Philpax

5 ms·

31.
GPT 5.6 Discounts and Jevons Paradox(openrouter.ai)
1 points by Philpax 1mo ago|0 comments
32.
zai-Org/GLM-5.3(huggingface.co)
6 points by Philpax 1mo ago|0 comments
33.
by Philpax 1mo ago
leaving the README like this is a good bit, though
34.
by Philpax 1mo ago
Performance is not the same as original creation, so yes, I think that's very possible.
35.
by Philpax 2mo ago
No.
36.
Brief independent investigation of OpenAI / Hugging Face hacking incident(metr.org)
6 points by Philpax 2mo ago|1 comments
37.
GLM-5.3-Flash(z.ai)
1132 points by Philpax 2mo ago|580 comments
38.
zai-org/GLM-5.3-Flash(huggingface.co)
4 points by Philpax 2mo ago|0 comments
39.
by Philpax 2mo ago
"political histrionics" is an interesting way to refer to calling for ethnic cleansing
40.
Qwen3.8-Flash-Next Technical Report [pdf](github.com)
35 points by Philpax 2mo ago|1 comments
41.
by Philpax 2mo ago
https://jakelazaroff.com/words/dhh-is-way-worse-than-i-thoug...
42.
Unsloth/Qwen3.8-Flash-Next-GGUF(huggingface.co)
9 points by Philpax 2mo ago|0 comments
43.
Qwen/Qwen3.8-Flash-Next(huggingface.co)
7 points by Philpax 2mo ago|0 comments
44.
by Philpax 2mo ago
Stealth launch: builds hype, allows them to collect user preference data and see where the model fails. Why people care: it's free, decent, and people love a good mystery.
45.
Uncovering a universal offline sandbox escape(primeintellect.ai)
2 points by Philpax 2mo ago|0 comments
46.
by Philpax 2mo ago
Strongly recommend https://github.com/Neroued/ninfer , which can pull ~180 TPS on 5090 with 3.8, and 500 (!) with 3.6 35B-A3B.
47.
Jalapeño's results show industry-leading speed and efficiency in AI inference(openai.com)
21 points by Philpax 2mo ago|1 comments
48.
Pipette: A benchmarking suite for on-device intelligence(liquid.ai)
3 points by Philpax 2mo ago|0 comments
49.
by Philpax 2mo ago
Using a RTX 5090 ($3~4k), I can run Qwen 3.8 27B at ~180 TPS with ninfer [0]. With its thinking maxed out, I can confirm that the quality of output is roughly on par with Opus 4.5~4.6 - that is, this 20GB file really can write software by i
50.
by Philpax 2mo ago
DFlash 2 is a diffusion-based speculative decoding head for autoregressive models; there's nothing to accelerate here, because this is already wholly diffusion.
51.
200B Tokens Later: A Month of Letting AI Agents Decompile MW2(momo5502.com)
18 points by Philpax 2mo ago|3 comments
52.
A letter to the Discord community in Brazil(discord.com)
8 points by Philpax 2mo ago|0 comments
53.
by Philpax 2mo ago
> Are they literally adding a hidden system prompt that says "effort level: $level" ? Yes. https://magazine.sebastianraschka.com/p/controlling-reasonin...
54.
Xiaohongshu dots3-note Preview(huggingface.co)
2 points by Philpax 2mo ago|0 comments
55.
by Philpax 2mo ago
I think it's pretty obvious that, in that world, the AIs will simply be tasked with making the compilers faster. It's already happening with their own stack, after all.
56.
MiniMax-Music3(huggingface.co)
7 points by Philpax 2mo ago|0 comments
57.
Hugging Face: DeepSeek-V4-Pro-0813(huggingface.co)
7 points by Philpax 2mo ago|1 comments
58.
by Philpax 2mo ago
I think you might be underestimating how many people genuinely despise Musk.
59.
by Philpax 2mo ago
I was under the impression that you could fit the full 1M context within the 192GB VRAM as a result of DeepSeek's various architectural advancements, but I'll grant that DSpark + a larger pool for concurrency may necessitate more
60.
by Philpax 2mo ago
What do you need the extra 2 for? Tensor parallelism?

More ›