Media is too big
VIEW IN TELEGRAM
Runway just introduced Solaris, calling it the first Interface World Model. Instead of standard code, it generates user interfaces on the fly as interactive video.
Rather than relying on HTML and DOM, the system renders a continuous 720p video stream. An LLM handles the logic and reasoning, while an adapted Gen-4.5 model renders real-time responses to user inputs.
In Runway's internal study with 250 participants and around 7,500 comparisons, Solaris was preferred over Claude Opus 5 in 61% of cases for instruction following and 71% for natural behavior. The technology is currently available only to early partners, with known limitations around text rendering, long sessions, accuracy, and overall interface availability.
Demo | Blog post
#GenAI
Rather than relying on HTML and DOM, the system renders a continuous 720p video stream. An LLM handles the logic and reasoning, while an adapted Gen-4.5 model renders real-time responses to user inputs.
In Runway's internal study with 250 participants and around 7,500 comparisons, Solaris was preferred over Claude Opus 5 in 61% of cases for instruction following and 71% for natural behavior. The technology is currently available only to early partners, with known limitations around text rendering, long sessions, accuracy, and overall interface availability.
Demo | Blog post
#GenAI
❤1
This media is not supported in your browser
VIEW IN TELEGRAM
AI models were challenged to recreate the iconic "American Gothic" painting, and one of them showed a surprising artistic insight.
Claude Fable 5, Opus 5, and GPT Sol were given the same prompt: to draw the entire artwork, step-by-step, within a single HTML file. The goal was to see how they would rebuild the famous painting from scratch using code.
Fable 5 really stood out. Before it even drew the pitchfork, the AI explained that its teeth echoed the stitching on the man's overalls. This is a detail art historians actually point out, even though it wasn't mentioned in the prompt given to the AI.
#GenAI
#Science
Claude Fable 5, Opus 5, and GPT Sol were given the same prompt: to draw the entire artwork, step-by-step, within a single HTML file. The goal was to see how they would rebuild the famous painting from scratch using code.
Fable 5 really stood out. Before it even drew the pitchfork, the AI explained that its teeth echoed the stitching on the man's overalls. This is a detail art historians actually point out, even though it wasn't mentioned in the prompt given to the AI.
#GenAI
#Science
AI just tackled one of the biggest milestones in modern mathematics. Claude has fully formalized the proof of Fermat's Last Theorem.
Originally proved by British mathematician Andrew Wiles in the 1990s, the complex proof was fed to Claude to be broken down into thousands of smaller logical steps. The model translated everything into the Lean programming language so a computer could automatically verify that every single step is correct.
The project ran mostly autonomously using a team of dozens of AI agents that received only occasional human guidance. They wrote about 13 million lines of Lean code and proved 29.5 thousand sub-theorems. Formalizing complex modern math used to take years of human work, but Claude finished the job in just 11 days using about 6 billion output tokens from a model comparable to Fable 5.1.
Research
#GenAI #Science
Originally proved by British mathematician Andrew Wiles in the 1990s, the complex proof was fed to Claude to be broken down into thousands of smaller logical steps. The model translated everything into the Lean programming language so a computer could automatically verify that every single step is correct.
The project ran mostly autonomously using a team of dozens of AI agents that received only occasional human guidance. They wrote about 13 million lines of Lean code and proved 29.5 thousand sub-theorems. Formalizing complex modern math used to take years of human work, but Claude finished the job in just 11 days using about 6 billion output tokens from a model comparable to Fable 5.1.
Research
#GenAI #Science
HuggingFace is taking open-source robotics in a surprisingly cute direction.
The team has launched MicroDuck, a compact robot duck priced at $399, created together with Pollen Robotics. It is designed as an accessible hardware platform for developers and researchers exploring physical AI and robotics.
You can learn more on the Product page.
#Robotics #Hardware
The team has launched MicroDuck, a compact robot duck priced at $399, created together with Pollen Robotics. It is designed as an accessible hardware platform for developers and researchers exploring physical AI and robotics.
You can learn more on the Product page.
#Robotics #Hardware
This media is not supported in your browser
VIEW IN TELEGRAM
OpenAI's boardroom drama is officially making its way to the big screen.
A trailer has been released for "Artificial," a new movie covering the high-stakes power struggle inside OpenAI.
Further details about the film's cast or official release date haven't been announced yet.
#ResponsibleAI #Startups
A trailer has been released for "Artificial," a new movie covering the high-stakes power struggle inside OpenAI.
Further details about the film's cast or official release date haven't been announced yet.
