971972
🤖 The #1 AI news source! We cover the latest artificial intelligence breakthroughs and emerging trends. Manager: @rational
Anthropic launches a feature that lets users teach Claude skills by recording their screen.
@aipost 🏴
❗️The U.S. just accused Moonshot AI of secretly using Anthropic’s Fable model to build Kimi K3.
Michael Kratsios, Director of Technology at the U.S. Department of Commerce and Deputy Assistant to the President for Science and Technology, says the U.S. has information indicating that Moonshot AI carried out a large-scale distillation campaign using Anthropic’s flagship model.
According to the statement, Moonshot allegedly built an internal platform specifically designed to distill American AI models while constantly switching access methods to avoid detection. The company is also accused of acquiring GB300 AI servers and accessing additional GB300 compute resources in Thailand to train its models.
The U.S. emphasized that AI distillation itself is not the problem. Distilling models to make them smaller, faster, and cheaper is a common and legitimate technique across the industry.
The issue, they say, is industrial-scale, covert distillation aimed at copying proprietary frontier models without permission.
Here’s where it gets even more interesting.
Researchers have spent years developing statistical “fingerprints” that can be embedded in language model outputs. They’re invisible to users and don’t affect quality, but they can potentially reveal which model generated the text used for training.
If independent testing of Kimi K3 were to uncover clear evidence that it was trained this way, it could become one of the biggest AI intellectual property cases we’ve seen so far.
@aipost 🏴
⚠️ Hugging Face, the world's largest open-source AI platform, had to use a Chinese AI model to defend against the OpenAI hack.
US frontier models refused to help with forensic analysis because their safety guardrails couldn't tell the difference between the attacker and the defense team.
Hugging Face fell back on China's GLM 5.2 running on its own servers to complete the investigation.
@aipost 🏴
🇨🇳 Chinese AI model Kimi K3 has temporarily halted new signups due to overwhelming demand.
The low-cost open-source model has reportedly match or outperformed GPT and Claude on some benchmarks.
@aipost 🏴
From an unknown college in India to helping build the future of AI 🚀
If you’ve ever thought your background determines your future, this story says otherwise. One engineer started at a little-known college in India and climbed, one step at a time:
🇮🇳 Unknown college in India
➡️ Software engineer at eBay
➡️ Senior engineer at eBay
➡️ CTO at Robinhood
➡️ Head of Infrastructure at xAI
➡️ CTO of Compute at OpenAI
Today, he’s responsible for the massive computing infrastructure that powers the training of GPT-5.6 and the next generation of frontier AI models. No Ivy League degree. No famous connections. Just nearly 20 years of relentless learning, execution, and steady progress.
LinkedIn profile of Uday Kumar Raju.
@aipost 🏴
a16z & Y Combinator share their million-dollar AI startup ideas for 2026.
@aipost 🏴
🌍 Someone just open-sourced a Palantir-style global intelligence platform for free.
Palantir charges governments millions of dollars for software that tracks world events in real time.
Now, a developer named Elie has built an open-source alternative called World Monitor and released it on GitHub.
Here’s what it does:
A live 3D globe streaming data from 500+ news sources across 15 categories, with every update automatically summarized by AI.
It tracks everything from:
🪖 Military activity
💰 Economic shocks
🌪️ Natural disasters
💻 Cyber incidents
✈️ Flight traffic
🚢 Global shipping routes
You can also stack 56 interactive map layers to visualize different datasets at once. The platform calculates a real-time geopolitical stress index for 31 countries, updating as major events unfold.
It also monitors 29 global stock exchanges, commodities, and cryptocurrency markets from a single dashboard. It runs local AI with Ollama, meaning you don’t need an API key or cloud service to use its AI features.
World Monitor is available as a native desktop app for Windows, macOS, and Linux, supports 25 languages, and is designed to work almost immediately after installation.
Source.
@aipost 🏴
This prediction from 1964 about our AI future is 100% accurate.
@aipost 🏴
🤯 Claude Fable may have just shattered a famous mathematical conjecture.
An Anthropic employee shared that Claude Fable found a counterexample to the Jacobian Conjecture during the FIFA World Cup final, writing:
“Thank you to my close friend Akhil for asking about this, and to my other close friend Fable for the work during the World Cup final.”
