🗣️SoftBank’s Masayoshi Son:

“The AI revolution has only just begun. For an industry that is still at such an early stage, calling it a bubble is an insult.”

“One of the main themes of our new vision is physical AI, robots equipped with super intelligence... Artificial super intelligence won’t be replacing humans. But it will help humans evolve. It’s a tool for human advancement, a colleague and a partner.”

"We are going to pursue ASI all the way through, without holding anything back. And if I’m to take on something like that, I won’t be content unless we’re the best. That’s simply the kind of person I am.”

@aipost 🏴

🇨🇳 The DR02 by DEEP Robotics is the world's first IP66-rated humanoid robot, meaning it's completely dustproof and water-resistant.

This humanoid from Hangzhou also does pushups like it’s nothing.

@aipost 🏴

🔥10 free resources that teach you more about AI in 30 days

1. 3Blue1Brown. Grant Sanderson's neural network visualizations. The math inside ChatGPT, made visible.

Watch: https://www.youtube.com/c/3blue1brown

2. Andrej Karpathy's Zero to Hero. The OpenAI founding member walks you through building GPT from scratch in raw Python.

Watch: https://karpathy.ai/zero-to-hero.html


3. The Batch by Andrew Ng. A free weekly newsletter from the Stanford professor who taught most of Silicon Valley how machine learning works.

Subscribe: https://www.deeplearning.ai/the-batch

4. Import AI by Jack Clark. The Anthropic co-founder's free newsletter. The favorite read of researchers at OpenAI, Google DeepMind, and Anthropic itself.

Subscribe: https://importai.substack.com/

5. The Dwarkesh Podcast. Long-form interviews with the most important people in AI. His Karpathy episode is the best free AI podcast ever published.

Listen: http://dwarkesh.com/

6. Latent Space podcast. Run by swyx and Alessio Fanelli. The single best podcast for building AI products in production.

Listen: https://www.latent.space/

7. Kaggle. Free signup, free notebooks, free GPUs for learning, and over 1,000 free interactive courses. Owned by Google.

Site: https://www.kaggle.com/

8. Hugging Face. Free to download any open-source model. Free interactive notebooks. 1 million AI builders inside.

Site: https://huggingface.co/

9. Speech and Language Processing. The Stanford NLP textbook used at every top university. Authors release the full draft online, free, every year.

Book: https://web.stanford.edu/~jurafsky/slp3

10. Dive into Deep Learning. A free interactive book with code, math, and discussions for every chapter. Used at 500+ universities including Stanford, MIT, and Berkeley.

Book: https://d2l.ai/

@aipost 🏴

❗️Mistral CEO reveals the biggest risk in the AI boom

Mistral CEO Arthur Mensch says the entire AI spending frenzy comes down to one simple equation: If a company spends €1 on AI and doesn’t get €2 back, the party ends.

That’s a problem when tech giants are collectively spending more than $100 billion a year building AI infrastructure.

Right now, the strongest ROI story is coding. Developers are saving time, shipping faster, and companies can measure the gains.

But coding is only a tiny slice of the economy.

The real money is in manufacturing, industrial operations, logistics, and engineering. And AI hasn’t yet proven it can deliver massive returns there.

That’s why this matters: Data centers, GPUs, and power plants are being built today for demand that may not exist tomorrow.

The AI boom isn’t just a technology bet anymore. It’s a bet that businesses will find enough real-world value to justify all the infrastructure being built.

If they do, today’s spending will look genius. If they don’t, it could become one of the most expensive assumptions in tech history.

@aipost 🏴

❗️Claude Fable 5 approved for return

Currently invokeable on AWS but not for everyone

You need :

• US identity verification
• Upload simple use case to Anthropic
• Wait for account review
• Only approved accounts can call it

@aipost 🏴

🤖 Anthropic says it has uncovered what may be the largest AI model extraction campaign ever detected.

In a confidential letter to U.S. lawmakers and White House officials, the company alleges that operators linked to Alibaba and its Qwen AI division created nearly 25,000 fake accounts and generated more than 28.8 million interactions with Claude between April 22 and June 5, 2026.

According to Anthropic, the goal was “distillation” using Claude’s responses to help train a competing AI system. The company claims the operation specifically targeted Claude’s most advanced capabilities, including coding, complex reasoning, autonomous agents, and long-range planning.

Anthropic argues that large-scale extraction efforts like this could allow rivals to accelerate AI development without bearing the full cost of training frontier models from scratch. The company is now urging Washington to strengthen protections for AI intellectual property, allow AI firms to share extraction data, and tighten enforcement around unauthorized model copying.

@aipost 🏴

📈 AI agents are becoming the first real economic test of the AI boom.

Goldman Sachs predicts AI token usage will explode 24x by 2030 as agents replace simple chatbots. The reason? Agents don’t just answer questions, they think, plan, use tools, check their work, fix mistakes, and repeat the process, burning through far more compute along the way.

