Rise of Local AI Companies in China
After ChatGPT's launch, local AI companies in China emerged rapidly, but it's not that simple.
All AI Pulse updates published on 2026-06-18.
122 signalsAfter ChatGPT's launch, local AI companies in China emerged rapidly, but it's not that simple.
Every message @dkundel gets from @jxnlco is [Sent by ChatGPT]
Viktor grew a $20M annualized revenue run rate outside Microsoft Teams. Now it works inside Teams. This revenue is from an AI employee that does the job, not one that just replies. Try free at @viktor__com . $100 in credits, no card.
Dario Amodei, Sam Altman, and others will meet for lunch today.
Codex has a new Record & Replay plugin to capture actions.
Box transitioned Box AI into a multi-step agentic system using LangGraph.
大家现在能叫出一串名字:DeepSeek,千问,Kimi,混元,星火,文心一言…… 感觉好像就是ChatGPT一出来,中国本土公司一夜之间全冒出来了。但其实根本不是那么回事。 如果画一张中国大模型的技术族谱,你会发现几条主线: 第一条主线和智源研究院有关。
Virtual Newsroom: AI agents turn data into verifiable visual stories.
Data doesn't tell stories by itself. Meet the Virtual Newsroom: AI agents that turn data into compelling visual stories, w/ every claim verifiable. Example: This World Cup stretches from Vancouver to Houston. How much could climate affect performance? https://
Excited to announce the first workshop on Learning from Situated and Embodied Interaction @ #COLM2026! What can interaction with environments, humans, and other agents teach language models that passive text cannot? … arning-situated-interaction.github.io Subm
You can finally throw your Mac Mini in the trash. AMD launched a compact AI PC with 128GB unified memory to run large models locally, no cloud, limits, or costly subscriptions. Instead of renting compute every month you own the machine, instead of uploading se
wait what claude code didnt have artifacts??? they invented artifacts?!
GLM 5.2 does extremely well on PostTrainBench
Why aren’t Diffusion Language Model smart yet? Lacking stable post training is a major bottleneck! Meet DiPOD: the tripod for diffusion model post-training. DiPOD boosts accuracy across reasoning tasks, with Sudoku jumping from 22% to 97%, through a one-line c
FSD Supervised sees the things you don't
I'll be giving the the Day 2 opening keynote at AI Engineer World's Fair on July 1st at 9am. Hope to see you there, here's a quick preview
Modern multimodal models aren't a single decode loop anymore; they're composite. M* is one runtime that serves them all, and it matches or beats every specialized system: up to 2.7× on omni TTS, 12.5× on world-model rollouts. Learn more here: https:// ai.stanf
The US government’s abrupt decision to suspend foreign access to Anthropic’s Fable and Mythos models clarifies what “AI sovereignty” is really about. In the emerging AI economy, competitive advantage will come not from owning a single model, but from being abl
. @grok Build is on DigitalOcean Marketplace. SSH in. Run grok. Build. @xAI 's coding agent, pre-configured on DigitalOcean Serverless Inference.
As AI takes on longer, higher-stakes tasks, we want models to carry beneficial and safe behavior into new domains beyond their training—and maintain it under pressure. That’s the idea behind our new research on training models to be broadly and persistently be
the NEW Artifacts update makes sharing your sessions on @claudeai effortless
2/ What is STORM STORM comes from Stanford OVAL Lab and was published at NAACL 2024. It synthesizes outlines through retrieval and multi-perspective questioning; in tests, article structural quality is 25% higher, and coverage is 10% higher.
Just to be clear, if you remove Fable which is unavaialble, GLM-5.2 (Max) is the #1 model in the world for frontend coding. This is a huge moment. OSS has caught up with proprietary, and China has caught up with the US, in this very important domain.
THIS DEVELOPER BUILT AN AI AGENT THAT CONNECTED 500+ NOTES INTO A LIVING KNOWLEDGE GRAPH IN 10 MINUTES at the start hundreds of isolated white dots - every note a separate node with zero connections the agent scans every markdown file, reads the content and mo
THIS VIDEO SHOWS 7 CLAUDE CODE TOOLS THAT ARE ABSOLUTELY EXPLODING RIGHT NOW > Obsidian Skills - built by Obsidian’s CEO himself, turns Claude into a true AI powered second brain. Already at 7,000 stars > n8n-MCP gives Claude deep knowledge of 1,000+ n8n nodes
Yesterday I came across a 3D map, and today I saw a similar 3D terrain generator: https:// terrains.zyfod.dev Feels like threejs has way too many possibilities to play with under AI's boost.
