Elektrine lite

← Feed

Quantisan

quantisan@hachyderm.io

<p>I&#39;m curious whether agents can help more people take on problems where a bad answer is worse than none. I have lived in Japan since 2018, where I bake sourdough without keeping a starter, and travel with my family to old country inns -- ryokan 旅館</p>

Posts

  • View post

    AI coding agents still feel like a very early stage, in terms of actual developer experience. Think about how Vim came about, Emacs, text editors, IDEs, all that. It took years and years. The human interaction, the human-computer interface aspects of AI use, are still so nascent.

  • View post

    Drafted many PRDs with AI. They come out pretty shallow. Obvious, but not quite right. Always ends up like that, regardless of how much Claude Code changes or the models change. Reverting to the mean? So now I ask the AI to never draft for me. Fill in the blank, frame it, add bullet points, bring out the evidence relevant to that section, just dump it there. Then I draft. Super clunky. The automation dial&amp;#39;s turned a bit too far off. But there&amp;#39;s something to forcing it make less...

  • View post

    Good thing with LLM is that I dont need to care about my grammar anymore. Because I&amp;#39;m real!

  • View post

    For Product Discovery work, how do you set yourself up to tackle a tree of separate opportunities vs a stack of prerequisites? I thought I have 3 separate bets, but now first feedback is the list seem like a sequence of requirements until the last, which is the differentiator... sigh

  • View post

    Claude Code added a `/branch` command last week!

  • View post

    Been stepping back and focusing on one idea only...

  • View post

    From 2024 How spammers and scammers leverage AI-generated images on Facebook for audience growth https://misinforeview.hks.harvard.edu/article/how-spammers-and-scammers-leverage-ai-generated-images-on-facebook-for-audience-growth/

  • View post

    Been a Opus fan for the past year (Claude Code main driver, Codex for second opinion), but this week is the first time I&amp;#39;m switching the two for non-coding work. I&amp;#39;m liking Sol for slow thinking work. It just reads better. I find the quality maybe a bit less, but it&amp;#39;s just easier to understand instead of repeatedly needing to prompt Claude to &amp;quot;say it again simpler for clarity&amp;quot;. Same task, word counts of response (making similar points): ``` Fable 5.0...

  • View post

    Got curious about this year’s Fields Medal. Watched a couple of explainer videos on the Kakeya conjecture. I have particular trouble seeing the connection between Kakeya and Harmonics. So I did what every AI fan would do: asked Fable 5 to do the research and build an educational game. https://quantisan.github.io/kakeya-game/ 5M tokens later, I’m still clueless. The game Fable built seems to repeat one of the videos almost step by step. Normally, you’d prompt an LLM with your question and ask...

  • View post

    Spent the weekend occasionally driving Claude on my phone to refactor my Pelican blog onto a leaner generator (https://github.com/borkdude/quickblog) with hands-free publishing process (Github Actions, and Cloudflare Pages). Now I can write a post once in a blue moon, `git push`, and it’ll just work. Had too much fun and redesigned the site with Claude Design too. https://www.quantisan.com

  • View post

    Been heads down building for 6 months. I looked up in July and just feel lost. Fast forward to today, the problem is that I&amp;#39;ve been building what I think should be instead of what&amp;#39;s needed. Since August, I&amp;#39;ve been focusing on stopping and listening. Reached out and been listening to people ever since. I still can&amp;#39;t shake the feeling that I&amp;#39;m not doing enough though.