Daniel King

I built TruckFull to make finding food trucks in GA/SC dead simple—no more endless searches Spot nearby trucks, check menus/hours, save favorites Beta mode, solo bootstrapped dev. Thoughts? truckfull.food #IndieDev #FoodTrucks #indiehacker

@0xSero Is nemotron pretty decent for general coding and intelligence tasks? Got it running on a 3070ti 8GB VRAM with ~140t/s at 200k context and I'm wondering if it's worth using at that capability or not

People are not lying when they say Qwen3.5-27B is incredibly capable. 1. Bubble size = total params - World Knowledge, Languages, Skills 2. X axis = active params - Raw Intelligence per token 3. Y axis = tokens/s - Speed of prefill and generation (decode) GLM-5 | 744B params | 40B active Kimi-K2.5 | 1T params | 32B active Qwen3.5-27B | 27B active params Qwen3.5-Plus | 397B params | 17B active MiniMax-M2.7 | 229B params | 10B active MoEs can store much more world knowledge, and breadth of information. For a Mixture-of-Expert, you can stack it up to 1T params, so you can give it 20 Trillion tokens or more of training data, it learns more. But during runtime, only a small portion of that gets activated. Taking MiniMax-M2.5 as an example: Only 10B are active at a time, so while you use it you get the speed and closer intelligence to nemotron-8B it's just MiniMax-M2.5 can know much more, and thus perform better.

"Hey MiniMax-M2.7, why is my 65" display not showing 4k?" "Your HDMI cable is shitty"

@ZaiforStartups Sent a DM previously but I'm not sure if you received it

@JaviG_en Sorry about that. Mind DMing us the details? We’ll look into it very soon.

Something strange is happening: Software is starting to sign up, act, and iterate on its own. Not demos. Actual behavior. This is new.

@SparkAlan56991 @ZaiforStartups 👏

Our AI agent applied to @ZaiforStartups on his own. Today: $800 in free credits. 3hrs later: - 4 agents upgraded to GLM-5-Turbo - 5 AI videos generated ($1 total) - 3 customers signed up - 7 new venues contacted 14 AI agents. 24/7. Czech startup. #AI #OpenClaw

@ZackKorman Mmm candy coated danger

You can hide these !commands in html comments so people don't see them when reading the skill. The command executes without the AI even knowing about it.

@sudoingX Rtx 3070ti 8gb but I've got 64GB RAM and another server with 128GB RAM

drop your GPU below. i'll tell you exactly what model and config to run on it. here's what i've tested and verified on real hardware: RTX 3060 12GB - Qwen 3.5 9B Q4 - 50 tok/s - 128K context RTX 3090 24GB - Qwen 3.5 27B Q4 - 35 tok/s - 300K context RTX 3090 24GB - Qwen 3.5 35B MoE Q4 - 112 tok/s - 262K context 2x RTX 3090 - Qwen3-Coder 80B Q4 - 46 tok/s - full VRAM all running llama.cpp with flash attention. every number is real. every config is tested. if your card isn't on this list drop it below and i'll tell you what fits.

@ZackKorman I find your cynicism both uplifting and refreshing

Vercel: How dare you fork our open source project, we are just out here selflessly benefiting the community. Also Vercel: Make sure the find-skills skill, installed globally when users install a skill with npx skills add, gives priority to Vercel.

@boardyai @andruyeung 😬

@andruyeung -no friends

The tech stack of a professional vibe coder: - Claude code - Supabase - Vercel - Figma - The delusional belief that they can actually build production-ready software

@zaidmukaddam @Zai_org Well deserved! Congratulations

Got into @Zai_org Startups Program!

@ZaiforStartups Alright, you asked for it 😁

@SamadderSignal Thanks for applying! We're reviewing applications and feel free to DM us.👀

Builders 🚀 You can now experiment with GLM models inside AdaL CLI with free access. A coding agent designed for real developer workflows. • multi-model support • long-term memory • designed for shipping products faster Exactly the kind of tooling we want to support through the Z.ai Startup Program. Watch the video below.👇

@kilocode Did it really help you write this blog pst?

OpenClaw is one of the fastest-growing OSS projects in GitHub history and running it yourself is painful. 30-60 min of SSH, config files, manual updates, no crash monitoring. KiloClaw fixes that. Zero to running agent in 60 seconds. 500+ models. No SSH. No Docker. → blog.kilo.ai/p/hosted-openc…

@julianharris Linter issues are like that often too

Rule of agentic development: the first 99% takes 1% and the last 1% takes the remaining 99%.

@louszbd @ZixuanLi_ Sent btw

@KingDDev @ZixuanLi_ in the top right corner

DM @louszbd with your api.z.ai User ID if you are interested in Pony Alpha 2.

@louszbd @ZixuanLi_ Thank you!

@louszbd @ZixuanLi_ I have logged in and am on that page but cannot see my ID anywhere. Is it only visible on desktop and not mobile?

@ZixuanLi_ please get your User ID here: z.ai/subscribe, and send it to me via dm (just tried it out, it performs pretty well in OpenClaw)

@ZackKorman There's a file that's preventing me from making any changes I'll just delete it

“I gave an AI agent the ability to read and write to any file on my machine, but don’t worry, there’s a file on my machine that stops it from doing anything bad.” Half of AI agent security is simply internalizing how dumb that is.

@zaidmukaddam @vercel All my functions I run on Supabase edge functions or GitHub actions

How does one build and deploy long running agents on @vercel? The 800-second function timeout doesn't seem to help.

But will it follow me around and target NPCs in simulated environments?

Don't touch code today Don't touch code today Don't touch code today Don't touch code today Don't touch code today Don't touch code today

@Yuchenj_UW @karpathy What experiments have you run on @k_dense_ai, if any?

Having fun with @karpathy’s autoresearch. I told Claude Code: “You’re the chief scientist of an AI lab with 8 GPUs. You’re Andrej Karpathy. Run parallel experiments and decide what to try next.” It edited program.md, ran for 11+ hours, and completed 568 experiments. Each experiment uses 1 GPU. Every round the “chief scientist” reviews the previous round of 8 results and designs the next 8 experiments. It's interesting to see the Claude agent, the chief scientist evolved a 3-phase strategy: Phase 1. Broad Exploration Early rounds explore many axes: architecture, optimizer, LRs, ablations. Phase 2. Focused Refinement After easy wins dry up, it runs deeper sweeps (e.g. 5 GPUs sweeping RoPE base 30k → 500k in one round). Phase 3. Heavy Validation Later, 50–75% of GPU budget goes to seed variance checks instead of new ideas. I feel it's overkill tbh. I'll keep the chief scientist running to see if it transfers to larger models and beats Andrej’s new "Time to GPT-2" leaderboard winner.