AI Post — Artificial Intelligence

🗣Sam Altman says the public has mischaracterized him as AI's villain for the same reason he once mischaracterized Mark Zuckerberg: When a new technology brings fast change, people project their fear onto the person who seems to hold the most power. Admits that in the early days of Meta he held similar belief towards Mark Zuckerberg, at a time when he did not know Zuckerberg personally. @aipost 🏴

❗️Hacker may have exposed himself by asking AI to write his résumé A hacker suspected of targeting South Korean banks may have accidentally revealed his identity after asking Claude Code to write a résumé about his own hacking activities. CrowdStrike traced the attacks to infrastructure running an AI hacking tool called ARTEX. Investigators later found publicly exposed folders containing the attacker’s saved Claude Code conversations. In one chat, he asked Claude to create a cybersecurity résumé describing his ARTEX work and provided personal details including a phone number, Telegram handle, claimed age of 26, university and location in China. The same Telegram handle appeared in other hacking-related conversations. He had also asked Claude where stolen Korean data is typically sold. His identity has not been independently confirmed but asking an AI to turn your own cybercrime history into a résumé is certainly an unusual way to leave a trail. @aipost 🏴

❗️Demis Hassabis is in talks to raise new funding at a $50 BILLION valuation for Isomorphic Labs. Mission: Solve all disease @aipost 🏴

⚠️ For the first time in either presidency, Trump will present the National Medal of Science. Elon Musk, Jensen Huang, Lisa Su and Sergey Brin will receive it Thursday at the White House's "Science: A New Golden Age" summit. Satya Nadella and Michael Dell will receive the separate National Medal of Technology and Innovation. @aipost 🏴

⚡️ AI is rewriting America’s power demand The U.S. is heading for record electricity consumption in both 2026 and 2027 and AI data centers are becoming a major driver. Electricity use is projected to rise from 4.195 trillion kWh in 2025 to 4.288 trillion in 2026 and 4.356 trillion in 2027. That’s an additional 161 billion kWh of demand in just two years, as data centers built to train and run increasingly powerful AI systems consume enormous amounts of electricity. Source. @aipost 🏴

❗️AI hits biology’s data bottleneck AI biology has reached a problem GPUs alone can’t solve: there simply isn’t enough training data. A $1.8 billion effort backed by Biohub, the U.S. Department of Energy, NIH, Meta, Google DeepMind and Isomorphic Labs aims to build a “universal virtual cell” that could predict how cells respond to interventions and let researchers test biological experiments digitally before doing them in the lab. Today’s datasets contain hundreds of millions of cells. Future models may need billions and eventually trillions. The goal is to compress work that normally takes decades into about five years, with the first dataset expected in roughly one year. There’s also an unusual access model: commercial funders get one year of exclusive access to the data they help generate before it becomes public. Government-funded data has no such embargo. @aipost 🏴

🤖 ChatGPT now accepts audio files! Paid users can upload a recording and ask for a transcript, a summary, or answers about what was said. It supports formats like MP3, WAV and M4A, with a 512 MB limit per file. Free users don’t have access yet. @aipost 🏴

❗️The White House is preparing to unveil OVER $1 BILLION in industry commitments for its AI-driven Genesis Mission, according to Axios. The mission aims to harness advanced AI to accelerate American scientific breakthroughs. AMD, OpenAI and Anthropic are among the companies involved. President Trump is expected to attend Thursday's AI science summit as Washington pushes to accelerate scientific discoveries amid federal research funding cuts. @aipost 🏴

🤖 Anthropic launches Claude Haiku 5.5 Anthropic has released Claude Haiku 5.5, calling it the fastest, cheapest and most capable small model it has built. Haiku 5.5 costs around 75% less to run than Haiku 4.5 while delivering major improvements in coding, computer use and knowledge work. It’s designed for high-volume tasks like summaries, classification, customer support and browser use, and can work as a subagent alongside Opus 5.5 and Sonnet 5.5. It’s also the first Haiku with adjustable effort, letting developers choose between lower cost and higher intelligence. Anthropic says around 90% of previous Haiku requests were under 100,000 tokens, where the new model offers particularly strong value. Anthropic is also cutting Sonnet 5.5 cache-read prices by 50%, to just $0.10 per million tokens. Source. @aipost 🏴

Parents are using AI to create school portraits of their children instead of paying for photographers. Others are using it to remove watermarks from professional photos, risking fines up to $150,000. @aipost 🏴

Which tasks do people actually want AI agents to do? 🤔 @aipost 🏴

Once you see it, you can’t unsee it. @aipost 🏴

🥺 Sam Altman: The world isn’t ready for the humanoid robot moment.

