🥺 ChatGPT faces another major lawsuit over medical advice

OpenAI and CEO Sam Altman are being sued after a U.S. pastor claimed ChatGPT downplayed symptoms of a life-threatening pulmonary embolism and discouraged him from seeking medical care.

The lawsuit argues ChatGPT crossed into practicing medicine without a license and calls for its health features to be suspended until they’re independently verified as safe.

OpenAI says ChatGPT isn’t meant to diagnose or treat medical conditions, but the case adds to growing legal scrutiny over the chatbot’s health advice.

Source.

@aipost 🏴

🗣 Sam Altman reveals he flew to 28 countries in 35 days after GPT-4 launched because world leaders were asking whether to shut AI down, and ended the trip waking up convinced he was back in his childhood bed.

“Three years ago we had just launched GPT-4. The world was melting down. Everybody was freaking out.”

“World leaders were like, do we need to take control and shut things down? Is this thing waking up?”

“I think we did 28 countries, 35 days. I lived on an airplane.”

“By the end of it, the last few days, I started having this vision, feeling, whatever, of being in my childhood bed. I would sort of vaguely wake up and I would think I was in my childhood bed in my childhood room. And I was like, okay, this is some deep ‘it’s time to go home’ thing.”

@aipost 🏴

❗️Satya Nadella just picked a side in AI’s biggest debate.

Microsoft says open-weight AI models are critical to keeping America competitive.

Here’s why:

If only a handful of companies own the most powerful AI models, they also control the prices, access, infrastructure, and ultimately the future of AI.

Open-weight models flip that equation.

Developers can download them, customize them, run them on their own hardware, and build products without relying on a single AI provider.

Of course, that comes with security risks.

But Microsoft argues that putting the future of AI in the hands of just a few closed labs could be an even greater danger.

Source.

@aipost 🏴

🇨🇳🇺🇸Beijing is hitting back at Washington’s threats to investigate and punish Chinese AI firms...

China’s Commerce Ministry labeled the U.S. approach as “AI hegemonism” and warned it will take all necessary countermeasures if restrictions are imposed.

The dispute centers on allegations that companies like Moonshot AI improperly extracted advanced American AI models to power their own systems.

Chinese authorities and the companies reject these accusations, insisting their advances are the result of independent research and innovation.

@aipost 🏴

🧠 Demis Hassabis says DeepMind had no choice but to sell.

Before AI became a trillion-dollar race, DeepMind was struggling to survive. Demis Hassabis says he realized staying independent would require $7 billion in funding.

The problem? They could barely raise $10 million at a time. After DeepMind’s breakthrough Atari demos, Big Tech started aggressively recruiting its researchers.

They had no product, no revenue, and were focused entirely on building AGI. Hassabis says he paid himself nothing, while engineers earning around $100,000 a year were suddenly receiving offers of $10 million… or even $50 million annually.

At that point, competing as an independent company became almost impossible. That’s why DeepMind joined Google in 2014. Looking back, it wasn’t just a business decision.

It was a reminder that building frontier AI has always been as much about capital and talent as it is about breakthrough research.

@aipost 🏴

A new milestone has been reached in the open source AI space. Open models from China are now outperforming recently released competitors, such as Opus 4.8, just two months after their debut.

Kimi K3, now holding the top spot among open weights models at 57 in the Artificial Analysis Intelligence Index, has made its weights publicly available. The 2.6T parameter model, released under the 'Kimi K3 License,' is labeled as 'Commercial Use Restricted.'

This licensing means businesses offering model-as-a-service solutions with over $20 million in annual revenue must seek additional agreements. Products exceeding 100 million monthly active users or generating $20 million in monthly revenue must clearly display "Kimi K3" in their user interface.

A technical report outlining the model's architecture and training methods has also been released.

📰 @aipost

OpenAI research shows that 43.5% of occupation-specific ChatGPT messages from U.S. business users related to tasks typically handled by a different profession. This indicates users are leveraging the model to address specialist work outside their primary field.

Financial calculations and computer troubleshooting were among the top three most common tasks handled beyond users’ professions in all groups. Marketing-related tasks also appeared frequently across various occupations.

The data captured messages within ChatGPT Business, not finished work, so it does not measure quality, productivity, or impact on jobs.

Researchers note that, especially in small businesses, employees often solve problems outside their usual roles. AI is emerging as a versatile support tool across different functions.

📰 @aipost

🤖 Anthropic explains why it still won’t release frontier open-weight AI models

While much of the AI industry is rallying behind open-weight models, Anthropic is holding its ground.

The company says today’s most powerful AI systems are becoming capable enough that releasing their weights could make it much easier for bad actors to build dangerous cyber, biological, or autonomous attack tools. Instead of full public releases, Anthropic argues for controlled access through APIs, where safety measures can be monitored and updated.

At the same time, Anthropic says it isn’t against open-weight AI. It supports open research and believes open-weight models have an important role in innovation. The company says it may release open-weight models in the future but only when they can be deployed with what it considers adequate safety protections.

@aipost 🏴

⚡️Microsoft’s new AI cybersecurity system just hit a 96% score on CyberGym.

That’s 12 points higher than Anthropic’s Mythos while costing roughly half as much as Microsoft’s previous system.

The secret? Microsoft didn’t rely on one giant AI model.

Instead:

• MAI-Cyber-1-Flash handles about 90% of security tasks.
• Only the toughest 10% are escalated to GPT-5.4.

Behind the scenes, a system called MDASH orchestrates 100+ specialized AI agents that:

• Search code for vulnerabilities
• Investigate and debate whether issues are real
• Remove duplicate findings
• Test whether bugs are actually exploitable
• Gather evidence before reporting results

@aipost 🏴

🚀 Elon Musk says Grok 4.6 is arriving on August 7. And that’s just the beginning.

According to Musk, Grok 4.6 launches next week with 1.5 trillion parameters. Just a few weeks later, Grok 4.7 is expected to follow with an even larger 2.1 trillion-parameter model. Reports about the timeline have been circulating over the past few days, although some technical details remain unverified.

But that’s not the only rumor heating up.

There’s growing speculation that Composer 3 is also close to launch. The coding model is said to have been trained from scratch on the massive Colossus 2 cluster after SpaceX’s reported acquisition of Cursor. If the rumors are accurate, Composer 3 could deliver Claude Opus 5 and GPT-5.6 Sol-level coding performance at a fraction of the cost.

@aipost 🏴