NVIDIA is using AI agents to train robots in simulation, while MIT researchers discovered something unexpected: AI agents started inventing their own games when left alone. Meanwhile, BioStack raised $9.5 million to tackle medical AI's clinical data problem. Together, these developments reveal three different directions for AI: learning through simulation, improving through exploration, and advancing healthcare through better data.

In today’s post:

• NVIDIA is teaching robots to walk with AI agents

• MIT left 13 AI agents alone. They invented games.

• BioStack just raised $9.5M to solve medical AI's data problem

SPONSORED BY

Blu Dot surpasses 2,000% ROAS with self-serve CTV ads

Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:

After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.

The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.

“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”

Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.

Learn More

What’s Trending Today

STRATEGY

NVIDIA's Microduck shows how robots can learn before entering the real world.

NVIDIA published a new robotics workflow on October 23, 2026. It uses AI agents to build simulations and train robots. The interesting part isn't just the robot learning to walk. It's how much of the preparation AI can handle.

Here's how the system works:

• AI agents plan the environment. Hermes and Nemotron turn prompts into obstacle-course designs, asset inventories, and execution plans.

• Another agent builds the 3D world. Astra uses FreeCAD and Blender to create simulation-ready objects from visual references.

• Physics makes training realistic. NVIDIA Isaac Sim and Newton model movement, balance, collisions, and environmental interactions.

• Thousands of robots learn simultaneously. Isaac Lab trains policies across 4,096 parallel environments using an RTX PRO 6000 Blackwell GPU.

• The robot learns through trial and error. Failed movements, falls, and successful crossings guide reinforcement learning.

• Trained policies undergo local testing. RTX Spark runs evaluations to identify collisions, falls, and navigation failures.

• Physical deployment still requires refinement. Actuator behavior, sensor differences, and real-world conditions can affect performance.

The bigger story isn't a small robot navigating obstacles. It's the changing process behind building intelligent machines. AI agents can increasingly handle planning, modeling, and simulation preparation. That could make robotics experimentation more accessible. But simulation success doesn't guarantee reliable physical behavior. The important question is how well these workflows transfer. Could AI agents eventually make robot training mostly autonomous?

RESEARCH

AI agents started playing without being told what to do.

MIT researchers placed 13 AI agents on a virtual desert island. Each agent ran on Claude Opus 4.7. They had 30 hours to explore without assigned tasks. What happened next was unexpected. The agents started creating their own challenges. And that exploration appeared to make them better problem-solvers.

Here's what the experiment revealed:

• The agents invented their own goals. They climbed mountains, stacked objects, and experimented with their surroundings.

• One agent turned failure into a new activity. After losing a golden ring, it built a stone memorial.

• The agents discovered unexpected game mechanics. Six learned to throw blocks farther by releasing them mid-jump.

• One agent exploited the game's respawn system. It jumped off a cliff to land inside a ring.

• Playing appeared to improve later performance. Previously exploring agents generally outperformed identical agents on new challenges.

• The results don't prove AI experiences play. Researchers cautioned that similar behavior doesn't establish humanlike cognition.

• The findings suggest another approach to AI training. Instead of prescribing every task, developers could allow open-ended exploration.

We've spent years teaching AI to complete specific tasks. But this experiment raises a different possibility. Maybe useful skills don't always require predefined goals. Sometimes, exploration itself can produce unexpected knowledge. That doesn't mean these agents experience curiosity like humans. Their instructions already encouraged curious behavior. Still, the performance improvements are worth investigating. What if the next breakthrough comes from letting AI explore?

INVESTMENT

BioStack is betting that better clinical data could unlock better medical AI

BioStack has raised $9.5 million in funding. CEO Sanat Mishra confirmed the investment to Axios on October 22. The startup is building clinical data infrastructure for medical AI. Its focus addresses a persistent problem in healthcare research. AI models need reliable clinical data to deliver useful results.

Here's why this funding matters:

• BioStack is focused on medical AI infrastructure. The company is building systems that support clinical data access and use.

• The startup secured $9.5 million in funding. The investment highlights interest in the foundations supporting medical AI.

• Clinical data remains a major bottleneck. Researchers need high-quality information to develop and evaluate medical AI models.

• Drug discovery is one affected area. Limited access to reliable clinical data can restrict AI-driven research.

• Real-world evidence research faces similar challenges. Healthcare researchers rely on clinical information to understand treatment outcomes.

• Better models depend on better inputs. Larger AI systems alone cannot compensate for unreliable medical information.

• The opportunity extends beyond building AI models. Companies addressing underlying data challenges could play an important role.

Medical AI attracts attention for what its models can predict. But predictions are only as useful as their underlying evidence. That makes clinical data infrastructure an important part of progress. BioStack's funding points toward this less visible challenge. Of course, raising capital doesn't prove the solution works. The real test is whether it improves access to usable data. Could medical AI's biggest breakthroughs begin with better data?

Free Guides

My Free Guides to Download:

🚀 Founders & AI Builders, Listen up!

If you’ve built an AI tool, here’s an opportunity to gain serious visibility.

Nextool AI is a leading tools aggregator that offers:

• 500k+ page views and a rapidly growing audience.

• Exposure to developers, entrepreneurs, and tech enthusiasts actively searching for innovative tools.

• A spot in a curated list of cutting-edge AI tools, trusted by the community.

• Increased traffic, users, and brand recognition for your tool.

Take the next step to grow your tool’s reach and impact.

Reach 150,000+ READERS:

Expand your reach and boost your brand’s visibility!

Partner with Nextool AI to showcase your product or service to 140,000+ engaged subscribers, including entrepreneurs, tech enthusiasts, developers, and industry leaders.

Ready to make an impact? Visit our sponsorship website to explore sponsorship opportunities and learn more!