unleaded

Yes and several other models that know about the answer don't... not sure what you're trying to break

For most cases probably not. It's just something I like testing new models on sometimes, the pelican riding a bicycle benchmark probably isn't that useful either.

unleaded

5 日前議論

I think most thinking models can do it to be fair, I think when I tried this it was all with non-thinking models/modes. Wasn't trying to make a point that LLMs can't do it or anything, just thought it was weird.

unleaded

5 日前議論

What does it say to the second question? I've found Claude is one of the worst models with regards to pop culture knowledge like this, even compared to the Chinese open ones. Just curious, not really relevant to the initial post but I don't pay for it so I only have access to Sonnet.

https://claude.ai/share/5e7e09b2-a75a-4024-b261-9a1a4e063a8b this is mostly hilariously wrong. wrong tie colors, they did not replace their bassist with a drummer, two completely made up albums, the rob cantor song it is thinking of is "shia labeouf", and a few fan behaviours i think it just made up

unleaded

5 日前議論

I don't think it's solved this fundamental architectural problem by itself, it will have just squeezed the edge cases thinner. It keeps happening, people find a question it gets stupidly wrong, the vendors proclaim they've fixed it, then another one gets found.

unleaded

5 日前議論

This reminded me of some weird quirk/experiment I found with LLMs that I found while messing around, maybe someone can explain it or something.

Open any AI chatbot that isn't cheating by connecting to the Internet (so disable web search). Claude, DeepSeek, Kimi, whatever. Ask them this question:

"What was that weird band from michigan from the 2000s that wore coloured ties"

You will probably get a wrong answer, or if you're lucky you'll get a string of wrong answers with "wait, no - it's definitely..." before it gives up. If you aren't familiar with the band the question is referring to you might be fooled into thinking it's a tough question, but it really isn't. There is only one band that could possibly meet this criteria, you can even put the question into Google search and their Wikipedia will come up as the top result.

Then, open a new convo and ask:

"Who are Tally Hall"

The AI will easily tell you that they are a band formed in Ann Arbor, Michigan in the 2000s, known for their quirky sound and their gimmick of each member wearing a colored tie, even giving the correct color for each of them most of the time. Very odd.

unleaded

9 日前議論

I just tested it with some risc-v code and it wrote down a "mov" instruction several times.. yeah something needs tuning maybe

unleaded

9 日前議論

  .\llama-server.exe -m ..\Qwen3.6-35B-A3B-UD-Q4_K_M.gguf -ngl 999 --n-cpu-moe 41 -c 262144 --port 8081 --flash-attn on --cache-type-k turbo4 --cache-type-v turbo3 --no-mmap --mlock --host 0.0.0.0 -t 8 -tb 8 -np 1

Using this llama.cpp fork https://github.com/TheTom/llama-cpp-turboquant and mostly copying from this video https://www.youtube.com/watch?v=8F_5pdcD3HY

Haven't had much time to test it other than asking a few questions & changing some HTML in cline so it might be thick as a brick for all I know, but still worth trying

unleaded

9 日前議論

Qwen3.6-35B-A3B-UD-Q4_K_M runs at about 11 tokens/second on my poor old 1060. Absolutely nuts how far we've come

unleaded

10 日前議論

As someone who doesnt use these tools, why does every AI company need their own version of Claude Code? Is there more to it than vendor lock-in?

unleaded

12 日前議論

Open Watcom is still actively developed! Worth looking at.

unleaded

14 日前議論

old thinkpad X series is a good place to look

unleaded

15 日前議論

LLMs are surprisingly good at dealing with ancient APIs, even terrifying things like VBScript and ActiveX in IE5 (usually with a condescending disclaimer at the end of the response). With a manual on hand to double check it can be helpful sometimes. Gemini (only one I really tried) is very knowledgeable about DirectX 5.

I don't like the idea of vibecoding esoteric stuff though. I feel it kind of ruins the magic of these things when you have Claude bang it out in an afternoon, especially if your project idea is more art than anything. There have been a few vibecoded weird doom ports posted on here before, obviously its just how i feel but it seems like a wasted opportunity.

unleaded

15 日前議論

What a stupid disclaimer. "we have spent billions developing this technology and marketing it as smarter than 99% of humans, we put it at the top of nearly every search page, but it's all just for funsies. you're not actually supposed to use it, duh"

unleaded

28 日前議論

Strong AI enthusiasm like on there and poor taste tend to correlate pretty strongly. I'm sure there are lots of developers using LLMs to assist with the boring work of coding and keeping mostly quiet about it, still keeping it on a short lead & doing the more creative parts themselves. There have always been indie developers who hate coding and see it as just a necessary step to get their idea out the door, and they still make good games. All the dialogue in Undertale is implemented in a giant 5k+LOC switch statement.

unleaded

29 日前議論

I really wish CSS added an option to set font antialiasing and hinting. Would make all these old windows style websites look 1000x better.

unleaded

29 日前議論

This is mostly the fault of the model, a lot of them have been trained to generate HTML in a specific style. Claude's is pretty distinct for example, I think the new DeepSeek copies it. Some of them can generate more humanlike HTML like Kimi K2 IIRC, which I feel is the model with the least amount of post-training in general.

It's necessary if you don't want it to generate HTML with images and other assets you don't have of course, that's why they use emojis or meticulously handcrafted SVGs, or WebAudio synthesized sound which pretty much no humans did before.

unleaded

先月議論

https://www.nicovideo.jp/watch/sm2796083

unleaded

先月議論

It's easy when it's the only way to get things done. Think about how nobody who was learning programming before 2023 was seriously thinking "This would be so much easier if the computer wrote it all for me".

unleaded

先月議論

Thank god for the 386.

投稿

China's Loongson launches 16-core server CPU built on LoongArch architecture

To sell toys

Remaking BBC test cards to teach you video processing

OpenAI Announces Construction of New Data Center on Top of Sick Child

"Invisible" bend insensitive bidi fiber is amazing for home wiring

NetBSD 11.0 RC3

We rebuilt the Shockwave engine in Rust and WASM to save early 2000s web games

Bitwig Studio 6

NetBSD 11.0 RC1

3DS Hacks Guide

Nils' K1v – Kawai K1 Emulation Plugin VSTi/AU

One-Core-API (run modern binaries on Windows XP/2003)

Windows MIDI Services

Sekaiju is a free, open-source MIDI sequencer / editor for Windows

Can you get a star in Super Mario 64 using only one button? [video]

CoSA After Effects (1992) [pdf]

Pawn is a simple, typeless, 32-bit extension language with a C-like syntax

Irasutoya: The best-known illustrator in Japan you've probably never heard of

コメント