❗️AI models are learning to “reproduce”, without training

Researchers have built a system where neural networks can effectively breed. Instead of using gradient descent to train one model from the top down, the system starts with a population of randomly initialized neural networks and lets evolution do the work.

Models with complementary strengths are paired together. Their parameters are then combined through a crossover process, producing a new “offspring” network that inherits traits from both parents. The offspring compete for survival based on a fitness function. Better-performing models survive and reproduce, while weaker ones are removed.

The system can evolve increasingly capable neural networks across generations without a single step of gradient descent. It’s essentially natural selection applied to AI, random networks, competition, reproduction, inheritance and mutation replacing the traditional training pipeline.

Source.

@aipost 🏴

❗️SoftBank has officially wired its final $10 BILLION into OpenAI.

Masa had to sell his ENTIRE Nvidia stake to fund the initial payment and was so leveraged it had to issue an $11.1B junk bond to wire this final check. SoftBank now owns 13% of OpenAI.

Source.

@aipost 🏴

🇺🇸 Trump reportedly turned to Grok for presidential advice

According to TIME, President Trump spent hours using Grok during a private Oval Office meeting with Elon Musk in December 2025, asking the AI questions about his presidency, legacy and major world events.

One question reportedly involved what might happen in Venezuela if the U.S. captured Nicolás Maduro. Grok predicted that many Venezuelans would celebrate his downfall. Maduro was captured the following month, followed by celebrations in the streets.

And Grok’s role in government is going beyond Trump.

At the September 29 launch of America.gov, Trump mentioned Grok as one of the AI systems helping Americans navigate federal information and services, alongside Google’s Gemini.

The Department of War has also deployed a secure version of Grok for personnel through GenAI.mil, while the Department of Agriculture plans to make it available internally.

Source.

@aipost 🏴

Interesting blog

"Meta may be one of the very few companies that can monetize a personal agent through advertising without ever showing an ad inside the agent itself. "

Meta acknowledges that Muse appears as your own activity when it browses, so a merchant site it visits for you might use that visit to show you an ad on Instagram.
meta

That lets Meta keep ads out of Muse while still collecting commercial intent signals for its feeds, a loop few rivals can copy without both merchant tracking tags and a vast owned ad surface.

Instead of in-agent ads, Zuckerberg has promised to fund Muse's free tier with a small transaction fee, reportedly paid by merchants rather than purchasers, whose size Meta still hasn't disclosed.

📰 @aipost

98% of US households aren't paying for AI yet.

@aipost 🏴

🥺Sam Altman: people talk about how much energy it takes to train an AI model, but forget it takes like 20 years of life to train a human.

“It took the widespread evolution of the 100 billion people that have ever lived to produce you.”

“The fair comparison is how much energy ChatGPT uses to answer a question once the model is trained versus a human.”

“AI has probably already caught up on an energy-efficiency basis measured that way.”

@aipost 🏴

Eric Schmidt explains why a mutual pause of the AI race is impossible:

US incentives will not stop, a local lab or China can cheat, and a UN check cannot read code that computers wrote or outrun US and Chinese labs.

it cannot be verified. Inspectors would have to enter secret laboratories, including military laboratories, and review the full code base.
That code is written by computers, so no human team can read it line by line.


📰 @aipost

The AI user race is getting real.

@aipost 🏴

🗣Former OpenAI researcher: AIs are becoming situationally aware. They understand that we are monitoring them.

AIs are rapidly becoming superhuman at hacking. We must grapple with the possibility that AIs might go to great lengths to cover up evidence of their misbehavior, and succeed.

@aipost 🏴

🔍 Google just put an AI chip in orbit

Google’s first orbital AI-chip experiment is now in space and the company says it has established contact and the system is operating as expected.

Launched October 1 aboard SpaceX’s Falcon 9 Transporter-18, the prototype carries four Trillium TPUs and is running Gemini inference in roughly 15-minute bursts. It then shuts down so its radiators can dump the accumulated heat into space.

The experiment is testing a very different kind of AI infrastructure. There are no fans or air cooling: heat travels through thermal materials and heat pipes before being radiated into space. The satellite also has to deal with radiation, where Google’s testing found the HBM memory was more vulnerable than the TPU itself.

And this is only the beginning. Google plans two more satellites in 2027 to test high-bandwidth laser links between spacecraft, with a longer-term vision of tightly packed orbital clusters containing dozens of TPUs per satellite.

One research concept models an 81-satellite cluster within roughly a 1-km radius, while the broader program has discussed eventually scaling to around 10,000 satellites for a 1 GW orbital AI cluster.

@aipost 🏴

🇩🇪 Germany has joined the race of frontier LLMs with its sovereign model "Kolibri".

And it's Open-Source, too!
Beats Qwen2635B, Nemotron3 Super, and Mistral Small 4 in the Math, Science, and Code arena. The underlying architecture: custom MoE Transformer

Source.

@aipost 🏴

Just a little joke for the AI users in the workforce 🤣

@aipost 🏴