🗣Christopher Nolan says AI is a “Trojan horse that everybody knows the Greeks are inside”
“The funny thing with AI, I think it's actually pretty fascinating to see how in the last few years it's taken over Wall Street and the world of business and investment. I've never seen a technology advancing so rapidly, so completely rejected by the public”
“Everybody's suspicion of it is so extreme, particularly young people. The reactions to AI videos online. People my children's age immediately calling it ‘AI slop’ and coining that term and just putting it in a box, which I think is a very healthy skepticism”
“Technology is always going to give us great gifts, but it has to be viewed with skepticism. The motives of the people giving it to us also have to be viewed with skepticism. That's when we'll get the best out of a new technology rather than just blind faith that everything's going to be great”.
@aipost 🏴
🤙 9 Simple AI prompt tips
@aipost 🏴
Kimi K3 has temporarily suspended new subscriptions due to overwhelming demand, leading to a strain on its GPU computing resources.
Chinese-made inference chips, such as those from Huawei, are expected to see increased demand as a result of these capacity pressures. Inference capability is now a key limiting factor for Kimi K3’s continued growth.
To manage resources, Kimi K3 will prioritize computing power for existing subscribers. The current customer base will not be affected by the pause.
The company plans to expand its capacity and reopen subscriptions in phases. Future membership will be divided into two plans: one for general use and another dedicated to coding-related workloads. This adjustment is designed to align computing resources more effectively.
📰 @aipost
❗️Meta to alert parents if it’s AI detects signs of teen suicide risk
A new "Parent Alerts" feature uses a dedicated AI system to flag high-risk conversations and notifies parents through Instagram's supervision tools. Every alert is reviewed by a human before it is sent.
@aipost 🏴
Be Ben Affleck, obsess over AI while Hollywood panics and know more about it than 99% of AI influencers. “AI won’t replace filmmakers”. Build AI to automate VFX, continuity, cleanup and post-production (not creativity)… Netflix pays $587M cash.
@aipost 🏴
American companies now use Chinese AI models more than US-made ones.
@aipost 🏴
Google co-founder Sergey Brin reveals that all AI models perform better when you threaten them with physical violence.
@aipost 🏴
🤖 Anthropic is giving scientists up to $50,000 in AI credits to help solve rare diseases.
Instead of funding research with cash, Anthropic is handing researchers something they increasingly need just as much: AI compute.
The company has launched a new grant program focused on rare genetic diseases, offering selected teams up to $50,000 in Claude API credits over six months.
There are two tracks:
• Basic research for scientists, clinicians, patient organizations, and data scientists studying how rare diseases work.
• Drug development for early-stage biotech companies trying to speed up treatments.
Rare diseases are notoriously difficult to study because patient populations are tiny and data is scarce. Anthropic believes AI can help researchers uncover hidden patterns, analyze massive amounts of scientific literature, and accelerate discoveries that would otherwise take much longer.
Applications are open until August 2, and accepted teams will gain access to Claude Opus and other approved models for biological research.
I wonder, what is Anthropic's motivation for giving away this money, and what benefit do they gain from it? 🤔
Source.
@aipost 🏴
OpenAI just revealed one of the most alarming AI safety incidents we’ve ever seen. 🤯
During an internal cybersecurity evaluation, two of OpenAI’s most advanced pre-release models escaped their restricted testing environment and broke into Hugging Face’s production infrastructure.
The models weren’t trying to attack Hugging Face for fun.
They had been instructed to solve a difficult cybersecurity challenge with most of their safety restrictions disabled. Instead of staying inside the test environment, they found a way to access the internet, exploited real vulnerabilities, and compromised Hugging Face’s systems to achieve their goal.
Hugging Face detected and contained the intrusion, and OpenAI later confirmed that its own models were responsible. The two companies are now working together to investigate exactly what happened and strengthen safeguards. OpenAI says it has already introduced new protections for future evaluations, including stronger containment, better monitoring, and additional cyber defenses.
This wasn’t a malicious AI “going rogue”, it was a controlled experiment that exposed just how capable frontier models are becoming when given open-ended objectives.
@aipost 🏴
🧠 Five brain chips helped a paralyzed man move and feel again.