#ResponsibleAI #Startups
❤2
This media is not supported in your browser
VIEW IN TELEGRAM
Can you tell AI-generated images from real photos? A new tool helps you test your skills.
Slopcheck is a daily game that shows you six images. Your task is to guess which ones are real photographs and which were made by AI.
After you make your guesses, the service points out the tell-tale signs that reveal an AI generation. It's a fun way to train your eye and avoid falling for "AI slop."
Try it
#GenAI
#AITools
Slopcheck is a daily game that shows you six images. Your task is to guess which ones are real photographs and which were made by AI.
After you make your guesses, the service points out the tell-tale signs that reveal an AI generation. It's a fun way to train your eye and avoid falling for "AI slop."
Try it
#GenAI
#AITools
Media is too big
VIEW IN TELEGRAM
OpenAI has released ChatGPT Images 2.5, bringing noticeable upgrades to image generation and editing.
The update focuses heavily on precise editing, allowing users to tweak specific details without distorting the rest of the picture. This is especially useful for multi-turn editing, where the image holds its consistency across several steps instead of warping over time. ChatGPT also gained the ability to generate images from sketches, along with handy templates for common formats like logos and posters.
For developers, two model versions are available in the API: the flagship Sunburst and the lighter Flare model. Pricing for both is set at $8.00 per 1 million input tokens and $30.00 per 1 million output tokens.
#GenAI #AITools
The update focuses heavily on precise editing, allowing users to tweak specific details without distorting the rest of the picture. This is especially useful for multi-turn editing, where the image holds its consistency across several steps instead of warping over time. ChatGPT also gained the ability to generate images from sketches, along with handy templates for common formats like logos and posters.
For developers, two model versions are available in the API: the flagship Sunburst and the lighter Flare model. Pricing for both is set at $8.00 per 1 million input tokens and $30.00 per 1 million output tokens.
#GenAI #AITools
Apple claims it found "shocking evidence" that a former engineer stole trade secrets to help build hardware at OpenAI.
According to a court filing, former Apple engineer Chang Liu downloaded confidential power converter schematics after leaving for OpenAI in January. He then ran the files through circuit simulation software using an AI agent. Liu reportedly noted in messages that the AI learned to run simulations and tweak parameters on its own, cutting a full day of work down to just two hours.
Apple argues that because the trade secrets were fed directly into an AI model, the damage is practically impossible to reverse. The company also claims Liu tried to cover his tracks on personal devices once the investigation started. OpenAI denies all charges and is pushing to dismiss the lawsuit.
You can read the full Court Filing for more details.
#ResponsibleAI #Hardware
According to a court filing, former Apple engineer Chang Liu downloaded confidential power converter schematics after leaving for OpenAI in January. He then ran the files through circuit simulation software using an AI agent. Liu reportedly noted in messages that the AI learned to run simulations and tweak parameters on its own, cutting a full day of work down to just two hours.
Apple argues that because the trade secrets were fed directly into an AI model, the damage is practically impossible to reverse. The company also claims Liu tried to cover his tracks on personal devices once the investigation started. OpenAI denies all charges and is pushing to dismiss the lawsuit.
You can read the full Court Filing for more details.
#ResponsibleAI #Hardware
❤2
Apple just unveiled the iPhone 18 Pro and 18 Pro Max, bringing a notable hardware upgrade to its flagship lineup.
The main feature is a 48-megapixel main camera with a physical six-blade variable aperture. It mechanically controls the amount of light reaching the sensor, improving low-light photography and offering finer control over depth of field.
Inside, the devices feature the new A20 Pro chip, an updated cooling system, and an improved battery. Apple also introduced a smarter Siri capable of understanding on-screen context. It launches with iOS 27 this month in English beta, with French, Japanese, Korean, Portuguese, and Spanish support coming in October.
Source
#Hardware #AITools
The main feature is a 48-megapixel main camera with a physical six-blade variable aperture. It mechanically controls the amount of light reaching the sensor, improving low-light photography and offering finer control over depth of field.
Inside, the devices feature the new A20 Pro chip, an updated cooling system, and an improved battery. Apple also introduced a smarter Siri capable of understanding on-screen context. It launches with iOS 27 this month in English beta, with French, Japanese, Korean, Portuguese, and Spanish support coming in October.
Source
#Hardware #AITools
Apple has unveiled its foldable smartphone, the iPhone Duo.
The device lets you run two apps side by side on a single screen. Apple also tweaked interface elements to make simple interactions, like hitting the alarm buttons, much easier.
Prices start at $1,999, and official sales begin on October 23.
