davepoon

Inspired by the incredible real-time AI vision demos and ngxson's live SmolVLM + llama.cpp project, I wanted to see what was possible on Apple Silicon! 🍏 Super excited to share my SmolVLM Real-Time Webcam Demo with MLX-VLM! 😙

The more you do with AI, the more time you’ll spend with AI, anxious about falling behind those who use it even more.

@ivanzugec I wish! It makes no money but costs money and personal time to run… I need someone to sponsor me 😆😆

@davepoon Nice!!! You ready to retire now.

It's now easier than ever to get your Skills discovered, with over 43k visitors in the past 30 days, and growing.

On the flip side, I spent a few days using the Codex app with GPT-5.4 Extra High and GPT-5.3-Codex Extra High (fast mode) to plan and build an app. About 60% of the time was burned just writing test cases, the end result was completely unusable, and I ultimately had to rewrite the entire codebase with Opus just to get it running.

Just let Opus go for over an hour on a new feature. When it was done, I asked how I can test it. 20 minutes later, it realized I can't test it because it did the whole thing entirely wrong. Idk how you guys use this model every day for real work 🙃

Simply add your skills to your marketplace.json, and we will automatically index them. You can also share your skills directly using a GitHub URL. Once submitted, your skill will be reviewed and featured on the BuildWithClaude website: buildwithclaude.com/skills

GitHub Copilot CLI is getting a lot better now, it produces better results than Gemini CLI, what a surprise!

@itspoidaman It's the same when coding in the Codex app. It takes a long time and seems to get stuck in a loop just to change 100 lines of code for a test case. I'm a bit disappointed with GPT-5.4 for coding, I'm considering cancelling my subscription.

I asked gpt 5.4 to write me a prompt and do some research about it to make it effective and it's been like 10 minutes and it's still thinking in a loop about it amazing

Just updated the OpenClaw template to v2026.3.13 with bug fixes. It will now automatically detect the gateway token for the Gateway Dashboard. Grab all the latest UI updates in v2026.3.13! railway.com/deploy/opencla…

AI coding agents are accelerating releases like never before. Seeing 10 new versions of the same npm package in 8 hours is becoming normal. But how do we manage the flood of updates and breaking changes? Feels like we're seeing more breaking changes than ever in the AI era.

@ashishguleria_ Hi, nice to meet you 🙂

If you're in tech. Let's connect. 🫡

Opus 4.6 High is still significantly superior and faster than GPT-5.4 Extra High. GPT-5.4 Extra High couldn’t figure out how to fix the Drizzle DB schema mismatch between my production and local databases. Opus immediately knew the fix, and told me just run and done…😅

@Prince_Canuma @neural_avb Wow, that's great! I'm still using an M1 Max and thinking about a big upgrade by buying a MacBook Pro M5 Max.

@neural_avb It can run LTX-2 even 16GB M1 We have some settings you can adjust to ensure you don’t run OOM

MLX-Video updates coming soon 🚀 Took a bit longer to debug LTX-2 dev model and 2.3 but we are almost there! Note: PR is tbd, so bear with me :) github.com/Blaizzy/mlx-vi…

@rabi_guha Wow, great work! I’ll definitely give it a try over the weekend.

We built a better alternative to json-render & A2UI Up to 3x faster rendering & 67% fewer tokens. Far fewer broken outputs. We've been shipping Generative UI interfaces to 10,000+ developers for the past year. We know where JSON-based rendering breaks in production. So we built a new format - one that works with how models actually generate, not against it. And today we're open sourcing it: openui [dot] com

@ctatedev Interesting~ it’s already giving me ideas for a few use cases for a vibe app.

Generative UI in json-render has two modes: inline - AI outputs text + JSONL patches interleaved. It decides when UI beats text. For chatbots, copilots, assistants. standalone - JSONL patches only. No prose. Prompt in, UI out. For website builders, text-to-widget, dashboards.

Do you use GPT-5.4 xhigh or GPT-5.4 high? Which one produces better results for you? Many people say GPT-5.4 xhigh tends to overthink. Ironically, quite often GPT-5.4 high produces better results. Is that true?

Built this site when Claude introduced subagents. Originally just to share my own curated collection for Claude Code. Didn't expect it to grow into a discovery hub for Claude plugins and skills, but here we are. For plugin creators: Publish your plugins on GitHub and the site indexes them automatically. For skill creators: Add your skills to marketplace.json for automatic indexing, or share a URL and we'll review and add it to the site. If you're building for the Claude ecosystem, this is where people will find your work. buildwithclaude.com

You can now add and index your skills on buildwithclaude.com/skills just by sharing a URL. No more manual setup. Submit your skill link, get it indexed faster, and make it discoverable to 10k+ users looking for useful AI workflows, tools, and skills.