bel8

10 ms·

1. 211.▲
   by bel8 4mo ago

   I wouldn't skip at least testing the original. Model distilling done by Cursor could be the culprit.
2. 212.▲
   by bel8 4mo ago

   I had a similar experience with DeepSeek Flash. I'm developing a webgl game in TypeScript using my little custom vibesloped game engine that runs in the browser and live reloads whenever a file is saved. I told the LLM to implement Mul
3. 213.▲
   by bel8 4mo ago

   I write prompt for another feature or fix.
4. 214.▲
   by bel8 4mo ago

   What works for me is: - Use a fast model like DeepSeek Flash V4 on high (it's Sonnet level, but fast and cheap). - While the LLM is working, start writting your next prompt. A good prompt usually takes between 1 and 10 minutes to write
5. 215.▲
   by bel8 4mo ago

   the OS is broken, in this case.
6. 216.▲
   by bel8 4mo ago

   What I like about MiMo too is that it is multimodal. For example, I can send screenshots of what I'm developing and it understands.
7. 217.▲
   by bel8 4mo ago

   From the notes, they seem humble and empathic. We're lucky to have China imposing competiton to the western AI megacorps. If it wasn't for China, I would probably have to spend $100/mo on AI instead of $10 like I do currently
8. 218.▲
   by bel8 4mo ago

   I don't think so. This is a VM, closer to WSL2. Proton is based on Wine which translates Windows instructions to Linux. Besides there's already Wine for mac. But I would love to be wrong here.
9. 219.▲
   by bel8 4mo ago

   For server side, which I believe is the context here, Linux and open source are king. Even Microsoft gave up on Windows and just runs Linux most things except niche cases. Heck, even SQL Server which is expensive piece of machinery got port
10. 220.▲
    by bel8 4mo ago

    Sure but we're talking about the unified law of almost an entire wealthy continent here. It's EU ffs. Not some small island country in the middle of the ocean. This looks to me like yet another bet from Apple: "they'll b
11. 221.▲
    by bel8 4mo ago

    It worked for OpenAI when GPT 3 was deemed too dangerous to be released. This is just a spin of that.
12. 222.▲
    by bel8 4mo ago

    With billions/trillions of dollars floating around, is it hard to imagine benchmarks could be biased? I think it's safe to assume everything AI related is heavily biased until proven otherwise. Just like in pharma.
13. 223.▲
    by bel8 4mo ago

    I sniff iPhone prompts ending up in Musk's xAI servers... Apple -> Google -> xAI datacenter see: Google to pay SpaceX $920 million a month for compute capacity at xAI data centers https://www.cnbc.com/2026/06
14. 224.▲
    by bel8 4mo ago

    Yep. Even many of our jobs became remote and less social.
15. 225.▲
    by bel8 4mo ago

    I'll bite. Maybe it's not the iPhone itself, but social media. I have seen studies about damages that social media can cause in behaviours. This might be one of them. Smartphones are the catalyst to social media consumption as we
16. 226.▲
    by bel8 4mo ago

    You might be interested in this: > With $3.88 & 690,003,591 tokens and 5 hours, Deepseek Pro & Flash combined, managed to reverse engineer Teamspeak's Licensing System for 3.13.8 (latest of post) https://www.reddi
17. 227.▲
    by bel8 4mo ago

    The power of nostalgia.
18. 228.▲
    by bel8 4mo ago

    Some corporate demand Windows as host (contractual/compliance bs) but I want to work in Linux.
19. 229.▲
    by bel8 4mo ago

    WSL supports GPU passthrough.
20. 230.▲
    by bel8 4mo ago

    Add MiMo 2.5 to the list. Priced like DeepSeek, performs similarly but it also has vision capability.
21. 231.▲
    by bel8 5mo ago

    > In your experience, is max worse or you suggest it for less token use? Yes. DS4 Flash max is incredibly chatty for minimal gain over DS4 high. I asked the same question a month ago: https://news.ycombinator.com/item?id=
22. 232.▲
    by bel8 5mo ago

    Unfortunately there's no defending Haiku 4.5 at this point when cheaper and better options are available. TLDR: https://artificialanalysis.ai/models?models=gemini-3-5-flash... and: https://i.imgur.com/n
23. 233.▲
    by bel8 5mo ago

    It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub C
24. 234.▲
    by bel8 5mo ago

    They know Google has a ton of data to train LLMs on. Recently I have been asking YouTube's new AI about some videos ("when is Steam metrics mentioned in the video?" for example), which means they also index videos. This is an
25. 235.▲
    by bel8 5mo ago

    The XML extension I use in VSCode is by Red Hat. Oh dear. Here we go again.
26. 236.▲
    by bel8 5mo ago

    I would buy a Neo to escape Windows but it's not like macOS is a pleasant experience either. My next mobile workstation will probably be an arm laptop with Linux for great battery life.
27. 237.▲
    by bel8 5mo ago

    > You are to be congratulated on the sheer looming height of your standards. The angels cry out to you from the heavens. Sheesh. - akkartik Snarky ad-hominem comments are forbidden in HN, FYI. https://news.ycombinator.com&#x2F
28. 238.▲
    by bel8 5mo ago

    > "Use as-is until e-waste" the Neo is not That's a very low bar to clear
29. 239.▲
    by bel8 5mo ago

    DeepSeek V4 Flash - High (answered in 3.4s, $0.00 cost) Thought: Let me think about this question. The days of the week are: Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday. Now I need to count how many have the
30. 240.▲
    by bel8 5mo ago

    If Opus 4.8 is just slightly better than 4.7 then it maybe ties with GPT 5.4, maybe. And it gets completely outclassed by GPT 5.5 for my workload. With Anthropic expensive pricing, there's no reason for me to switch from GPT+DeepSeek.