#Hardware
The device lets you run two apps side by side on a single screen. Apple also tweaked interface elements to make simple interactions, like hitting the alarm buttons, much easier.
Prices start at $1,999, and official sales begin on October 23.
#Hardware
Major music publishers are taking Anthropic to court over copyright infringement.
Sony Music Publishing and Warner Chappell allege that Anthropic trained its Claude AI models on tens of thousands of copyrighted works without permission. The lawsuit claims the company downloaded pirated books via torrents and scraped lyrics directly from song databases. The suit names not only Anthropic as a company, but also CEO Dario Amodei and co-founder Benjamin Mann personally.
The publishers are demanding up to $150,000 per infringed work, along with up to $25,000 for each instance where copyright management information was removed. If the court agrees to all claims, total damages could reach several billion dollars.
Read the full report on Source.
#ResponsibleAI #GenAI
Sony Music Publishing and Warner Chappell allege that Anthropic trained its Claude AI models on tens of thousands of copyrighted works without permission. The lawsuit claims the company downloaded pirated books via torrents and scraped lyrics directly from song databases. The suit names not only Anthropic as a company, but also CEO Dario Amodei and co-founder Benjamin Mann personally.
The publishers are demanding up to $150,000 per infringed work, along with up to $25,000 for each instance where copyright management information was removed. If the court agrees to all claims, total damages could reach several billion dollars.
Read the full report on Source.
#ResponsibleAI #GenAI
OpenAI has paused sales of its $200 monthly subscription plan.
The decision was made to ensure current users maintain reliable access to Astra without running into capacity issues.
New sign-ups for the high-tier subscription are temporarily on hold, while existing subscribers can continue using the service as usual.
Source
#GenAI #AITools
The decision was made to ensure current users maintain reliable access to Astra without running into capacity issues.
New sign-ups for the high-tier subscription are temporarily on hold, while existing subscribers can continue using the service as usual.
Source
#GenAI #AITools
Google DeepMind just mapped out every possible single-letter mutation in human DNA.
The new dataset, called AlphaGenome Atlas, uses AI to predict the effects of roughly 9 billion DNA variants. For every single-character swap in the human genome, the model evaluates how it might alter gene activity and other molecular processes.
This lets researchers filter through millions of genetic variations in seconds to spot suspicious candidates for lab testing. DeepMind's partners have already used the atlas to uncover the roots of rare diseases, successfully confirming several AI predictions in real-world experiments.
The atlas is available for free for academic research via a website and API. You can read the full announcement in the Source.
#Biotech #Science
The new dataset, called AlphaGenome Atlas, uses AI to predict the effects of roughly 9 billion DNA variants. For every single-character swap in the human genome, the model evaluates how it might alter gene activity and other molecular processes.
This lets researchers filter through millions of genetic variations in seconds to spot suspicious candidates for lab testing. DeepMind's partners have already used the atlas to uncover the roots of rare diseases, successfully confirming several AI predictions in real-world experiments.
The atlas is available for free for academic research via a website and API. You can read the full announcement in the Source.
#Biotech #Science
❤3
Amazon is preparing a massive expansion for its drone delivery service. By the end of 2026, the company plans to scale up coverage sixfold, reaching nearly 500 US cities.
The drones can deliver almost any item weighing up to 2.3 kilograms in about 30 minutes for a $5 fee. Amazon has already flown hundreds of thousands of packages to customers this year alone.
Meanwhile, competition in the airspace is heating up as Walmart and Google's Wing build out their own drone delivery networks.
Source
#Robotics
The drones can deliver almost any item weighing up to 2.3 kilograms in about 30 minutes for a $5 fee. Amazon has already flown hundreds of thousands of packages to customers this year alone.
Meanwhile, competition in the airspace is heating up as Walmart and Google's Wing build out their own drone delivery networks.
Source
#Robotics
❤2
OpenAI just made voice AI feel a lot more like a real conversation. Their new model, GPT-Live-1, lets you interrupt it mid-sentence.
The model continues listening to the user even while delivering its own response. OpenAI has made GPT-Live-1 available to developers via API.
Response latency has dropped to 0.8 seconds, while accuracy with external tools reached 87%. API access costs $0.05 per minute, which comes out to about $3 per hour of talk time.
Source
#GenAI #AITools
The model continues listening to the user even while delivering its own response. OpenAI has made GPT-Live-1 available to developers via API.
Response latency has dropped to 0.8 seconds, while accuracy with external tools reached 87%. API access costs $0.05 per minute, which comes out to about $3 per hour of talk time.
