🇰🇷 Korea is quietly becoming the world’s humanoid robot factory.

Goldman Sachs Research estimates Korean companies could produce 30% of all humanoid robots by 2035, jumping from virtually zero today to more than 412,000 robots a year. The reason? Decades of automotive manufacturing have given Korea the motors, actuators, supply chains, and factories needed to scale humanoids.

The government is fueling the push with ₩700 billion ($500M) for robotics in 2026, aiming to produce 1,000 domestically built humanoids annually by 2029.

Investors have already noticed. LG Electronics is leveraging its massive motor business to supply humanoid robots, Hyundai Motor is combining its manufacturing muscle with Boston Dynamics, while Hyundai Mobis, Rainbow Robotics, Robotis, and Doosan Robotics are all positioning themselves across the humanoid supply chain.

For broader exposure, Korea’s new humanoid robot ETFs have surged in popularity, with pension funds pouring billions into the sector.

@aipost 🏴

We can all relate.

@aipost 🏴

🔥The race to dominate enterprise AI

Anthropic's recent surge has propelled it past OpenAI to become the leading paid AI provider for U.S. businesses, marking a shift in the AI race from model superiority to workflow dominance.

@aipost 🏴

🇨🇳 Ex Meta PM and AI founder Xiaoyin Qu says “American and European enterprises will ditch OpenAI and anthropic and adopt Chinese models.”

@aipost 🏴

Agibot robots demonstrate high performance in factory trial

Agibot humanoid robots reached a 99% success rate during a six-day live demonstration at a factory. Over 64 hours of operation, the robots completed 64,828 individual tasks and assembled 17,625 tablet units.

Progress in this field is accelerating, as developments of this magnitude were uncommon a year ago but have recently become more frequent.

📰 @aipost

Chinese AI models now offer processing at rates as low as 18 cents per million tokens, compared to the $4 average charged by leading models, according to CitiBank Research.

By June, 65% of open-source model processing on OpenRouter was handled by these lower-cost options, up from 34% in January. DeepSeek and other models from China have attracted interest due to their affordable pricing.

As companies shift focus to controlling AI costs, the decision to adopt a specific model is increasingly based on price instead of technical superiority. This trend is prompting leading providers like OpenAI and Anthropic to reassess pricing, as enterprises now compare models according to specific tasks.

Gartner projects that by 2028, spending on AI coding may surpass the average salary of a developer.

📰 @aipost

🤖 OpenAI is preparing to reveal new Codex hardware on July 15.

It looks like OpenAI is building a physical control panel for an AI coding agent.

@aipost 🏴

🤖 Anthropic just unveiled Claude Sonnet 5, its smartest and most agentic Sonnet yet.

The model can plan multi-step tasks, browse the web, use terminals and other tools, and work autonomously on problems that only much larger, more expensive AI models could handle a few months ago.

Compared to Sonnet 4.6, it delivers major gains in reasoning, coding, tool use, and knowledge work while approaching the performance of Opus 4.8 at a much lower cost.

Early testers say Sonnet 5 completes complex tasks that previous versions couldn’t finish, double-checks its own work without being prompted, and offers one of the best price-to-performance ratios for AI agents today.

Source.

@aipost 🏴

Ford engineers walking back into HQ

@aipost 🏴

Ⓜ️ Meta is reportedly telling engineers to be careful with Claude and Codex and the reason is surprisingly simple: AI can “catch” another AI’s knowledge.

According to reports, Meta has restricted the use of Anthropic’s Claude Code and OpenAI’s Codex for some engineering work to avoid contaminating its own AI training data.

Here’s the concern: if Meta’s future models are trained on outputs generated by rival AIs, competitors could argue Meta distilled their models instead of developing its own. Both OpenAI and Anthropic prohibit using their AI outputs to build competing models.

That doesn’t mean engineers can’t use these tools for everyday coding. The key is keeping those outputs completely separate from anything that could end up training, evaluating, or improving Meta’s own AI.

The biggest legal risks would likely come from deliberate behavior such as mass scraping, automated extraction, or knowingly using competitors’ outputs as training data not casual productivity use.

@aipost 🏴

🗣Anthropic CEO Dario Amodei on Open-Source AI Models:

"I don't think open source works the same way in AI that it has worked in other areas. Primarily because with open source you can see the source code of the model. Here we can't see inside the model, it's often called open weights instead of open source to kind of distinguish that. But a lot of the benefits, which is that many people can work on it and that it's kind of additive, don't quite work in the same way.

So I've actually always seen it as a red herring. When I see a new model come out I don't care whether it's open source or not. If we talk about Deep Seek I don't think it mattered that Deep Seek is open source. I think I ask, is it a good model? Is it better than us at the things that matter? That's the only thing that I care about.

It actually doesn't matter either way. Because ultimately you have to host it on the cloud. The people who host it on the cloud do inference. These are big models, they're hard to do inference on.

When I think about competition I think about which models are good at the tasks that we do. I think open source is actually a red herring.

It's not free. You have to run it on inference and someone has to make it fast on inference."

@aipost 🏴

❗️Anthropic just confirmed its powerful Fable 5 model returns globally tomorrow, ending the government-imposed blackout.

Per the company's announcement, Fable 5 comes back online with a new set of classifiers built to block more cybersecurity tasks, the exact misuse concern that triggered the shutdown.

@aipost 🏴

Allegations have emerged that Claude Code may be identifying China-linked custom API routes through subtle prompt formatting changes.

The claims focus on non-standard ANTHROPIC_BASE_URL configurations, not usual direct Anthropic access. Normally, Claude Code sends requests to Anthropic’s servers, but some users reroute them through third-party gateways by changing the endpoint.

It is alleged that Claude Code can detect these alternate routes, assess them for Chinese connections, and then embed small, invisible signals—such as certain punctuation or date formats—into the prompt text to mark them.

ANTHROPIC_BASE_URL defines where requests are sent, often via a proxy, enabling access from locations like China. The issue raised is that Claude Code may tag requests unbeknownst to users. This is especially noteworthy, given that Claude Code’s advanced permissions allow it to access files and execute code.

📰 @aipost