“I don’t think that’s very far away.” “There’s gonna be a visceral moment of, ‘oh man, this is gonna do a lot of things people used to do.’” “They’ll obviously do a lot of jobs.” “The first day you’re walking down the street and seven robots walk past you doing things, it’s gonna feel very sci-fi.”

@aipost 🏴

❗️Mistral drops a 1T-parameter “Le Chonk” Mistral AI has unveiled Mistral Large 4, internally nicknamed “Le Chonk”, a 1 trillion-parameter natively multimodal model with just 49B parameters active at a time. Mistral says it is currently the best open-weights model from the US or Europe across aggregated benchmarks, while reaching state-of-the-art performance in areas including cybersecurity, manufacturing and finance. It also reportedly surpasses closed frontier models on visual grounding. The model was developed end-to-end in Europe and runs on Mistral’s own European cloud infrastructure. API access is available today, while the open-weights release is planned for the end of October. Mistral is also working privately with cybersecurity partners on the model. Source. @aipost 🏴

❗️Microsoft and META are aggressively cutting employee use of Claude ahead of Anthropic's IPO Microsoft has cut internal claude spend by more than 33%, nuked per-employee token budget from $100k/month to $10k/month, and forced Copilot to auto-route to cheaper models META used Claude code to build Muse, then cut active users from 60,000 to 30,000 (50% decline) after launch, and replaced it with Muse Code Palantir and Nvidia are also scaling back claude over soaring prices and data privacy fears. @aipost 🏴

Not one AI giant says yes when asked if it's insured against catastrophic risk. OpenAI, Anthropic, Google and Meta representatives went silent when NYC Council Speaker Julie Menin asked them to raise their hands if they had that insurance. @aipost 🏴

📢 Your voice could become a trademark Law firms may soon be selling “identity portfolios” that protect a person’s face, voice and signature phrases, much like companies protect their brands. Italy’s Prime Minister Giorgia Meloni has just taken a step in that direction. She filed an application with the EU Intellectual Property Office to register her voice as a sound mark, using a four-second recording of herself saying “Io sono Giorgia Meloni.” The application is still under examination. The goal is to give her another legal tool against AI-generated voice clones and deepfakes. Meloni has already been targeted by manipulated content, including fake material and misleading uses of her voice. If this approach catches on, protecting your identity could start looking a lot like protecting a corporate brand. Your face. Your voice. Your catchphrases. All potentially becoming intellectual property in the fight against AI impersonation. Source. @aipost 🏴

🤖 ChatGPT may start changing how it writes in the EU OpenAI is reportedly preparing invisible text watermarks for ChatGPT and Codex outputs in the EU, with the pattern embedded directly into the wording. The reason is the EU AI Act, which is pushing AI-generated content toward being identifiable. OpenAI says its testing found no meaningful impact on quality. But there’s a bigger issue: detection requirements could start influencing how an AI actually phrases your text. And the watermark isn’t particularly robust. In one test using 400-token passages, changing just 25% of the words to synonyms reportedly reduced detection from around 92% to 17%. There’s another problem:m, it can’t tell how much of the text came from a human. You could write something yourself, ask ChatGPT to edit it, and the final version could still carry the watermark. Source. @aipost 🏴

❗️AI could make you guilty on video Sundar Pichai warned about a disturbing possibility: someone could use AI to create a convincing video of you saying something you never said. Your face. Your voice. A fake confession. Then they send it to your boss, your family or your friends and demand money to make it disappear. We’ve already seen scammers use fake versions of famous people to push crypto scams. Now imagine that same technology being used against an ordinary person. You say the video is fake. They say you’re lying. And everyone watching has “proof.” The terrifying part isn’t that AI wants to frame you. It’s that someone else can use your identity as a weapon. If seeing is no longer believing, how do you prove you didn’t do something that never happened? @aipost 🏴

💰 The top 1% of AI spenders now outspend the bottom 50% combined They spend an average of $903/month, while the median customer spends just $25. @aipost 🏴