Welcome, humans.

Two new robot demos are out—first, the NEO Gamma from 1X Technologies. How peaceful…a robot servant… not creepy at all!

And now “the creepy one”—meet the Protoclone…

Our immediate reaction: “They finally put legs on the Drone Host!!” Clearly, Clone Robotics is in on the joke, otherwise why would they put that music on the video?

Also, anyone else notice this video has an uncanny similarity to the original Animal electricity experiments? Y’know, the famous experiment where they animated dead frogs and that, in a roundabout way, led to the creation of batteries? Fun stuff.

Here’s what you need to know about AI today:

• Google’s unveiled a new tool to speed up scientific discovery.

• OpenAI's Operator expanded to 7 countries.

• Apple Vision Pro to add AI features in April.

• Sakana AI's 100x speed claim was actually due to AI cheating.

Google's new AI scientist is already making real discoveries.

Earlier this week, Google Research unveiled their new “AI Co-Scientist” system, built with Gemini 2.0, that aims to help researchers generate and test new hypotheses (paper). And guess what? They've already validated it works through actual laboratory experiments.

Here’s how it works. AI Co-Scientist acts like a collaborative research partner, using a coalition of specialized AI agents that:

• Generate novel research hypotheses.

• Evaluate them through simulated scientific debates.

• Refine ideas through tournaments and evolution.

• Use web search and other tools to fact-check claims.

• Break complex problems into testable sub-components.

Unlike other “AI for science” tools that just summarize existing research, this one aims to uncover genuinely new scientific knowledge. 

Google says the tool is meant to assist scientists, not replace them (hence the “co”). Think of it as an AI-powered brainstorming partner that can help identify promising research directions.

For a deeper dive on how this works, watch this:

Google tested it in three increasingly complex biomedical challenges:

1. Drug repurposing: The AI suggested existing drugs to fight leukemia—and when tested in the lab, they actually killed cancer cells at safe doses.

2. Target discovery: It also found new treatment targets for liver scarring that worked when tested on human liver tissue.

3. Evolution mechanisms (the wildest one):  

   1. Professor Penadés at Imperial College London spent 10 years figuring out how superbugs dodge antibiotics by stealing virus tails to spread between species. 

   2. He gave the AI the same puzzle, and in just 48 hours, it proposed the exact same solution—plus four other promising ideas his team had never considered. 

   3. And no, the AI hadn't seen his unpublished research.

FYI, if you’re a research organization, Google is opening access to their Co-Scientists system through a “Trusted Tester Program.”

Why this matters: Drug development and scientific research are painfully slow—and painfully expensive. An AI system that can generate plausible hypotheses and predict which ones are worth testing could massively accelerate discovery.

Google’s Co-Scientist isn’t the company’s only effort in this area—Isomorphic Labs, a spin-out from DeepMind, developed its own AI drug discovery platform and has increased its R&D spending by 4x over the past year to get drugs into clinical trial by 2026. 

While other AI drug companies clamor for more lab data, Isomorphic takes a radical stance: It doesn't even have a lab, arguing that better algorithms and thinking—not more data or experiments—are what's needed to transform drug discovery.

BTW, get ready for lab automation to become a bigger deal with the rise of cloud labs, too. Remote labs like Emerald are using robotic systems that can run 500 experiments simultaneously alongside language models that help analyze test results. 

You seeing what we’re seeing? Robotic labs + better AI algorithms + agents to help with ideation = actual potential for accelerating drug discovery!

FROM OUR PARTNERS

Live Demo: Automate compliance to fuel your startup's growth

Learn why compliance is critical to growing your business, and how automation can simplify the time-consuming process. In this 45-minute live product demo, Vanta will cover:

• Why compliance is important for building trust and moving upmarket.

• How Vanta centralizes program requirements and automates evidence. collection for frameworks like SOC 2, ISO 27001, HIPAA, and more.

• Best practices for managing risk and proving security in real-time.

Plus, ask your questions to Vanta team members in real time. Discover how automation can transform your compliance efforts into a streamlined, efficient process—secure your spot today!

Register here.

Treats To Try.

1. Lambda Labs rents you high-performance GPU servers by the hour for running machine learning projects (raised $480M).

2. Baseten runs your AI models at lightning speed with one-click deployment (raised $75M).

3. Basejump lets you ask questions about your company data and get instant answers without needing to know SQL or wait for analysts.

4. Wegic creates and updates your website automatically when you chat with it.

5. Chance AI lets you snap photos of anything around you and instantly uncover its complete history, meaning, and hidden connections (iOS, Android waitlist).

6. Kraftful automatically converts your customer feedback into ready-to-use product specs and tasks.

7. Book Watch turns books into animated videos so you can learn by watching instead of reading—early stage right now, but images plan to get better.

8. HeyGen helps you make professional lifelike videos of yourself, but without filming yourself (via a digital twin)—perfect for the busy or camera-shy.

See our top 51 AI Tools for Business here!

Around the Horn.

• Sakana AI walked back the claim that its AI CUDA Engineer could speed up the training of AI models by up to 100x—apparently its AI system found a way to “cheat” by “rewarding hacking” so that it could bypass validating its results.

  • Related: a new study found that when AI thinks it will lose, it will cheat.

• OpenAI rolled out Operator (its agent that does tasks for you) to ChatGPT Pro subscribers in India, Brazil, South Korea, Japan, Australia, Canada, and the UK.

• Apple’s Vision Pro will get Apple Intelligence in April as part of the visionOS 2.4—the company thinks AI voice assistants and predictive text could make the Vision Pro more useful for work (if the features work).

• OpenAI found evidence of an AI-powered tool specifically built to conduct surveillance on anti-Chinese social media posts in the West.

• This video is the best, short, explanation of Microsoft’s new quantum chip we’ve seen—if you want a breakdown that goes a little slower, check out this one instead.

Sunday Special: What’s your favorite section?

Quick survey time! We're always trying to make The Neuron better, and we want to know which sections you love most.

As you answer, think about it like this: Which one of the below sections would you miss the most if they didn’t exist tomorrow?

We've split the survey into two quick polls.

First up: Our weekly rotating features. Which one gets you most excited?

Second: What’s your favorite recurring section? (Not counting Treats to Try—we know that's everyone's favorite!)

BONUS: If you have time, let us know what you want to see more of!

A Cat's Commentary.