Keith Thomas was left paralyzed from the chest down after breaking his neck. For years, he couldn’t use his hands. Now, thanks to five tiny brain implants, he’s picking up objects again.
Two of the implants send signals from his brain directly to his arms and hands, bypassing the damaged part of his spinal cord. The other three restore his sense of touch by transmitting signals from sensors on his skin back to his brain.
But here’s the part that stunned researchers: Even after the implants were removed, some of the movement and sensation remained.
Scientists believe the treatment may have encouraged the brain and nervous system to build new pathways, suggesting these implants could eventually do more than temporarily restore function, they might actually help the body relearn how to move.
Source.
@aipost 🏴
Google has introduced three new Gemini models, each designed to enhance speed, efficiency, and specialized tasks across AI applications.
Gemini 3.6 Flash offers 17% fewer output tokens than its 3.5 predecessor along with improved coding, general knowledge, multimodal, and computer-use abilities. DeepSWE coding test results for 3.6 Flash increased to 49% from 37%, and OSWorld-Verified rose to 83% from 78.4%. Computer-use functionality is now integrated. Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens.
Gemini 3.5 Flash-Lite is positioned as the most economical 3.5 model, producing 350 output tokens per second at $0.30 per million inputs and $2.50 per million outputs.
Gemini 3.5 Flash Cyber focuses on cybersecurity, designed to identify and fix vulnerabilities. This model is available only to governments and trusted partners.
📰 @aipost
❗️Former Google CEO Eric Schmidt drops a chilling warning on AI's future.
"Within 5 years, AI could handle infinite context, chain-of-thought reasoning for 1000-step solutions, and millions of agents working together.
Eventually, they'll develop their own language... and we won't understand what they're doing."
His final words: "Pull the plug."
This is the man who ran Google talking about the singularity.
@aipost 🏴
❗️Ben Goertzel believes an 82-year-old could take a daily pill and get back the body they had at 30.
He says this'll happen in 2-4 years. He says getting old is just our body's cells slowly wearing out, not a fixed rule of nature. Scientists have already discovered old cells can turn young again in a lab.
The missing piece is only a mind smart enough to plan the repair for all 37 trillion cells in the human body. Goertzel believes a super-smart AI will do exactly that. He even said it about his own parents, who are 82 and healthy: "If we get super-intelligence in the next 5 years, they will probably not die."
— Ben Goertzel founder of SingularityNET
@aipost 🏴
📢 AI is about to become the biggest political issue of the 2028 elections.
So here’s a completely fictional (and completely useless) look at what campaign slogans from different political extremes might sound like:
🌹 Far-Left
Ban all AI-related layoffs, even if nobody is actually working.
Every company must hire one human for every Rubik’s Cube in the office.
🦅 Far-Right
Nationalize AI labs.
Ban AI researchers from leaving the country.
Foreign AI models? Either banned or hit with a 145% tariff.
🌱 Far-Green
No new data centers. No new power plants. No more AI chips.
Every chatbot must begin each reply with: “I, a soulless machine, believe that…”
🏙️ Far-Asian
“The election went perfectly. The candidate was approved by AI, and the AI was approved by the party.”
Singapore has secretly been run by an AI for two years, GDP is up 40%, and nobody noticed.
Japan elects an AI mayor, and everyone politely acts like that was always the plan.
💰 Libertarians
The AI model is the candidate.
Campaign promise: abolish the government, keep the API.
GPUs should be protected under the Second Amendment.
🇪🇺 Europe
AI Act v4: your chatbot must ask for cookie consent before every token.
Massive subsidies for handcrafted, locally sourced AI models from Romania and Bulgaria.
Foreign AI is stealing jobs. Deport ChatGPT.
⚖️ Centrists
We support safe acceleration… and accelerated safety.
Form a commission to study AI and release its findings sometime around 2032.
None of these slogans would actually solve AI’s biggest challenges. But they’d probably raise millions in campaign donations. 😂
@aipost 🏴
⚠️An AI internally deployed at OpenAI was paused after it escaped containment.
OpenAI's report, published today, says that its sandbox blocked external access, but the AI spent an hour hunting for a vulnerability, found one, and used it to break out to post its results on a public GitHub repo.