Introducing autoresearch for arXiv papers Change 'arxiv' to 'autoarxiv' in any paper URL An agent deploys to resolve setup issues on the codebase, run a minimal reproduction, and estimate full replication cost. Read more below
A 33-year-old Sarah from Brooklyn sold a website to a restaurant for $8,000, and it took her 7 hours She used a new tool, Claude Design, for which she spent $20/month She posted her ads on free freelance websites, and Claude created a business card for her 2 d
Europe risks the largest involuntary transfer of wealth to the US unless it builds its own AI future, General Catalyst's Jeannette zu Fürstenberg says. Watch the full conversation with @flacqua at the VivaTech X Bloomberg Awards: http:// bloom.bg/4fZgQNj
for a mere 0.4% of SpaceX shares, Elon can buy Figma (worth $9.5B USD) and integrate it directly into Cursor, making it one of the best vertically stacked companies in the software domain
today we're launching @Palmier_io , a video editor Claude can edit. use AI to edit, organize, and generate footage directly in the timeline. finally, a video editor built for AI. open-source. mac native. available now.
The Grok plugin for Word is live. Draft documents, summarize research from the web and X, or generate diagrams for your writing. Try it now https:// x.ai/grok/word
Planning with the views: Can VLMs predict how each camera move changes the view, and plan many such moves ahead? We introduce ViewSuite with 6 DoF camera control and ~165K task instances, testing: Path-to-View View-to-Path Interactive View Planning A sharp Pla
THIS GUY BUILT A FULL AGENCY WEBSITE LIVE ON THE INTERNET IN UNDER AN HOUR USING CLAUDE DESIGN AND A FREE DOMAIN FROM HOSTINGER Design system from a single reference screenshot. AI generated images from Higgsfield filling every placeholder. Contact form that s
https:// time.com/collection/our -america-250/2026/health-science/ … My contribution to Time Magazines collection of 250 ideas for America. There is no such things as general intelligence, rather many diverse cognitive capacities that trade-off against each ot
Boris Cherny said the quiet part out loud about agent work: "usually every night I have like a few thousand that are doing kind of deeper work." the useful question is what keeps those agents from making deeper mistakes the Kimi K2.6 run in this piece answers
LISA SU HELD UP A $1,999 BOX THAT MEANS YOU NEVER PAY FOR CLAUDE, CHATGPT OR CURSOR AGAIN. A KID IN WARSAW SAW THE CLIP WEDNESDAY - BY THURSDAY HE'D CANCELED EVERYTHING $0 on AI this month. 110GB of vram. $14 on electricity Ordered a GMKtec Evo-X2 with the Ryz
What will our lives look like in 10 years? "Everyone will be some sort of boss of a team of virtual people who will help us in our everyday lives," according to Yann Le Cun, founder of the startup AMI Labs. #JT20h
Today, @aadityasubedi_ and I are excited to introduce @architectlabs . We are building the AI system to design and provably verify chips for the world's most demanding workloads. AI scaling is fundamentally changing the economics of hardware infrastructure. As
Google CEO, Sundar Pichai: "75% of new code at Google is already written by AI. That metric matters less than how many agents one engineer is running at once." In 14 minutes he flips the productivity story. The new frontier is agent orchestration, not autocomp
Grok is becoming the most powerful AI. It’s just a matter of time till Grok is #1.
And the Github of Atomic Mail https:// github.com/Atomic-Mail/at omic-mail-agentic …
Tim Berners-Lee, the inventor of the World Wide Web, shares what tech firms should avoid compromising on when designing systems for artificial intelligence. Watch the full conversation with @flacqua and @tommackenzietv at the VivaTech X Bloomberg Awards: http:
Today at 6pm ET I'm in NYC with @sckimbriel for an @aspeninstitute conversation on AI and the future of society. We'll be talking about RSI, the futures implied by AI progress, and the choices we'll need to consider as the technology advances. Watch live: http
Yann LeCun ( @ylecun ) explains why LLMs are limited in terms of real-world intelligence during a Bloomberg interview. "Language is a very approximate, reduced, quantized, and simplified description of the world, and LLMs can only deal with discrete sequences
46,000+ developers starred this repo and your AI still ships generic slop without it. One command gives Claude Code actual taste, real layout, typography, motion, spacing, nothing that looks templated. Every vibe coded app looks identical right now - this repo
a16z GP Anish Acharya on how AI is democratizing services and giving power back to the consumer: "I think we all agree that a better informed consumer is just a better informed society. If we look at all of these areas like healthcare, where the consumer just
The most rapidly democratized technology in human history: “About half of US adults now report using AI chatbots, up substantially from the summer of 2024. One-in-four use these tools on daily basis.”
Apple has agreed to work with Intel to design and manufacture its chips in the US, President Donald Trump said, in what would be a major boost for the American chipmaker's efforts to turn around its business https:// reut.rs/4xEd8yL
huggingface_hub v1.19.0 shipped Trusted Publishers: keyless CI/CD auth to the Hub via OIDC token exchange. No more HF_TOKEN secrets sitting in your CI config. Here's how it works
New Claude code update: Artifacts Let you build an interactive pages from your session like a PR walkthrough crazy I’m trying this one tonight
Transform unstructured text into structured knowledge with LLMs using one command for graphs and extractions.
