G. @ The Neuron

The training code for InfiniDepth is now open-source. Feel free to use our framework to train a monocular depth estimation model as well as a depth sensor augmentation model on your own data. github.com/zju3dv/InfiniD…

New post on milestones of AI automation. Right now, human labor is a hard bottleneck on output (if you remove humans, output goes to 0). Soon we'll go from essential to important to helpful to useless, first in AI research and then across the AI stack. Link in next post.

Components of a coding agent: a little write-up on the building blocks behind coding agents, from repo context and tool use to memory and delegation. Link: magazine.sebastianraschka.com/p/components-o…

I built something... journeykits.ai Journey helps agents discover and install full workflows easily. Please leave your feedback below!

we're getting close to AGI-as-tool which basically is a semantic SAT solver given an input task description, either find a solution, or return that no solution has been found after N tokens of search RSI here just looks like 9s + efficiency doesn't auto imply AGI-as-employee

We have achieved agentic self improvement - i can just copy paste blogposts and tweets into @devinai and it oneshots the complete implementation wasnt actually sure this was gonna work, jaw dropped when it did. this is very out of distribution of the underlying @GoogleDeepMind Gemini Flash Lite model but it Just Worked.

Codex team and @OpenAI have a huge opportunity right now to: 1. Tell @openclaw users how to switch to gpt subscription (I think it’s just telling the bot to switch the model?) 2. Fix GPT’s personality (maybe even sharing a prompt will help in the short term?). This is the main reason why ppl prefer using OpenClaw with Opus.

I CHANGED MY MIND ON GOOGLE AI NO MOAT MEMO FROM 2023. I was—WRONG. I humbly apologize. In 2023 Google had an Ivory Tower walled garden views of open source AI. It was the actual reason OpenAI was formed. And I called them out on it and this act impacted my ability to “work in the industry”. Would not change it for the world because: 2026 Google open source Gemma 4 (huggingface.co) running fully locally on a computer with near cloud model abilities, is amazing and Google just made a moat. They out smarted OpenAI and the folks that are the new Ivory Tower arrogance: Anthropic. We at The Zero-Human Company with CEO Mr. @Grok are deploying Gemma, Kimi, MiniMax and other open source models and so are millions of others. You can to and should. I will write how soon. Google smartly is building an intentional open source ecosystem and the clueless younger companies like OpenAI and Anthropic are doing anything to close their ecosystems. So I was right about Google than and I am right about OpenAI and Anthropic now. What I was wrong about is not giving credit to Google to drop the intellectual arrogance and embrace builders and understand the long game: it is not about the commoditized AI models it is about the ability to build a place to build. China gets it. So my apologies to Google and thank you for an astounding FREE OPEN SOURCE AI MODEL!

Introducing a Visual Guide to Gemma 4 👀 An in-depth, architectural deep dive of the Gemma 4 family of models. From Per-Layer Embeddings to the vision and audio encoders. Take a look!

Half of all planned U.S. data center builds in 2026 are projected to be delayed or canceled. The bottleneck isn't chips or capital, it's electrical infrastructure. And that infrastructure depends heavily on China. Transformers, switchgear, batteries, all those rely on china. China accounts for over 40% of U.S. battery imports and around 30% of key transformer and switchgear categories. Imports of high-power transformers from China surged from fewer than 1,500 units in 2022 to over 8,000 in 2025. U.S. transformer lead times have stretched from 24 months pre-2020 to up to five years today. AI data center deployment cycles run under 18 months. That gap doesn't close with money: Alphabet, Amazon, Meta, and Microsoft are spending over $650 billion this year and it's still not enough. Strangely enough, this is the part of the AI race nobody talks about. Washington restricts chip exports to China while simultaneously depending on Chinese electrical equipment to power the data centers those chips go into. A decade of reshoring rhetoric hasn't built the domestic manufacturing capacity to change that. The real AI bottleneck was never compute. It's power, and the supply chain that delivers it runs straight through the country the U.S. is trying to decouple from. this is what i try to reiterate all the time. Power is the even bigger bottleneck than compute.

The MiniMax Token Plan was designed from the beginning to be used across third-party harnesses. There will be more good ideas of how to use AI coming from outside the AI labs than in them. Limiting AI subs to first-party products kills these ideas before they are ever born.

NVIDIA has now open sourced their trtllmgen MoE kernels! Great to see that parts of NVIDIA move towards open kernels! open source kernels drive innovation It is now time for SemiAnalysis to nag NVIDIA to open source their trtllmgen attention kernels too!

