Robotics is entering a new phase as researchers at Yale and Princeton use AI coding agents to teach robots complex tasks like building furniture or brewing coffee. By leveraging Large Language Model agents that reason through commands and break them into subtasks, the team cut the time to assemble an IKEA coffee table from two weeks to just one hour. Simulations offer a cheaper and safer alternative to real hardware, and the goal is to let robots follow high-level human instructions with minimal intervention.
Meanwhile, UC Berkeley professor Ken Goldberg and NVIDIA's Spencer Huang introduced a concept called agentic robotics. Their approach uses AI agents to generate, test, and refine robot programs without requiring massive training datasets. Using NVIDIA's Isaac simulation libraries and OpenAI models, the team let AI agents rehearse robot behavior in simulation, and the resulting code worked when deployed on a physical robot. The group also published an industry-standard benchmark for evaluating robot performance on industrial tasks.
At Google, Tulsee Doshi, head of product for frontier models, argued that product managers are more valuable than ever because building with AI models has become cheaper and more unpredictable. She compared model work to a potions class where outcomes are uncertain and stressed the need to turn surprise capabilities into clear behavior specs. She also highlighted the balance between simple user experiences for mass products and more advanced controls for power tools, while emphasizing monitoring systems over formal proof for safety.
On the battlefield, Ukraine is actively using AI in its conflict with Russia, deploying AI-powered drones to track targets and software that assists with attack planning. Dr. Natalia Otel said the technology helps the military make more informed decisions, but warned that AI in warfare is a double-edged sword since systems could be hacked or compromised. She expects AI use in conflicts to grow even more widespread in the years ahead.
Concerns about bias in AI persist, with examples like facial recognition systems making more errors for people of color and AI assistants linking safety advice to certain ethnic backgrounds. Patrick Charlier of Belgium's Unia stated that artificial intelligence can be racist, though Hugues Bersini of the Université libre de Bruxelles noted that AI has no feelings or intentions. The technology can reproduce patterns tied to racial prejudice while appearing neutral, reinforcing existing societal biases.
In the startup world, Siggy, a virtual mental health company, emerged from stealth mode with $6 million in pre-seed funding. The startup uses AI to assist with therapy and prescription management, aiming to make mental health care more accessible and affordable. The funds will support technology improvements and service expansion.
Abridge, an AI healthcare company valued at over $5 billion, confirmed it carried out layoffs, though it did not disclose how many employees were affected or where they were based. The Pittsburgh-based company, which also has offices in the Bay Area and New York City, said impacted workers received transition packages and outplacement support.
A naming debate has erupted after President Donald Trump signed an order directing federal agencies to refer to AI as Super Intelligence rather than artificial intelligence. The rationale is that the word artificial makes the technology seem fake, though the term artificial intelligence has been in use since the 1950s. The change has sparked debate over what the name implies and who gets to shape how we talk about technology.
Security concerns around AI are also in focus. Experts describe AI jailbreaks as a method where attackers use specially crafted prompts to trick models into ignoring safety rules, potentially leading to data theft or unauthorized access. Separately, Varonis Threat Labs developed NotC2, a proof-of-concept command-and-control framework that operates inside Amazon Web Services. It uses Amazon Bedrock to convert plain English into cloud commands and hides malicious activity within normal logs, though it only works after an attacker already has AWS credentials.
French President Emmanuel Macron called for AI to be banned wherever it threatens children's mental health or brain development, and also proposed banning some social networks for people under 15. A committee of experts will advise him, while research labs will follow a charter requiring AI systems to be tested before reaching children. The charter will involve companies like OpenAI and Anthropic, and leaders from several countries discussed AI risks together.
Key Takeaways
- Yale and Princeton researchers used AI coding agents to reduce IKEA coffee table assembly time from two weeks to one hour in simulation.
- NVIDIA and UC Berkeley introduced agentic robotics, using AI agents to generate and refine robot programs without massive training data.
- NVIDIA's Isaac simulation libraries and OpenAI models were used to rehearse and deploy working robot behavior on physical hardware.
- Google's Tulsee Doshi says product managers are critical for turning unpredictable AI model outputs into usable products and safety evaluations.
- Ukraine deploys AI-powered drones and planning software in its conflict with Russia, but experts warn AI in warfare could be hacked or compromised.
- Facial recognition and AI assistants may reproduce racial prejudices, though experts note AI has no feelings or intentions behind its outputs.
- Siggy AI mental health startup launched with $6 million in pre-seed funding to improve therapy and prescription management access.
- Abridge, an AI healthcare company valued at over $5 billion, confirmed layoffs without disclosing the number of affected employees.
- President Trump signed an order directing federal agencies to call AI Super Intelligence instead of artificial intelligence.
- Macron proposes banning AI that threatens children's mental health, with a charter involving OpenAI and Anthropic to test systems before they reach children.
Yale team uses AI coding agents to train robots for complex tasks
Researchers at Yale and Princeton are using AI coding agents to teach robots complex tasks like building furniture or brewing coffee. The system uses Large Language Model agents to reason through commands and break them into subtasks. The team found that simulations powered by coding agents cut the time to assemble an IKEA coffee table from two weeks to one hour. They say simulations are cheaper and safer than using real hardware. The project aims to give robots the ability to follow high-level human commands.