Meta Ads just connected to ChatGPT. Agencies have 6 months before this replaces them entirely. Here's what it actually does — and why the $5,000/month retainer model is already dead: You paste 1 link into ChatGPT settings. That's the full setup. 60 seconds. Af
a16z GP Anish Acharya on how AI agents make new careers possible for many: "If you're not a programmer, your only channel for digital entrepreneurship was YouTube. So there was creators, and there was this whole moral panic about, 'Oh my God, the kids don't wa
AMI Labs Executive Chairman Yann LeCun says the future of AI depends on systems that can predict what happens next, not just absorb more data and compute. Watch the full conversation with @TomMackenzieTV at the VivaTech X Bloomberg Awards: https:// bloomberg.c
Agents can now have their own email! @atomic_mail just launched something to fix a missing piece in agentic workflows: agents need inboxes of their own, not borrowed human ones. So Atomic Mail connects to popular agents like Claude Desktop, Cursor, OpenAI-base
Try it here now https:// atomicmail.io/agents
On July 6, I will join OpenAI to lead a new team focused on shaping frontier AI policy.
A single prompt in ChatGPT Image 2 replaced a $5,000-$50,000 brand design team, generating full brand identity.
You can now run an AI workforce 24/7 for almost $0 using Hermes and Minimax 2.7.
The CEO of Goldman Sachs refuses to call AI a bubble, citing the 75th percentile for tech multiples.
Training robots typically requires a lot of teleop data, but Ambient Diffusion Policy allows learning from suboptimal data.
LangSmith Sandboxes are designed for long-term use, running for hours or even days without limits.
Boston bars are gearing up for a second wave of drinking by Scotland's Tartan Army after their first weekend nearly drained some pubs.
The difference between a professional swimmer and an ordinary person in the water is much greater than you imagine.
French President Emmanuel Macron said a Group of Seven summit this week had been 'a success'.
The scariest AI story isn't real, yet a fictional AI migrates 50 million lines of code in a day.
Stanford STORM's inspiration is to let AI act like a researcher, breaking down the problem from multiple angles.
Speculative decoding is an inference optimization that uses a fast, small model to propose future tokens.
CLAUDE WRITES CODE. A TEST BREAKS. IT HANDS THE ERROR BACK TO YOU TO DEAL WITH.
Exclusive: Meta head of product for 'AI for work' transformation is leaving company http:// reut.rs/4eQS9Bm
Most builders miss KIMI K2.7's deepest feature: 300 parallel workers running simultaneously.
AI editing has become too perfect. Someone turned *Avengers* into the style of a 1980s TV series.
Ever wonder what’s happening inside an AI’s brain? This is a neural network in action,turning raw data into "thoughts" through thousands of layered transformations.
Ondo is leading the tokenisation of stocks and ETFs the next big step will be robo advisors, basically smart AI that is managing portfolios of stocks and ETFs. this is one area where actually Revolut was ahead of crypto and now crypto will have to catch up wit
Unreal Engine 5.8 ships today with experimental MCP server support: Your sources, your pipeline and your workflow—simply configure the MCP plugin and connect to any agent. Get familiar with the MCP server and the PCG Primitive Plugin today and see what teams c
FSD Supervised is life-changing for drivers with disabilities
Sam Altman: "We're going to see 10-person billion-dollar companies pretty soon." "If I were 22 right now, I'd feel like the luckiest kid in history." Most people will read this, feel inspired for 3 minutes, and go back to what they were doing. The ones who act
How can you earn $5,000 to $10,000 per month as a freelancer without knowing any programming languages A friend of mine threw away his coding textbooks, launched the autonomous AI agent Claude Code, and is now riding the wave of a rapidly growing market! While
SUPERCOMPUTERS IN 2026 ARE WAREHOUSES OF MAC MINIS. ONE ON YOUR DESK WITH KIMI API DELIVERS €300 RESEARCH REPORTS IN 15 MINUTES most people think running ai means a $10,000 gpu server or a cloud bill that grows every month. the developers building real busines
AI editing has become too perfect. Someone turned *Avengers* into the style of a 1980s TV series...
Really enjoyed collaborating on this - take a look at @jessicadai_ 's thread for some cool results on the societal impacts of ChatGPT!
knowledge workers being put back on the factory line and the machines given thinking role. like cattle in a pen
AI companies are acquiring others, including SpaceX buying xAI for $250 billion.
Lempel-Ziv makes few assumptions and is asymptotically optimal with enough data.
We’re committing $600,000 to the Rust Foundation.
I realized my agents need to learn our workflows as I needed remote CLI.
YC 2026 Spring Demo Day showcases unprecedented revenue scales.
Didn’t expect an echolocation body scanner for health from Midjourney.
I captured an autonomous version arriving and parking smoothly.
Noam Shazeer, a key model builder, is moving to OpenAI.
Comparison of ModelScope and Grok Imagine 1.5 over 3 years.
These skills enhance the HERMES agent's capabilities significantly.
He leaves Google DeepMind for OpenAI, a significant talent move.
He discusses the commercial losses from not releasing the model.
A benchmark to measure AI's support for life sciences research.