pixelesque

Those still have to be rasterised to a pixel-based frame buffer at render/compositing time.

Yeah, even in like 1996/1997 for certain industries there were hints as to the way things were going, even if it took 4/5 years for the transition to fully take place.

For example, in 1996/1997, Digital Domain (VFX industry) used a 'render farm' cluster of Carrera Alpha workstations running NT to render the Titanic film, instead of SGIs running IRIX. (SGIs were still often used on the artists workstations though, but progressively that shifted).

By 2001, many of those machines were x86 which were then often as fast as the SGIs and Alphas, even with x86's stack-based floating point architecture which handicapped it a bit, and the significantly higher memory bandwidth and larger caches of the SGIs.

Well, to some extent you might be able to argue they're "just" brute-forcing things (especially with unlimited tokens and hours to spend on a task), but they obviously have detailed knowledge to guide them in their attempts, can learn (or at least, persist their newly-gained knowledge), and can use tools.

With a swarm of them working together at speeds humans would be unlikely to match (in terms of iterating on different attempts progressively), it's a lot easier to see how they could overwhelm targets.

It was the same in June I think...

They promised weights for Qwen Image 2 around 6 months ago and there's still nothing, so wouldn't get your hopes up...

The really annoying thing to me about the "Ads" in the system settings for various things like Apple Music, iCloud, Fitness, etc (other than the fact they're there in the first place) is that they appear after a bit of a delay (around 500 ms?), so other items I actually want have to slide down to make space as I'm about to click something.

Old Apple would never have done that I don't think.

> We deal mostly with source code, so the text we fold is overwhelmingly ASCII and making it run at memory speed is the single most important thing we can do. Everything else just has to keep the rare non-ASCII path from spoiling it.

Semi-on-topic: I've noticed that many LLMs via coding agents (ChatGPT and Claude at work with my CoPilot account, and DeepSeek 4 and ChatGPT in pi.dev at home) really seem to like using unicode / emoji characters for things like arrows (for things like test value ranges), crosses and ticks (for pass vs fail in test comments), instead of plain ASCII. Codebases are almost exclusively ASCII chars to my knowledge, although they're UTF-8 files.

I'm not yet using agents to write code (only do code reviews, write example prototypes I then copy bits of, and helping craft tests), but I'm likely to get there soon, and I'm sure it's possible to prompt them NOT to do this, but has anyone else noticed this? I wonder if that changes things over time for them if this is a common theme of increased non-ASCII output?

I mean, it's a spectrum, surely...

What about stock photography, IMO as much as I personally hate "overcooked" AI-generated imagery, it doesn't annoy me that much more than stock photography does on blogs.

I do however use it as a gauge of "if the images are generated by AI, how much of the text content is?" though, so...

I mean, the VFX part of the Film industry (and actual location shooting to some degree) has been chasing subsidies / tax credits since around 2002 with the Harry Potter movies that mostly were shot and produced (and more specifically the VFX were done in London, rather than in California) in the UK.

Since then Canada and the UK in particular (and NZ and Australia) have had various different fluctuations in subsidies for various parts of filming / doing post-production of movies over the intervening years.

Thanks - yeah, sorry, I mis-read that as you using both DS and K3...

What quant are you using, and what tps are you getting with K3?

Hit detection doesn't seem to be always getting the nearest surface to the camera: if there's a folded-over corner in front of the main surface, it almost always seems to move the 'main' part of the 'fabric' behind the folded corner, rather than allowing dragging just the corner in front.

Out of interest, are you using the DeepSeek plan?

(I've been using it via OpenRouter and it's much more than that, but still cheap).

Cheaper - not cheap!

Morph occasionally have lower prices than the standard rates, and:

https://telnyx.com/pricing/inference-api

Is one which is a bit cheaper... I haven't actually tried K3 myself...

I wish!!

No, I'm using it via OpenRouter in pi.dev - I just used it 30 mins ago... Providers (automatically selected): StreamLake and Baidu Qianfan.

