance appears to be "[t]his is not an anti-AI font. AI is in our lives. We are actually a pro-consent font." (https://github.com/isaqueseneda/shieldfont/issues/2) Might be a bit more niche market.

> So has anyone managed to separate the effects of semaglutide from the effects of weight loss?

I was looking for that in particular. The study does say:

"To assess whether the observed effect was mediated by weight reduction, we performed a sensitivity analysis adjusting for change in BMI from baseline to week 104. In this adjusted model, the treatment effect coefficient was attenuated from β −0.092 to β −0.066 (P < 0.001), corresponding to a reduction of 28% in the estimated effect size. This attenuation indicates that 72% of the treatment-associated difference in 20-year dementia risk persisted after accounting for BMI change, suggesting that mechanisms beyond weight loss contribute to the observed proteomic signature modification."

This particular study was done on people without diabetes, at least.

100%, the author believes they learned the lesson, but failed to recognize what it looks like when a person on the other side initiates this process. It's amazing the dissonance here:

> They should expect a possible answer to be 'No' and be prepared to negotiate the difference. I didn't throw these wise words back at him, but that's what was going through my head. And it emboldened me further. I simply told him the room is unusable for yoga with the AC on and we've paid rent for the room.

They were emboldened to discard even mentioning negotiation and stood their ground instead. That might be fine if the lesson you're trying to illustrate is one about firm boundaries but the advice they're holding up here was very much the opposite.

The author there reads the point "Two different data entry operators, given a person’s name, will by necessity enter bitwise equivalent strings on any single system, if the system is well-designed" by thinking in terms of erroneous spelling, but also relevant is encoding issues such as lookalike characters.

Peter Biľak gives a story on bumping into this for the accent in his name. https://www.typotheque.com/articles/lcaron

Officially, Fable/Mythos testing was delayed because of Anthropic's data retention policy. Don't know if there's word on them working that out yet.

https://x.com/arcprize/status/2064399134099153344

True, though I think the point they're aiming at is that symbols like [new driver mark] contrast with literal pictograms like [symbol of person in wheelchair]. You can infer/guess the meaning of [symbol of person in wheelchair] 'without words' in a way that you can't with [new driver mark], because [symbol of person in wheelchair] is communicating with a picture instead, while [new driver mark] appears to be purely convention. (At least, the article doesn't seem to suggest otherwise.)

> complacent ("smug satisfaction with oneself" - I disagree that complacency is intrinsically smug)

I agree that it doesn't seem 'smug', but weirdly both dictionary dot com and Wiktionary give 'smug' as a synonym or part of the definition.

But they also analyze 'smug' as equivalent to self-satisfied or self-complacent, so maybe that's the word whose meaning is not as expected.

(I would think of "smug" less as "self-" anything - it implies a relation, it's more like exulting in a superior situation one has over someone. And 'complacent' is at base being content with one's situation, but often with the negative implication that one should be acting to make things better instead)

I don't normally experience the time interval between when my input streams shut down every night and reboot every morning either, though.

Nothing prevents one from running an LLM whose harness has a clock and a while loop in it, and it would be weird if its mere lack were really so consequential to consciousness.

The Thaler case here is something different than "AI-generated = uncopyrightable" though. Thaler was not trying to copyright work in the way humans who make work with tools normally copyright their work ("Copyright 2026 by Me"), he was specifically trying to give AI the copyright ("Copyright 2026 by My-AI-Tool"). The court rejected this because only humans can own copyright.

I believe there are other cases where AI-generated works were found uncopyrightable but Thaler is not a good example* of them.

If it's not your thing, it's not your thing, but if 'lying' is really the only barrier, note that a lot of churches actually consider it part of their mission to work with nonbelievers and would take something like "I'm not a believer but I'd like to learn what it's like for you all" (or some other true formulation of your intentions) as a valid form of interest.

As someone who's done just that... if it helps, understand that this is the kind of person who would be spending hours writing that paragraph anyway, LLM or no.

There are many possible reasons for this, and sometimes people are laboring under several of them at once.

So, could this be an example of an LLM trained fully on public domain copyright-expired data? Or is this not intended to be the case.

In fact in this case, it's not the known limitation of floating point numbers to blame: this Calculator application gives you the ability (submenu under View > Decimal Places) to choose a precision between 0 to 15 decimal places, and it will do rounding beyond that point. I think the default is 8.

The original screenshot shows a number with 13 decimal places, and if you set it at or above 13, then the calculation will come out correct.

The application doesn't really go out of its way to communicate this to the user. For the most part maybe it doesn't matter, but "user entering more decimal places than they'll get back" might be one thing an application might usefully highlight.

> I'm on Mastodon. It's only very mildly nice. The reality is that it still suffers from all of this.

Most of it, at least - it does give the option to not be shown forwarded messages. There's controls on the home timeline to not show boosts, quotes, and/or replies.

Unfortunately English "fun" is used both for good wholesome fun and for the cruel fun that is "making fun of" people (laughing at their misfortune).

The site also says (at the bottom) that it grabbed the translations from Google Translate back in 2014 and hasn't updated them since.

Yeah, the bit that gets me is the assumption that an artist just _yields_ to the AI's choices.

The author says in the article "my own hand, the single most valuable asset I possess", but I'd say that much greater is the artist's eye——the artist doesn't just accept every line that flows out of their pencil, but knows what to throw out, what to redraw.

The infamous stereotype of AI art, the hand with too many fingers, is the outcome of a lot of people being given an artificially skilled hand without the skilled eye to go with it...

The AI artist who doesn't just cede to the AI, though, is bringing their own hand and eye to the work, not accepting the mere output of a prompt, but actively remaking it as an expression of their own craft and vision.

Lots of black-and-white pictures, lots of pictures in silhouette.

Pictures of the tower can also have a lot of... texture? If you don't know you're looking at "brown with shadows" you might think you're looking at "black with highlights."

Internet Archive suggests the original discussion was a Usenet thread; here is the link to Google Groups' version of it:

https://groups.google.com/d/topic/geometry.research/7pyFhAAy...