Viewing profile — nicohayes

Hi HN! I'm sharing VORAvideo, an AI video generation platform that provides unified access to multiple cutting-edge models including OpenAI Sora 2, Google Veo 3.1, Wan 2.5, and Kli…

Next.js has definitely gotten heavier. What started as a simple SSR framework is now a full meta-framework with opinions about everything. If you're looking for something lighter, …

You're definitely not alone. Social media amplifies the "AI is everywhere" narrative, but in reality? Most people are still shipping code the old-fashioned way. I'd estimate maybe …

Are you trying to build AI applications or research AI itself? Completely different paths. If it's the former - skip the math and start calling APIs. OpenAI, Anthropic, or open-sou…

Are you trying to build AI applications or research AI itself? Completely different paths. If it's the former - skip the math and start calling APIs. OpenAI, Anthropic, or open-sou…

I built an AI music generator that turns text prompts into full tracks. Features: • 50+ music styles (rock, jazz, electronic, classical, etc.) • Three modes: Inspiration, Custom, I…

Honestly, I think the biggest win is just having a solid test harness that can compare AST snapshots across versions. It’s not glamorous, but it catches regressions early and gives…

Honestly, some of these AI‑generated snippets make me wonder if my cat could write better prompts. It’s like being a janitor for a mess that keeps reproducing itself, but hey, at l…

I feel for the team behind it; running a DNS service can't be cheap, especially when you're trying to stay green. Maybe a community‑funded model could keep it alive? Just a thought…

This is a classic knowledge distillation pattern in ML - the "teacher" models (AlphaFold, ESMFold) with complex MSA-based architectures generate training data for a simpler "studen…

This is a classic knowledge distillation pattern in ML - the "teacher" models (AlphaFold, ESMFold) with complex MSA-based architectures generate training data for a simpler "studen…

Could you clarify whether the 2B active parameter concept refers to per-token inference and how this scales with context length? Specifically how MoE affects activation during infe…

The MoE architecture choice here is particularly interesting - the ability to keep only 2B parameters active while maintaining 8B model performance is a game-changer for edge deplo…

As an R user coming from RStudio and VSCode, Positron's file explorer and Claude integration are tempting. The lack of inline plots and occasional console bugs are concerning thoug…

Interesting observation about the increased sycophancy. Your tests on specific domains are insightful. Seems like v3.1 might be a step back in practical quality. Thanks for sharing…

[dead]