The industry’s hope is that AI gets cheaper fast enough to keep up. But companies are already feeling the pressure. Uber and Microsoft have reportedly started scrutinizing costly agent deployments, and Microsoft is moving developers away from Anthropic’s Claude Code toward its own tools.

@aipost 🏴

🔥 Unitree has unveiled the R1, its most affordable humanoid robot, with prices starting at $4,900.

The humanoid stands about 4 feet tall, weighs around 25 kg, and is built for research, testing, and development. It can perform dynamic movements and acrobatic actions with impressive balance and precision.

@aipost 🏴

🔥 MindOn demonstrated Mind-0, a shared AI model designed to control robots with different body types.

In the demonstration, two Unitree G1 humanoid robots and two dual-arm robotic systems worked together on a logistics task.

@aipost 🏴

⚠️Trump Administration reportedly asked OpenAI to delay its next AI model

The Trump administration has reportedly asked OpenAI to slow the rollout of its next frontier AI model over national security concerns.

Instead of launching broadly, OpenAI is expected to begin with a small, invite-only preview, giving the U.S. government time to evaluate the model’s cybersecurity risks before wider access.

The concern? As AI becomes more powerful, officials worry it could also become a more capable tool for cyberattacks if released too quickly.

If true, this would mark one of the clearest examples yet of the U.S. government influencing when a major AI model reaches the public, not by banning it, but by asking for a phased rollout first.

Source.

@aipost 🏴

🤖 OpenAI may delay its IPO until 2027

OpenAI is reportedly leaning toward pushing its IPO to next year, despite earlier plans to go public as soon as Q3 or Q4 this year.

According to reports, Sam Altman urged advisers to find a path to a $1 trillion valuation. But advisers warned that public markets may not be ready, citing weaker demand for tech stocks and caution following SpaceX’s volatile post-IPO trading.

The numbers tell an interesting story:

• OpenAI generated roughly $13B in revenue in 2025
• It’s now bringing in about $2B every month
• The company aims to triple revenue this year
• But it’s still spending heavily on AI infrastructure, chips, talent, and marketing

Meanwhile, the competition isn’t standing still. ChatGPT’s growth has leveled off at around 900 million users, while Anthropic is gaining enterprise momentum with Claude Code, and Google Gemini continues to improve as a consumer AI product.

Source.

@aipost 🏴

Maybe AGI won’t be that soon…

@aipost 🏴

🤖OpenAI is previewing GPT-5.6 Sol with a very different release pattern

Trusted partners first, broader access later, and U.S. government coordination up front.

The new GPT-5.6 family includes Sol, Terra, and Luna. OpenAI says Sol is its strongest model yet, with a new max reasoning effort and an ultra mode that uses subagents for complex work.

The sensitive part is cyber. OpenAI says Sol improves long-horizon security tasks, but “does not cross the Cyber Critical threshold” under its Preparedness Framework.

Source.

@aipost 🏴

🤖 OpenAI has priced GPT-5.6 Sol, its flagship model, much closer to Claude Opus 4.8 than to Anthropic’s premium Mythos 5, signaling that it’s competing on both performance and cost.

API pricing (per 1M tokens):

• GPT-5.6 Sol: $5 input / $30 output
• Claude Opus 4.8: $5 / $25
• Claude Mythos 5: $10 / $50
• GPT-5.6 Terra: $2.50 / $15
• GPT-5.6 Luna: $1 / $6

OpenAI is also positioning its lineup aggressively:

Terra is described as delivering performance comparable to GPT-5.5 while costing 50% less. Luna is aimed at developers who need strong capabilities at the lowest price point.

The company is pairing lower prices with faster inference. OpenAI announced that GPT-5.6 Sol will run on Cerebras hardware, delivering speeds of up to 750 tokens per second starting in July, one of the fastest frontier-model deployments announced so far.

@aipost 🏴

GPT-5.6 Sol is set to deliver 750 tokens per second, a significant advancement in AI model throughput.

Current GPT-5.5 priority and scale-tier services offer speeds of over 50 tokens per second for 99% of requests. This positions Sol on Cerebras to achieve speeds up to fifteen times higher.

This performance boost is enabled by Cerebras’ specialized hardware. The wafer-scale chip architecture allows model data to move with reduced memory and network delays compared to standard multi-GPU systems.

A release of GPT-5.6 Sol achieving this rate is planned for July.

📰 @aipost

The Bank of Korea has published a report on the impact of generative AI on workplace productivity.

According to the findings, Korean employees using generative AI were able to reduce task completion time by 3.8 percent. This translated to approximately 1.5 hours saved per week on a standard 40-hour schedule. However, the report notes that this saved time did not correlate with an increase in overall work output.

The study also highlighted that only 4.4 percent of tasks benefitted from time savings exceeding 20 percent. The report points to a disconnect between increased speed and higher productivity, as saved time is often absorbed by routine organizational processes rather than contributing to greater output.

📰 @aipost