XCSme

21 ms·

1. 421.▲
   by XCSme 8mo ago

   Do you have instructions.md and docs.md files? Also, I noticed it skips instructions if I steer it with prompts while it is doing stuff instead of queueing my instructions.
2. 422.▲
   by XCSme 8mo ago

   For coding, I agree, Codex-5.3 is the best out there. But for the chat, I feel like ChatGPT got worse and worse.
3. 423.▲
   by XCSme 8mo ago

   This reminds me a bit of the Black Mirror episodes with the bees. Where the people whose names tweeted something were actually the targets... https://en.wikipedia.org/wiki/Hated_in_the_Nation
4. 424.▲
   by XCSme 8mo ago

   Why the downvote? LE: Searching for "LinkedIn" I do get results, but not for "Reddit", maybe they haven't updated the policy yet, or they are not required to disclose that? But in the email it says "We'll
5. 425.▲
   by XCSme 8mo ago

   I tried to find the exact mention of this in their privacy policy, but I find it hard to navigate with their OS ui and tweet-like privacy policy. I saw no mention of Reddit/LinkedIn when I did a CTRL+F search
6. 426.▲
   PostHog now shares hashed emails of new users with Reddit and LinkedIn(uxwizz.com)

   4 points by XCSme 8mo ago|2 comments
7. 427.▲
   by XCSme 8mo ago

   The problem is not the models, is the moat and budget. Google and X still have money and are profitable, all the other AI companies are losing billions per year. And customers will happily switch from one model to another in a heartbeat.
8. 428.▲
   by XCSme 8mo ago

   X? Don't they have the biggest budget and largest GPU farm? Also grok-4.1-fast was one of the top models for a long time, especially in real world usage.
9. 429.▲
   by XCSme 8mo ago

   Well, the public reception seems to be changing, after tweets like this: https://x.com/AnthropicAI/status/2025997928242811253 Also, I liked Anthropic because they were focused a lot on safety, but after the Pentag
10. 430.▲
    by XCSme 8mo ago

    Anthropic* lol, unintentional...
11. 431.▲
    by XCSme 8mo ago

    Same question for Atrophic. Personally I only see Google (Gemini), X (Grok) and the Chinese models having a chances to still be alive in 1-2 years.
12. 432.▲
    by XCSme 8mo ago

    I love it too, because it's easy to write and copy, compared to formatted text
13. 433.▲
    by XCSme 8mo ago

    Love the stack you've built, really impressive to see a fully EU-based setup in action. For anyone trying to keep analytics fully European and self-hosted, I built UXWizz. It's EU-based, runs on your own servers, and adds session
14. 434.▲
    by XCSme 8mo ago

    Nice stack so far. Coolify + Hetzner is a solid combo once you start consolidating services. Since you're already self-hosting Umami, you might also want to look at UXWizz. I built it, and it fits the same privacy-first, self-hosted mo
15. 435.▲
    by XCSme 8mo ago

    Oh, and one thing that I noticed, your clickable cards look the same as non-clickable ones, I kept ending up clicking on the ones that do nothing.
16. 436.▲
    by XCSme 8mo ago

    Cool experiment, I like the idea of going deep on a micro-niche instead of chasing broad keywords. Since you're comparing structured overviews vs blog-style content, I'd watch scroll depth and click behavior closely. I built UXWiz
17. 437.▲
    by XCSme 8mo ago

    This matches what I've been noticing. A lot of AI crawler traffic just doesn't show up clearly in typical analytics dashboards, especially when tools aggressively filter or sample. Part of why I built UXWizz was to avoid black-box
18. 438.▲
    by XCSme 8mo ago

    Strong agree on clicks over surveys. Once I move from a video prototype to a live MVP, the real signal comes from watching what people actually do. I built UXWizz mainly for this. Self-hosted heatmaps and recordings make it pretty obvious w
19. 439.▲
    by XCSme 8mo ago

    You can hover over some stuff, click on the model to get more info like tested categories, hover the correct test numbers to see some info about what they got wrong. I just started on this, so currently adding more tests and I keep improvin
20. 440.▲
    by XCSme 8mo ago

    Yup, for example I tell Claude to return ONLY the answer as "LEFT" or "RIGHT". And it outputs: **RIGHT** With markdown bold formatting... This is probably fine in a chat app, but when you use this in a workflow, it will
21. 441.▲
    by XCSme 8mo ago

    Lands 6th of my random benchmarks: https://aibenchy.com/
22. 442.▲
    by XCSme 8mo ago

    No AGI yet: https://aibenchy.com/model/openai-gpt-5-3-codex-medium/
23. 443.▲
    by XCSme 8mo ago

    Also, not really tied, Opus has a lot better consistency and reasoning score (which means the reasoning made sense, only the final output was wrong).
24. 444.▲
    by XCSme 8mo ago

    Yes. Opus could do a lot better, but fails a lot because it doesn't respect the given formatting instructions/output format. I could modify the tests to emphasize the requirements, but then, what's the point of a test. In rea
25. 445.▲
    by XCSme 8mo ago

    That's interesting. There's not much we can do to test whether we get the same model...
26. 446.▲
    by XCSme 8mo ago

    I also don't trust the maxbenched results. I am thus making my own benchmarks: https://aibenchy.com
27. 447.▲
    by XCSme 8mo ago

    I got similar results for most models, with gemini 3 flash (with reasoning) being the most consistent/reliable model: https://aibenchy.com I also noticed the same thing: some models reason correctly but draw the wrong concl
28. 448.▲
    by XCSme 8mo ago

    Fun read. Math makes so much intuitive sense in his head.
29. 449.▲
    by XCSme 8mo ago

    Thanks, just added UXWizz there. First one in its city in Romania :D
30. 450.▲
    by XCSme 8mo ago

    How reliable is sendBeacon though? Last time I checked, maybe browsers had it disabled by default.