Agentic robotics marks a major shift in how robots are controlled
UC Berkeley professor Ken Goldberg and NVIDIA's Spencer Huang introduced a new approach called agentic robotics. The method uses AI agents to generate, test, and refine robot programs without needing massive amounts of training data. Researchers used NVIDIA's Isaac simulation libraries and OpenAI models to let AI agents rehearse and improve robot behavior in simulation. When the final code was run on a real robot, it worked. The team also released an industry-standard benchmark for evaluating robot performance on industrial tasks.
Google product lead says AI model work needs stronger product focus
Tulsee Doshi, head of product for Google's frontier models, argued that product managers are valuable again because building with AI models has become cheaper. She said model work is unpredictable and compared it to a potions class where results are uncertain. She stressed that product teams must turn surprise model capabilities into behavior specs and evaluations. She also said mass products like AI Mode need simple user experiences while power tools demand more advanced controls. On safety, she emphasized monitoring systems over proof, but noted that no eval numbers, incident rates, or launch dates were provided.
Ukraine uses AI on battlefields to track targets and plan attacks
Ukraine is using artificial intelligence in its conflict with Russia, from AI-powered drones that track targets to software that helps plan attacks. Dr. Natalia Otel, an expert on AI and warfare, said the technology helps Ukraine's military make more informed decisions and gain an advantage. AI-powered drones have been especially effective at identifying and tracking targets. However, Otel warned that AI in warfare is a double-edged sword because the systems could be hacked or compromised. She added that AI in warfare is likely to become even more common in future conflicts.
AI systems may reinforce society's existing prejudices
Examples such as facial recognition systems making more errors for people of color and AI assistants linking safety advice to certain ethnic backgrounds raise concerns about racial prejudice in AI. Patrick Charlier, director of Belgium's Unia, stated that artificial intelligence can be racist. However, Hugues Bersini, a computer science professor at the Université libre de Bruxelles, noted that AI has no feelings or intentions. He explained that AI cannot be racist in the human sense even though it can reproduce patterns linked to racial prejudice. The article highlights how algorithms may amplify existing societal biases while appearing neutral.
Siggy AI mental health startup launches with $6M funding
Siggy is a new virtual mental health startup that uses AI to help with therapy and prescription management. The company recently came out of stealth mode with $6 million in pre-seed funding. Siggy aims to make mental health care easier to access and more affordable. The money will go toward improving its technology and growing its services.
AI healthcare company Abridge confirms layoffs
Abridge, an AI healthcare company valued at over $5 billion, confirmed it laid off some workers. The company did not say how many people lost their jobs or where they were located. Affected employees received transition packages and outplacement support. Abridge is based in Pittsburgh and has offices in the Bay Area and New York City.
Debate over calling AI Super Intelligence instead of artificial intelligence
President Donald Trump signed an order telling federal agencies to call AI Super Intelligence instead of artificial intelligence. He said the word artificial makes the technology seem fake. The term artificial intelligence has existed since the 1950s. The change has started a debate about what the name means and who should decide how we talk about technology.
What is AI jailbreak and why is it dangerous
AI jailbreak is when someone tricks an AI model into ignoring its safety rules using special prompts. The attacker designs an input that forces the AI to share sensitive information or do things it should not do. This is dangerous because it can lead to data theft or unauthorized access to important systems. Experts say AI jailbreak needs to be addressed as AI becomes more common.
NotC2 AI framework hides malicious commands in AWS services
Varonis Threat Labs created NotC2, a proof-of-concept command-and-control framework that runs inside Amazon Web Services. It uses Amazon Bedrock to turn plain English into cloud commands. The tool only works after an attacker already has AWS credentials. It hides malicious activity in normal cloud logs, making it hard to detect.
Macron calls for AI ban to protect children's mental health
French President Emmanuel Macron said AI should be banned wherever it threatens children's mental health or brain development. He also wants to ban some social networks for people under 15. A committee of experts will give him recommendations on AI protections. Research labs will follow a charter that includes testing AI systems before they reach children. The charter will involve companies like OpenAI and Anthropic, and leaders from several countries discussed AI risks together.
Sources
- Yale-led project uses AI coding agents to teach robots complex tasks
- Recent advances in AI are creating a 'paradigm shift' for robotics
- Google's PM revival hides a product gap
- The future of AI warfare is playing out on Ukraine’s battlefields
- How Algorithms Can Reinforce and Amplify Society's Prejudices
- AI-enabled virtual mental health startup Siggy launches from stealth
- $5+ billion AI healthcare company confirms layoffs
- Reject the Sand God. Call AI by its true name
- Explained: AI jailbreak - what is it, how does it work, why is it a concern
- Researchers Uncover NotC2 AI-Powered C2 Framework That Hides Malicious Commands Inside AWS Services
- Macron wants AI ban wherever it threatens children
Comments
Please log in to post a comment.