GaggiX

16 ms·

1. 241.▲
   by GaggiX 1y ago

   You can also use a "dumb" watch that has 10 years battery life.
2. 242.▲
   by GaggiX 1y ago

   Huge batches to find the perfect balance between compute and memory banthwidth, quantized models, speculative decoding or similar techniques, MoE models, routing of requests on smaller models if required, batch processing to fill the GPUs w
3. 243.▲
   by GaggiX 1y ago

   GPT-5 models are actually great models for the API, the nano model is finally good enough to handle complex structured responses and it's even cheaper than GPT-4.1-nano.
4. 244.▲
   by GaggiX 1y ago

   Compared to GPT-4, it is on a completely different level given that it is a reasoning model so on that regard it does delivers and it's not just scaling, but for this I guess the revolution was o1 and GPT-5 is just a much more mature v
5. 245.▲
   by GaggiX 1y ago

   There is a new Qwen3-30B-A3B, you are compare it to the old one. https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507
6. 246.▲
   by GaggiX 1y ago

   Claude Opus is in the top 10, also people via OpenRouter mostly use these models for coding and Claude models are particularly good at this, the benchmark doesn't account only for coding capacities tho
7. 247.▲
   by GaggiX 1y ago

   https://huggingface.co/KittenML/kitten-tts-nano-0.1 https://github.com/KittenML/KittenTTS This is the model and Github page, this blog post looks very much AI generated.
8. 248.▲
   by GaggiX 1y ago

   This is another article based on the study that had a population of 16 people, most of which never used the tool before. Edit: "amirhirsch" user probably explained this better than me in an above comment.
9. 249.▲
   by GaggiX 1y ago

   Did it face the right way?
10. 250.▲
    by GaggiX 1y ago

    OpenAI is not the only company making LLMs, there are plenty now, you can use Gemini 2.5 Pro for example. And of course you can just try a SOTA model like Gemini 2.5 Pro for free, you don't have to trust anything.
11. 251.▲
    by GaggiX 1y ago

    And the old ChatGPT was nothing compared to what we have today, nowadays reasoning models will eat through math problems no problem when this was a major limitation in the past.
12. 252.▲
    by GaggiX 1y ago

    There is no difference, you can choose a 6 bits quantization if you prefer, at that point it's essentially lossless.
13. 253.▲
    by GaggiX 1y ago

    Reasoning models like qwen3 are even better, and they have more options, for example you can choose the 14B model (at the usual 4KM quantization) instead of the 8B model.
14. 254.▲
    by GaggiX 1y ago

    https://apxml.com/tools/vram-calculator This one is very good in my opinion.
15. 255.▲
    by GaggiX 1y ago

    I wouldn't be surprised if the author suffers from Bipolar Disorder or Schizophrenia reading the repo and his own citations, also a quite dumb LLM if used.
16. 256.▲
    by GaggiX 1y ago

    >Won’t the government force these websites to do some kind of stronger identity verification in the future? These heckin' kids needs more protection. I suggest banning all VPNs too, only this way kids are truly protected like they a
17. 257.▲
    by GaggiX 1y ago

    You can literally read the developer response: https://x.com/siarate/status/1949739112182714819 "And ofc Mouthwashing got yoinked from itch search results lol"
18. 258.▲
    by GaggiX 1y ago

    Even Mouthwashing got delisted by itch io, it's pretty insane how censorship is escalating in these last few days.
19. 259.▲
    GLM-4.5: Reasoning, Coding, and Agentic Abililties(z.ai)

    247 points by GaggiX 1y ago|134 comments
20. 260.▲
    by GaggiX 1y ago

    >but fail to block tech-savvy children. If I were a kid, I could see myself downloading Opera GX and enabling the free VPN. It's probably not "tech-savvy" because the browser gets a lot of ad views on YouTube; it would be
21. 261.▲
    by GaggiX 1y ago

    The input is wrong tho Your website reports 0.30$ for input and that wouldn't make any sense as it would be priced the same as the bigger Flash model.
22. 262.▲
    by GaggiX 1y ago

    >Squoosh lets you disable subsampling for JPEG and AVIF, but it can't be disabled for WebP, as the format doesn't support it. To work around this, the WebP encoder has a "sharp RGB to YUV" option which tries to limit
23. 263.▲
    by GaggiX 1y ago

    This article is based on the study that had a population of 16 people, most of which never used the tool before.
24. 264.▲
    by GaggiX 1y ago

    >We are pretty much plateauing in base model performance since gpt4. Reasoning models didn't even exist at the time, LLMs were struggling a lot with math at the time, now it's completely different with SOTA models, there have b
25. 265.▲
    by GaggiX 1y ago

    Whisper frequently generates random credits. I guess they didn't curate the dataset much at the time.
26. 266.▲
    A retro gaming YouTuber faces possible jail time for reviewing gaming handhelds(androidauthority.com)

    12 points by GaggiX 1y ago|0 comments
27. 267.▲
    by GaggiX 1y ago

    There is also a Voxtral Small 24B small model available to be downloaded: https://huggingface.co/mistralai/Voxtral-Small-24B-2507
28. 268.▲
    by GaggiX 1y ago

    >If someone took Kimi k2 and trained a reasoning model with it I imagine that's what they are going at MoonshotAI right now
29. 269.▲
    by GaggiX 1y ago

    Is this website supposed to load? Is HN the deadliest poison of this website?
30. 270.▲
    by GaggiX 1y ago

    So many nuances are lost in translation. I also can't imagine speaking English with actual people through a machine instead of speaking it directly.