bel8

10 ms·

1. 241.▲
   by bel8 5mo ago

   Well if they have a big challenge ahead since DeepSeek offers an open model at Sonnet+ level while being cheaper than Haiku, plus 1 million context size.
2. 242.▲
   by bel8 5mo ago

   On this note, is there a benchmark aggregator to compile all benchmarks in a single large grid?
3. 243.▲
   by bel8 5mo ago

   One difference is that MiMo 2.5 (non-Pro) has image, audio and video input capabilities. DeepSeek does not understand image, audio or video.
4. 244.▲
   by bel8 5mo ago

   It's funny, in a good way, because their off-peak times match perfectly the werstern peak demand.
5. 245.▲
   by bel8 5mo ago

   DeepSeek Flash on high (not max) is a freak of nature indeed. Very disproportionate intelligence-to-cost ratio. I'm leveraging this temporary anomaly and using it as my coding workhorse.
6. 246.▲
   by bel8 5mo ago

   I'd keep the conversation to dogfooding and how it helped Microsoft achieve unique status, but you do you.
7. 247.▲
   by bel8 5mo ago

   Sure success is not guaranteed. But dogfooding helps. Should Microsoft have stopped dogfooding because of Zune? They are worth multiples of trillion dollars so I would wager they know a little more about success than you and me.
8. 248.▲
   by bel8 5mo ago

   I consulted for multiple enterprise C# projects in the last 5 years. At least two of them are 1mil+ lines of code each. All of them run in Linux servers. Some of them were ported from PHP and Python to C#. Plus LLMs thrive in strongly typed
9. 249.▲
   by bel8 5mo ago

   1) They can still use Anthropic models. 2) Opus is not even unambiguously best at coding anymore. GPT 5.5 splits that title for some time now. 3) I would have probably done the same in his position. Dogfood the product.
10. 250.▲
    by bel8 5mo ago

    JS can't really break the web. It's a trillion+ market. If that ever happens for some exquisite reason, it would be forked in under 2 milliseconds by the browser consortium.
11. 251.▲
    by bel8 5mo ago

    try high variant instead of max. max is really chatty for minimal gain.
12. 252.▲
    by bel8 5mo ago

    opencode allegedly has contractual no-data-retention policies with their providers. I recall reading about that in an issue or in their Discord server. But I would contact them formally to verify that.
13. 253.▲
    by bel8 5mo ago

    Great! I have been using DeepSeek 4 Flash high for everything lately. First accessible model with useable 1 million context window for me.
14. 254.▲
    by bel8 5mo ago

    It seems this confusing tooling change plus the new smaller quotas made enough users unsubscribe for them to hit the panic button: https://www.reddit.com/r/google_antigravity/comments/1tjbd1e... https:/
15. 255.▲
    by bel8 5mo ago

    Good question. I haven't tested openrouter but I expect it to be slightly less cheap because it charges per token and opencode Go plan is a $10/mo fixed price model. Economies of scale leads me to think that for heavy use, openrou
16. 256.▲
    by bel8 5mo ago

    I expect bun to run almost everything that node runs these days. They have an extensive test suit to ensure that. Even the complicated NextJS runs with Bun: https://nextjs.org/conf/session/nextjs-bun Do you have a
17. 257.▲
    by bel8 5mo ago

    Can't the previous body of work be from AI too?
18. 258.▲
    by bel8 5mo ago

    I use their VSCode extension: https://marketplace.visualstudio.com/items?itemName=sst-dev.... It adds a button to VSCode to open a tab with opencode loaded. It's a bit better than just opening the CLI because it has so
19. 259.▲
    by bel8 5mo ago

    same boat. Google Pro AI quota became barely useful for anything meaningful. I think they envision Pro plan as "just a taste of AI, enough to lure folks into the Ultra plan" but that won't work for me when Codex is half the p
20. 260.▲
    by bel8 5mo ago

    Hi! I'm using DeepSeek V4 Flash on high via opencode. It should work with opencode using the proxy server or middleware method right? Any tips? Does this need a GPU to work? Or is it CPU only? I ask because I plan to try to run this us
21. 261.▲
    by bel8 5mo ago

    My anecdote: smart but too stubborn to be useful. I have been trying Gemini since 2.5 for coding. It is the smartest for creative web stuff like HTML/CSS/JS. But it has been very stubborn with following instructions like AGENTS.md
22. 262.▲
    by bel8 5mo ago

    Generate 5x more value for the same amount of money.
23. 263.▲
    by bel8 5mo ago

    Regardless, I see lots of people using local AI as an excuse to buy over provisioned macs and AMD Ryzen machines. Then not using it at all for local AI. We all like our expensive toys.
24. 264.▲
    by bel8 5mo ago

    To put it in another way, Apple could fund Asahi Linux.
25. 265.▲
    by bel8 5mo ago

    AI will have to be very reliable, fast and free to replace md files with HTML. Because I often dump text into md files and the operation is instant. Same for small tweaks.
26. 266.▲
    by bel8 5mo ago

    I wish that was Apple hiring, to officially support Linux on macOS day 1.
27. 267.▲
    by bel8 5mo ago

    And they run slower and quantized.
28. 268.▲
    by bel8 5mo ago

    Ahh I see. Thanks for the clarification.
29. 269.▲
    by bel8 5mo ago

    Great article but I'm confused on one thing. The article claims steering only works in local models, but GitHub Copilot has a "steer with message" feature where I can course correct mid execution. I use it often. I think thes
30. 270.▲
    by bel8 5mo ago

    React maybe (because there are alternatives). But for mid/large projects, I find that TypeScript brings sanity to JavaScript. I love some quick and dirty JS project. But after a certain project size I begin the see runtime errors like