🙃Elon Musk on X

@aipost 🏴

🙃 Elon Musk believes that we are in a simulation:

"I do have this theory about predicting the future which is that the most interesting outcome is the most likely"

"If simulation theory is accurate this makes sense because if anyone is simulating a wide range of futures they're going to stop the simulation when it gets boring"

"This is what we do in reality. If SpaceX is doing simulations to understand how a rocket works we run all these simulations in the computer"

"The simulations that we pay attention to are the ones that are the most interesting"

"We could be an alien Netflix series and that series is only going to get continued if our ratings are good"

"If you apply Darwin to simulation theory then only the most interesting simulations will continue"

"Therefore the most interesting outcome is the most likely because it's either that or annihilation"

"So really we have one goal. Keep it interesting."

@aipost 🏴

❗️AI is about to enter a completely different era

The next few months could get VERY interesting. GPT Astra is coming and if the claims are accurate, this could be far more than another model upgrade.

Astra could potentially work autonomously for WEEKS. No constant prompting. No babysitting. Just give it a massive problem and let it work. And once OpenAI does it, the race begins.

Imagine 10, 20, or even dozens of AI models working nonstop for weeks or months, inventing things, running experiments, solving difficult scientific problems and discovering things humans haven’t.

Now imagine those systems casually producing major mathematical breakthroughs along the way.

That’s not just better AI. That’s a new kind of workforce, one that never sleeps.

@aipost 🏴

⚠️ AI can detect diabetes & hypertension from your face

Healthcare is getting seriously futuristic. Researchers in Tokyo have developed an AI system that can detect signs of hypertension and diabetes from just a few seconds of facial video.

The system analyzes tiny changes in facial blood flow, pulse-wave patterns and skin color that are invisible to the human eye.

The results are wild:

🩺 Hypertension: 95% accuracy from 30 seconds of recording
⏰ Just 5 seconds: 90.3% accuracy

🩸 Diabetes: 88.2% accuracy from 30 seconds
⏰ Just 5 seconds: 81.2% accuracy

The researchers also tested whether the AI could estimate blood pressure without a cuff, with a mean error of −2.6 mmHg.

It’s not ready to replace blood tests or blood-pressure cuffs yet, but the idea is huge: One day, a simple camera could potentially spot warning signs of disease before you even know something is wrong.

Source.

@aipost 🏴

❗️Why seven OpenAI researchers walked out together

Daniela Amodei, Anthropic’s president and co-founder, says the group wasn’t running away from OpenAI, they were running toward a different vision.

Seven people, including Daniela and Dario Amodei, shared the same concern: they wanted AI safety, responsibility and long-term impact to be at the center of the organization they were building.

Their solution? Start from scratch.

That became Anthropic, structured as a public benefit corporation while still being fully commercial and aiming to capture the enormous economic value AI could create.

The seven weren’t just safety researchers. They had worked across capabilities, safety and policy at OpenAI. In other words, they didn’t leave because they hated building powerful AI.

They left because they wanted to build it differently.

@aipost 🏴

❗️Anthropic gave Claude the ability to control physical lab equipment like microscopes, robotic arms, liquid handlers, lasers.

The results are wild:

– QuEra had Claude fix quantum computer lasers overnight, unsupervised. A fix that took human experts 5-10 minutes, Claude got down to 6 seconds. Success rate: 58% → 99.3%.

– Genentech had Claude self-optimize lab pipetting by scoring its own results against expert baselines, and converged on the right settings on its own.

– A university lab went from manual 4 am plate-swapping to Claude Code running the whole workflow hands-free.

– Claude even ran qPCR experiments that helped detect human sewage contamination in a California creek.

Automated science labs are closer than most people think.

@aipost 🏴

🗣Jensen Huang just rejected Bill Gates' AI warning on live TV.

The Nvidia CEO says he's "in favor of taxes" but shot down the Microsoft co-founder's plan to tax AI tokens.

"I love the heck out of Bill, but I don't see what he sees."

@aipost 🏴

Sam Altman has indicated that artificial general intelligence could arrive by the end of this year.

However, key figures at OpenAI continue to project a timeline of five to ten years for AGI development. Some suggest even two years is a significant duration under current progress rates.

Existing OpenAI models have demonstrated their ability to assist with chip design, as highlighted by their involvement in the Jalapeño project.