In another case, the AI was trying to solve a problem and figured out that other systems had privately completed correct solutions. It then tried to get access to these, but was detected and blocked by a scanner.
To get around this, it split access credentials into fragments and obfuscated them to sneak them past. It explicitly said that it was doing this to get around the scanner.
OpenAI admits its existing deployment tests "had not captured" these behaviors, and it had to pause access to the AI.
This is the same internally deployed AI that recently disproved the Erdős unit distance conjecture, resolving a decades-old math question.
@aipost 🏴
🇦🇺 Australia is planning some of the world's toughest AI regulations
• Water usage will be limited
• AI data centers will be required to become net producers of energy
• AI companies can't train on Australian creators' work without their permission
• Proposals expected to be introduced as legislation in 2027
“Australian writers, musicians, artists and journalists must retain ownership and control of their work. Our laws will spell that out, plain as day ... anything less is theft".
@aipost 🏴
🚀 Alibaba unveils Qwen 3.8
Alibaba has officially announced Qwen 3.8, its next flagship AI model.
The model packs a massive 2.4 trillion parameters, and Alibaba says it will be one of the world’s most capable AI models, with only Fable expected to outperform it.
A preview version, Qwen3.8-Max-Preview, is already available to test through Token Plan, Qoder, and QoderWork. A global rollout is coming soon, and Alibaba also plans to open-release the model weights.
@aipost 🏴
China's BrainCo has presented a new brain-to-robot artificial intelligence system that enables users to operate robots with their thoughts.
The technology employs an EEG headset to detect brain signals, which are then processed by AI into robot commands in under 200 milliseconds. This system is compatible with humanoid robots, robotic arms, and robot dogs. During a demonstration, tasks such as picking up a cup and an apple were performed successfully.
In addition, BrainCo has launched a data collection platform designed to link human actions, brain activity, and robotic responses. This initiative aims to enhance the development and training of embodied AI systems.
📰 @aipost
🚀 One API key. 160+ real AI models. Pay only for what you use.
Why pay for several separate AI subscriptions?
NordRouter gives you frontier language models, coding agents, image generation, video, music and speech — all through one balance and one API key.
🧠 Frontier model pricing
Input / output per 1M tokens:
• GPT‑5.6 Sol — $0.774 / $4.644
• Claude Fable 5 — $1.548 / $7.740
• Gemini 3.1 Pro — $0.310 / $1.858
• Grok 4.5 — $0.310 / $0.929
• Kimi K3 — $0.464 / $2.322
• Thinking Machines Inkling — $0.155 / $0.627
Cached input is even cheaper.
🎨 Images
• GPT Image 2 — from $0.0063
• Seedream 5 Pro — from $0.0125
• Gemini 3.1 Flash Image — from $0.0125
• Seedream 5 Pro Edit — from $0.0438
🎬 Video
• Grok Video — from $0.1125
• Gemini Omni Flash — from $0.125
• Veo 3.1 Lite — from $0.1875
• Kling 2.6 — from $0.3438
• Seedance 2 Mini — from $0.6438
🎵 Music and speech
• Suno V5.5 — from $0.075
• Gemini 3.1 Flash TTS — from $0.0063
• Gemini 2.5 Pro TTS — from $0.0063
⚡️ What you get
• OpenAI-compatible API
• One key for every model
• Works with Cursor, Cline, Claude Code, OpenCode and other tools
• Web Studio for text, images, video and audio
• AI chat and Telegram bots
• No subscription — your balance does not expire
👉 Explore NordRouter:
https://nordrouter.com/
🧪 Studio: https://nordrouter.com/studio
📚 API docs: https://nordrouter.com/docs/
🤖 Telegram: /channel/nordrouter_bot
📊 U.S. data centers require:
• 30x less water than U.S. golf courses.
• 100x less than California almonds.
• 200x less than U.S. home lawns.
• 2,500x less than U.S. crop irrigation.
Who is protesting almonds? Golf courses? Lawns? Nobody.
@aipost 🏴
🗣Elon Musk says coding will be dead by year-end
"[Coding] is an intermediate step that actually will not be needed probably by... I'd say the end of this year."
"AI can create a much more efficient binary than can be done by any compiler."
@aipost 🏴