It's cheaper currently on many of the inference providers.

Personally, I'm having surprisingly good results with DeepSeek 4 Pro at home, which is very good value for money: it's not as good as Claude / GPT 5.6 (I have Co-pilot license at work), but it's still really useful for code reviews, validating thoughts, and especially designing / writing unit tests for new (and old before refactoring) functionality.

And it's very cheap per task. (Flash is even cheaper, but I've had issues with that on more complex tasks where it starts forgetting things and arguing with itself "but wait, let me read the function again").

Possibly because it's easier to approximate the ball's position moving vs the table than simulating the table's position in space moving (which is what happens in real life with a nudge?) and the ball's trajectory not altering?

I assume you mean Q8_0, not QB_0? (was confused, as I've not obviously seen a QB_0 before and can't find one now).

Even for row-based data?

Why has one of the images in the header carrousel got it featuring what looks like a refueling probe for refueling other aircraft sticking out the back? The other two don't.

It looks like they've taken this mostly from an older Boeing BWB design for future air-to-air refueling aircraft (although the wing shape is a bit different).

16 GB of RAM for a car infotainment system!!

(Maybe it's one of those that supports split outputs for multiple screens for all passenger positions? - i.e. kids in the back? That might be a bit more understandable.)

I'm curious about this too, and it's difficult to get any information about this given everyone has different setups, workflows and use-cases.

I bizarrely had Opus 4.8 this week (in pi.dev within a podman container, using openrouter) start installing various python packages (and uv!) within the environment (not as root) when I asked it to code review some fairly basic Rust .rs files that were generally stand-alone (it did very nicely work out and write some stubs for them to build them and work out how they worked).

It only gave up with the weird Python installing stuff when it discovered one of the Python packages needed Tensorflow.

It seems pretty focused and persistent in continuing its initial approach, and I'm wondering if I need to alter some instructions / initial prompts to rein it in a bit...

Only if you host public FOSS projects - it's not a replacement for private repos.

It seems a lot of people on here don't understand that.

Claude Bot still (at least last month, and it's been doing it for over a year now) seems to have a bug when traversing (at least my sites), wherein it drops the trailing slash of a directory (which is present in the a href tag), then makes the request to the subdirectory without the slash I put in the link, then Caddy automatically responds to that (via the built-in File handler) with a redirect telling it to add the trailing slash, and Claude Bot then makes the request again with the trailing slash that should have been there in the first place.

So most subdirectory URLs get two requests from Claude bot, the first one needless because that wasn't the URL in the tag.

Semi-off-topic, but...

Is the release of this why Google's share price is down 4.5%?

Thanks!

Out of interest, is that all that's required for the power negotiation handshake? I would have expected more complexity?

Note: in some cases (not completely sure if it applies in this situation as I haven't looked into the details of this code enough), using pre-increment AND when the value is used is a dependency on that operation, so "+= 1" is actually faster for out of order execution.

So yeah, as you say, compilers can be very sensitive to this, and in the past (although more than 10 years ago now, so might not be relevant now), ICC often used to generate better (faster executing) code when using "val += 1" vs "++val".

Semi-off-topic but...

On my sites, I see ClaudeBot consistently (and has been like this for over a year) ask for "${SITE}.com/base_dir1" and then get the redirect (Caddy does this automatically) to get "${SITE}.com/base1/base_dir1/" (trailing slash).

The hrefs on my sites include the trailing slashes for directories, so looks like ClaudeBot's internal code is stripping them off before requesting them, and therefore essentially makes almost 2x the requests to my sites for directories, half of them ending up being redirects back to the same url but with the trailing slash.

If would likely need to track them well (not sure from this article/video if that's the case?) to be useful in that scenario...

Drawing a splodge in roughly the location (not sure if there's range info either? I doubt it if it's passive) overlaid on the video likely won't cut it...