Jabir | AI Systems

@mschoening Notion team, huge respect 🫡

I was very close to quitting Twitter. Now the algorithm is good again. God, damnit! Also, 👋 to people I actually know and care about.

@kilocode @anacondainc Congrats 🎉

🚨 BIG NEWS: Kilo Code has been acquired by Anaconda (@anacondainc)! We've grown our agentic engineering platform from zero to a thriving open-source community of 3M developers in only 16 months. Now, we're joining Anaconda's trusted foundation to cover the full AI-native dev lifecycle. 🐍💛 Read the full article below!

moved from grok build to codex after hearing and seeing 5.6 performance

@jun_song Sounds about right

Upcoming models performance predictions : Opus-5 = Slightly worse than Fable, similar to GPT-5.6 Kimi-K3 = Very fast, slightly worse than Fable (tested) Deepseek-V4 = Opus-4.8 GLM-5.5 = Fable MiniMax-M3 Pro = Slightly worse than Fable Gemini-3.5-Pro = Only good at benchmarks

Perfect for developers building: - SaaS products - APIs - Mobile apps - Internal tools - Developer platforms -/github.com/anomalyco/openauth

This repository is a great way to learn how modern auth systems actually work: - OAuth 2.0 - PKCE - Access & Refresh Tokens - Social Login - Identity Providers - Self-hosted Authentication Highly recommended for every backend engineer.

Open-source OAuth 2.0 authentication server you can host yourself. Build your own auth service, a self-hosted OAuth 2.0 authentication server that supports Google, GitHub, email login, custom providers, and modern deployment platforms.

Why this matters: wound monitoring, skin cancer screening, scar assessment—all benefit from 3D. Off-the-shelf cameras become 3D scanners. Paper: arxiv.org/abs/2607.13010…

Key challenge: monocular depth is scale-ambiguous. DermDepth solves it by learning a metric scale prior from training data, achieving sub-mm accuracy on skin lesions.

DermDepth: Monocular 3D for Dermatology DermDepth: monocular 3D reconstruction with metric scale for dermatology. Single RGB image → depth + surface normals → accurate skin measurements. No special camera needed.

@sama half price checks out exactly on the sticker ($5/$30 vs $10/$50). the token efficiency number moves around depending on task though, seen anywhere from 19% to 54%+ fewer tokens depending on the benchmark, so “twice as efficient in many cases” is fair as a range, not a flat maxing

GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task. happy to deliver at one-quarter of the price.

@tamrrat Amazing results

Introducing Mint MCP - It turns your coding agent into a 3D engine! ✅ Create Threejs apps and games ✅ Generate 3D models ✅ Rig and animate models ✅ Generate Gaussian Splat worlds ✅ Generate PBR materials ✅ Generate themed Asset packs ✅ Generate audio / sfx ✅ Retopologize 3D models ✅ Convert 3D models to any popular format All the scenes in this demo were created with 5.6 + Mint Mcp + Mint Threejs Skills (links below)

@CodeByNZ Agent Swarm scaling up from K2.6’s 300-agent setup is the part I’d actually want to see benchmarked, that’s usually where the marketing and the real-world session diverge hardest

👀 Kimi K3 is now live. Moonshot AI has released its new flagship model, Kimi K3, which is already appearing in the Kimi app, CLI, and desktop version. The standout feature is K3 Agent Swarm, which supports massive parallel search and batch processing allowing users to get significantly more done in a single session. The model builds on Kimi’s reputation for strong agentic performance and long-context handling. Early users are already testing it across coding, research, and multi-step workflows. It’s one of the more interesting releases from a Chinese lab in recent weeks.

@jun_song What about token limits

Kimi-K3 feels like Opus-5 with superfast mode enabled.

@MiaAI_lab makes sense honestly, the compression tradeoff usually shows up hardest on tasks needing precision over many tokens. coding is exactly that, one wrong token in a diff and the whole edit is garbage, versus chat where a slightly-off word doesn’t break anything

Unfortunately, Bonsai 27B falls short on coding tasks 🙁 It hallucinates quite often, tends to rewrite entire files instead of making clean edits, and can easily get stuck in loops. It’s noticeably weaker than vanilla Qwen 3.6 27B/35B. That said, it performs well in agentic workflows where tool calling is needed, and it might be decent for simpler coding tasks. I'm sad. See an example in the next post.

Full answer on Stack Overflow: stackoverflow.com/questions/1269…

TypeScript catches type mismatches at compile time, not runtime. No more string concatenation surprises. It compiles to plain JS, works everywhere JS does. Start small, add types gradually.

Why TypeScript? A Quick Example Why TypeScript? One function says it all. 👇 JS: function add(a, b) { return a + b; } add(5, "10") → "510" (bug!) TS: function add(a: number, b: number): number { return a + b; } add(5, "10") → compile error

AutoGPT is not the most advanced today, but it's the project that made autonomous agents accessible. Repo: github.com/Significant-Gr…

This design influenced modern agent frameworks like LangChain, CrewAI, and AutoGen. The idea of agent + tools + memory is now standard.

AutoGPT: The Open-Source AI Agent That Started a Movement AutoGPT was the first open-source project to show how GPT-4 could autonomously decompose goals into sub-tasks and execute them. Still worth studying today.