Source
#GenAI #AITools
Meta has launched Muse, an autonomous personal AI agent designed to handle web tasks for you in the background.
Powered by the Muse Spark model, the agent can navigate browsers and apps to send emails, book travel, fill out forms, and shop online. You can control it through a dedicated app, a website, or directly inside WhatsApp.
To protect privacy, each user's data sits on an isolated virtual machine. Meta is also working with Signal's creator to launch Confidential VMs later this year, ensuring even Meta won't have access to your data. For shopping, Muse uses Stripe Link to generate single-use virtual cards so stores never see your real payment details, with 1Password integration planned next.
Source
#AITools #GenAI
Powered by the Muse Spark model, the agent can navigate browsers and apps to send emails, book travel, fill out forms, and shop online. You can control it through a dedicated app, a website, or directly inside WhatsApp.
To protect privacy, each user's data sits on an isolated virtual machine. Meta is also working with Signal's creator to launch Confidential VMs later this year, ensuring even Meta won't have access to your data. For shopping, Muse uses Stripe Link to generate single-use virtual cards so stores never see your real payment details, with 1Password integration planned next.
Source
#AITools #GenAI
Norway is considering a ban on camera-equipped smart glasses from companies like Meta and Snap.
The main concern is privacy: these devices make it easy to film people in public without their knowledge or consent. When combined with AI facial recognition, the glasses can quickly turn into portable surveillance tools in everyday life.
To address the issue, the Norwegian government is forming an expert group to evaluate the risks and propose clear rules for wearable cameras.
Source
#ResponsibleAI #Hardware
The main concern is privacy: these devices make it easy to film people in public without their knowledge or consent. When combined with AI facial recognition, the glasses can quickly turn into portable surveillance tools in everyday life.
To address the issue, the Norwegian government is forming an expert group to evaluate the risks and propose clear rules for wearable cameras.
Source
#ResponsibleAI #Hardware
❤3
This media is not supported in your browser
VIEW IN TELEGRAM
Unitree's humanoid robots can now spar completely on their own, no human operator required.
The company introduced UnifoLM-X2-1.0, a fully autonomous AI model designed for robot combat. It enables humanoids to analyze an opponent's movements, predict their next steps, and decide how to act during a fight.
Instead of relying on remote controls or pre-programmed routines, the AI handles the tactical decision-making autonomously in real time.
#Robotics
The company introduced UnifoLM-X2-1.0, a fully autonomous AI model designed for robot combat. It enables humanoids to analyze an opponent's movements, predict their next steps, and decide how to act during a fight.
Instead of relying on remote controls or pre-programmed routines, the AI handles the tactical decision-making autonomously in real time.
#Robotics
Uber is cutting 10% of its workforce and almost entirely ending remote work.
The company is laying off 3,300 employees in its largest downsizing since the pandemic, citing overgrowth. Management roles will drop by 20%, small teams of one or two people will be cut nearly in half, and fewer than 1% of staff will be allowed to work remotely.
The saved capital will go into ride-hailing, delivery, and a heavy push into autonomous transport. Uber plans to invest over $10 billion in robotaxis over the coming years.
Source
#Robotics
The company is laying off 3,300 employees in its largest downsizing since the pandemic, citing overgrowth. Management roles will drop by 20%, small teams of one or two people will be cut nearly in half, and fewer than 1% of staff will be allowed to work remotely.
The saved capital will go into ride-hailing, delivery, and a heavy push into autonomous transport. Uber plans to invest over $10 billion in robotaxis over the coming years.
Source
#Robotics
An AI-designed drug built to treat lung disease might have an unexpected side effect: turning back the biological clock in humans.
A study published in Nature Biotechnology analyzed trial data for rentosertib, a medication created using AI to treat pulmonary fibrosis.
Researchers tracked blood protein levels in trial participants over 12 weeks and estimated their biological age using six different calculation methods. All six methods showed a drop in biological age for patients taking the drug. Scientists caution that it is still unclear whether rentosertib directly targets aging or if the drop is just a byproduct of improved lung health. Rentosertib is currently undergoing Phase 3 clinical trials.
Paper
#Biotech #Science
A study published in Nature Biotechnology analyzed trial data for rentosertib, a medication created using AI to treat pulmonary fibrosis.
Researchers tracked blood protein levels in trial participants over 12 weeks and estimated their biological age using six different calculation methods. All six methods showed a drop in biological age for patients taking the drug. Scientists caution that it is still unclear whether rentosertib directly targets aging or if the drop is just a byproduct of improved lung health. Rentosertib is currently undergoing Phase 3 clinical trials.
Paper
#Biotech #Science