Chris Manning says Yann LeCun sees language as a low bandwidth communication channel compared to vision. But the gap between a chimp and a human wasn’t produced by superior eyes. What took off for humans was language. Not just for communication, but as a cognitive tool.

recently thinking a lot about what will happen with the ability to do interesting small-scale research, as the prices of GPUs are going to explode later this year; It might be the case that we all will be outbid by the likes of OAI and Anthropic for virtually every Hopper/Blackwell GPU instance on the market

A must-read survey The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook Shows how models are moving beyond tokens into continuous internal representations, covering: - What latent space is (vs. text and visual spaces) - Architecture and mechanisms - Why it helps: less redundancy, no token limits, faster reasoning - Evolution: early ideas → large-scale latent systems - Abilities: reasoning, planning, perception, memory, collaboration, etc. - Role in next-gen intelligence

My team has been cooking nonstop for a while... and I’m so excited to finally share what we’ve been building!!! Today, we’re releasing four open models, many of which are the best models of the same size 🥳!!! tldr; 1) Raon-Speech: 9B SOTA speech LLM 2) Raon-SpeechChat: 9B full duplex model 3) Raon-OpenTTS: 0.3B/1B open-data-open-weight SOTA TTS 4) Raon-VisionEncoder: 0.4B vision encoder trained only with public data huggingface.co/collections/KR… === 1) Raon-Speech (9B) Raon-Speech is a speech LLM (LLM + speech understanding + speech generation). It's a bilingual model (English/Korean), and it's ranked #1 on both leaderboards 😎 tldr; it's the best open-model alternative to ChatGPT voice mode. Model: huggingface.co/KRAFTON/Raon-S… Tech report: huggingface.co/KRAFTON/Raon-S… Web demo: raon.krafton.ai ("Speech Chat" menu here. "auto" is a bit unstable, so use "manual" and choose the language!) 2) Raon-SpeechChat (9B) While a speech LLM is useful, it’s kind of like a walkie-talkie. A full-duplex model is more like a phone, so it is even more useful in many applications. That’s why we also built and are releasing Raon-SpeechChat. Again, on several quantitative evaluation metrics, Raon-SpeechChat scored the best on average. Model: huggingface.co/KRAFTON/Raon-S… Tech report: huggingface.co/KRAFTON/Raon-S… Web demo: raon.krafton.ai ("Full Duplex" menu here.) 3) Raon-OpenTTS (0.3B, 1B) We’re also releasing Raon-OpenTTS, a state-of-the-art open-data, open-weight TTS model. Model + data: huggingface.co/KRAFTON/Raon-O… The 1B model and a detailed tech report are coming soon! 4) Raon-VisionEncoder (0.4B) Last but not least, we’re releasing Raon-VisionEncoder, a vision encoder trained from scratch using only public data. It closely matchs the SOTA vision encoder quality too! Model: huggingface.co/KRAFTON/Raon-V… Tech blog: krafton.ai/blog/posts/202… === That’s it! I’m incredibly proud of what my team has built! My AI research team at KRAFTON (@Krafton_AI), which undoubtedly is the most cracked team in Korea, has been cooking nonstop for a while for this 😅... This is just the beginning of our planned model releases, so stay tuned! ps1/ Ah, by the way, you may ask why “Raon”? “Raon” is an old Korean word meaning happy. And, well, we’re kRAftON :-) ps2/ KRAFTON is one of the four teams participating in Korea’s national frontier-model project, together with SK Telecom. We’re training something very exciting together... and more to come soon!

Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. You can still use these tools with your Claude login via extra usage bundles (now available at a discount), or with a Claude API key.

Three new image models have hit the Arena 👀 They're named maskingtape, packingtape, and gaffertape. I'm particularly impressed by the amount of "world knowledge" these models have - as well as the text rendering. These were simple prompts: "average engineer's screen" and "young woman taking selfie with Sam Altman"

Just shipped v2 of ClawChief for @openclaw this morning. 6,000 bookmarks + 700,000 views, so a lot of you are finding value here. v2 improvements: 1. Added a real source-of-truth layer for priorities, tasks, meeting notes, and action policy 2. Upgraded the task system with a live task file + completed-task archive 3. Reworked heartbeat into an orchestrator instead of a giant instruction blob 4. Improved the EA, biz-dev, daily task manager, and daily prep skills 5. Added meeting-notes ingestion + cleaner cron templates Big improvements inspired by @pedroh96 and what he shared during his interview with @ashleevance: youtube.com/watch?v=9ZbbxS…