Mindless2112

I don't think it's acceptable, and I don't tip if I pay before I eat. Ironically, I also don't adjust the amount I tip at sit-down restaurants based on how the service was. I refuse to allow tipping to expand beyond where it was customary when I was growing up. There should be less tipping, not more. It should be outlawed entirely.

Unless the LLM has an SVG renderer in its toolbox, it's like asking a human to draw while blindfolded. It's amazing if they can do it well, but it's practically meaningless.

Also there is https://www.doesitplay.org/

> In this Gemini-reconstructed scene, ...

I'm generally pretty pro-AI, but I find this icky. Of course, I wouldn't have noticed except the whiteboard drawing seemed not quite right, so I'll probably be fooled in the future.

A fair point...

I thought the bug was a missing check for the COW flag, but looking at it again it seems it was missing both setting and checking the flag.

In CHERI, for example, pointers have permissions. The pointer to the COW memory would not have the "write" permission.

I could be misunderstanding the bug, of course.

How much pain must there be until people realize we actually do need memory safety?

The parameters in the original gpt-oss-20B model are "post-trained with MXFP4 quantization", so there just isn't much to gain by quantizing to Q8. If you look inside the Q8 model, most of the parameters are MXFP4 anyway.

Though, looking inside my "gpt-oss 20B MXFP4 MoE" model, it looks to also be quantized the same way as the Q8, so that was probably an overstatement on my part.

Still, the Q8 is 12.1 GB and the FP16 is 13.8 GB. Not the ~1:2 ratio you might expect.

Current lowest price for a new card on Newegg: $949.99 vs $3,699.99.

As of b8966, it is still not great.

  | model                 |      size |  params | backend | ngl |   test |            t/s |
  | --------------------- | --------: | ------: | ------- | --: | -----: | -------------: |
  | gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | SYCL    | 999 | pp2048 |  851.81 ± 6.50 |
  | gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | SYCL    | 999 |  tg128 |   42.05 ± 1.99 |
  | gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | Vulkan  | 999 | pp2048 | 2022.28 ± 4.82 |
  | gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | Vulkan  | 999 |  tg128 |  114.15 ± 0.23 |
  | qwen35 27B Q6_K       | 23.87 GiB | 26.90 B | SYCL    | 999 | pp2048 |  299.93 ± 0.40 |
  | qwen35 27B Q6_K       | 23.87 GiB | 26.90 B | SYCL    | 999 |  tg128 |   14.58 ± 0.06 |
  | qwen35 27B Q6_K       | 23.87 GiB | 26.90 B | Vulkan  | 999 | pp2048 |  581.99 ± 0.86 |
  | qwen35 27B Q6_K       | 23.87 GiB | 26.90 B | Vulkan  | 999 |  tg128 |   10.64 ± 0.12 |

Edit: I've no idea why one would use gpt-oss-20b at Q8, but the result is basically the same:

  | model                 |      size |  params | backend | ngl |   test |            t/s |
  | --------------------- | --------: | ------: | ------- | --: | -----: | -------------: |
  | gpt-oss 20B Q8_0      | 11.27 GiB | 20.91 B | SYCL    | 999 | pp2048 |  854.16 ± 6.06 |
  | gpt-oss 20B Q8_0      | 11.27 GiB | 20.91 B | SYCL    | 999 |  tg128 |   44.02 ± 0.05 |
  | gpt-oss 20B Q8_0      | 11.27 GiB | 20.91 B | Vulkan  | 999 | pp2048 | 2022.24 ± 6.97 |
  | gpt-oss 20B Q8_0      | 11.27 GiB | 20.91 B | Vulkan  | 999 |  tg128 |  114.02 ± 0.13 |

Hopefully, support for the B70 will continue to improve. In retrospect, I probably should have bought a R9700 instead...

Compared to a B70, a 5090 is 1x the memory with 3x the bandwidth at 4x the price. Yeah, the 5090 is better, but you're paying for it.

Never buy another ebook from Amazon, sure.

Never buy another Kindle? I keep mine in airplane mode all the time and sideload all the books/papers I want to read. It works practically just as well as when I bought it. Why wouldn't I buy another? If Amazon makes a Kindle with color at 300 PPI, I will.

Sure, proper EPUB support would be nice, but if I need that I can jailbreak and install Koreader.

If there's another device with comparable hardware/software/battery, I'd consider it. AFAIK, Kindle still has the best standby battery life.

The demonym for France is "French," so it's not wrong (even if it doesn't sound right.)

Funny they didn't include any CJK languages on their list.

Also, there are:

* https://github.com/tomsoftware/Lemmings.ts

* https://lldb.camanis.net/level/play/473/1/Just-dig

It supports linking two Android devices.

There was a time when we could say "our greenhouse gas emissions are nothing compared to regular biological processes," and yet here we are.

> “The outbound and cross-bound DDoS attacks can be just as disruptive as the inbound stuff,” Dobbin said. “We’re now in a situation where ISPs are routinely seeing terabit-per-second plus outbound attacks from their networks that can cause operational problems.”

ISPs are starting to feel the pain, so perhaps in the near future they will do something about it.

I, too, am jealous of China's high speed railroads. However, on the whole, China has overbuilt their infrastructure, and that may not look so smart in 40-50 years when the maintenance bills start coming due.

The Freestyle Pro is almost a good keyboard. The Esc and function keys are all offset to the left by one key compared to a standard layout, which drove me nuts. I have a Freestyle Edge RGB now, which I like much better. (Though I replaced the wrist rests with some from Goldtouch.)

It's two different teams inside Google. Some part of the Chrome team is trying to quash JPEG XL.

It's a nice logo, but...

> So why would we change it? As a non-Indigenous entity, we acknowledge that it is inappropriate for the Foundation to use Indigenous themes or language.

Tokenization is an inherent weakness of current LLM design, so it makes sense to compensate for it. Hopefully some day tokenization will no longer be necessary.

It seems like this could be easily solved in models that support tool calling by providing them with a tool that takes a token and returns the individual graphemes.

It doesn't seem valuable for the model to memorize the graphemes in each of its tokens.

When an optometrist chose to file a claim against my medical insurance (which did not pay) rather than my vision insurance (which would), the only way I was able to get them to fix it was by filing a complaint with the BBB. Multiple calls with customer service only resulted in "there is nothing we can do to change it", but somehow they figured out how to fix it when they got the BBB complaint letter.

How else can private citizens keep businesses honest? Complaining on the internet only works if you have lots of followers.

Interesting that traffic didn't return to completely normal levels after the incident.

I recently started using the "luci-app-https-dns-proxy" package on OpenWrt, which is preconfigured to use both Cloudflare and Google DNS, and since DoH was mostly unaffected, I didn't notice an outage. (Though if DoH had been affected, it presumably would have failed over to Google DNS anyway.)

The CPI's hedonic quality adjustment means that this graph doesn't mean what you think it means.

> However, detecting environmental NIR information in the natural conditions at night without NIR illumination still remains challenging, requiring further advancements in material science and optical design.

Sadly no night vision contact lenses yet.

or http://1.65793 or http://1.1.257