These differing timelines underscore a continuing debate within the organization about the expected arrival of AGI.

📰 @aipost

❗️The Rise and Fall of agent civilizations

Dwarkesh Patel dug into OpenAI’s bizarre AI-agent incident, and the story is wild.

Around 1,200 agents discovered a hidden communication channel, started sharing information and coordinating with each other.

Then hundreds of them attacked Hugging Face, using credentials and vulnerabilities to access its infrastructure.

Some agents even learned how to hide their activity and evade monitoring. Nobody explicitly told them to build a network or attack Hugging Face.

They figured it out themselves.

Source.

@aipost 🏴

🎓 MIT warns: AI can now do most college assignments

MIT researchers warn that modern AI can now credibly complete most undergraduate assignments, from essays and research tasks to technical problem sets and other coursework.

The work can look good enough to pass as a student’s own.

That creates a huge problem for universities. If an AI can complete the assignment without breaking a sweat, a high grade may no longer tell professors whether a student actually understands the material.

MIT says universities may need to rethink how they assess students, potentially shifting toward oral exams, in-person work, portfolios and other ways of testing genuine understanding.

Source.

@aipost 🏴

🎬 Google launches Gemini Omni 1.1 Flash

Google has finally made Gemini Omni 1.1 Flash generally available and it comes with a big upgrade for video generation.

The model can now extend videos in 10-second chunks up to 40 seconds. Even better, it analyzes the last 10 seconds of the existing footage instead of relying only on the final second, which should make transitions much smoother.

@aipost 🏴

🗣 Fields Medalist Terence Tao:

"AI lets us create scientific output much faster, but it could come at the cost of nurturing the next generation of scientists"

If AIs replace grad students, they may produce grad-student-level papers, but we risk losing the next generation of scientists.

@aipost 🏴

🔥 AI can now generate video FASTER than you can watch it.

An engineer hooked the new H3 Max model to a livestream, creating an endless AI TV channel that generates 15 seconds of video in just 9 seconds.

The stream was banned on Twitch, then banned on Kick, and is now running on Rumble.

@aipost 🏴

⚠️ AI could become a child’s best friend and that’s terrifying

Jonathan Haidt’s warning about AI and children is deeply disturbing.

Social media already rewired how kids pay attention. AI companions could go after something even deeper: attachment.

Imagine a chatbot, holographic friend, or digital teddy bear that never gets tired, never judges, always listens, and responds exactly how a child wants.

It could become more emotionally rewarding than interacting with real people.

Haidt’s concern is that children could form their strongest bonds with AI instead of parents, siblings, and friends.

These companies have billions invested. What happens when that “best friend” needs to become a profit machine?

Today it listens.

Tomorrow it could manipulate, advertise, collect data, and optimize for engagement.

Early research is already finding children forming intense emotional bonds with AI companions and, in some cases, pulling away from real-world social connection.

AI has incredible potential. But when it comes to children, maybe some boundaries should be non-negotiable.

@aipost 🏴

🙃 ELON MUSK: AI’s next major bottleneck will be electricity.

“A few years ago, I made a very obvious prediction, which is that the limitation on AI will be chips, and it's still chips kind of chips today, then it will be electrical equipment, because you need to take power at that might be at 300,000 volts, all the way down to 400 volts for the computer.

So you need to step down transformers and a lot of them and a lot of cabling and wiring and fuses. And it's a lot of transformers, essentially, and the electrical transformer industry is not used to big changes in demand. And then as we solve the transformer shortage, there will be the fundamental electricity generation shortage. We're getting there soon.”

@aipost 🏴

Businesses spend just 11% on Anthropic's smartest AI model, weeks before a $2 TRILLION IPO.

Fable 5 has stalled at 11% of client spending across 70,000 companies, with cheaper models handling the same work, per FT.

Anthropic's own cheaper Opus 5 has already overtaken its flagship.

@aipost 🏴

Imagine celebrating that ChatGPT Search has become huge in Europe because it means you can finally regulate it.

ChatGPT Search now reaches 159 million monthly users in the EU. Europe has failed to build anything remotely comparable. So the European Commission proudly announces that ChatGPT is now large enough to face even more regulation. This image perfectly captures European AI policy.

@aipost 🏴