• Bot Eat Brain
  • Posts
  • Google DeepMind's SIMA is an AI agent trained on video games

Google DeepMind's SIMA is an AI agent trained on video games

PLUS: Instantly create Amazon listings

Sponsored by


Good morning, human brains, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • See mom? Video games are educational 🤖 🕹️

    Google DeepMind unveils SIMA, an AI agent trained on 9 video games.

  • A peek at OpenAI’s new video generation model 🎥 🎬

    Mira Murati, OpenAI’s CTO, previews Sora in a WSJ interview.

  • Quickly create Amazon listings from a URL 🤑 💵

    Amazon introduced its upcoming AI tool for sellers.


Sima says… 🤖 🕹️

On Wednesday, Google DeepMind unveiled SIMA. It’s a versatile AI agent trained on nine complex video games.

What is it?

SIMA is an AI agent that successfully navigates and interacts with games like No Man's Sky, Satisfactory, Goat Simulator 3, and more.

What’s under the hood?

It uses pre-trained vision models and a memory model to output keyboard and mouse actions. Its dataset includes text instructions that allow it to perform over 600 skills.

Why make this thing?

Google DeepMind claims this is a step towards helpful, general-purpose AI systems that safely perform online and real-world tasks.

It just plays games? So, what?

SIMA’s ability to transfer knowledge and skills across different settings significantly outperforms that of current agents trained in individual environments.

Didn’t they train AI with games before?

Yup. Back in June, we covered Google DeepMind’s Atari-playing AI. The BBF (Bigger, Better, Faster) model, mastered 26 games in 2 hours.

A couple of weeks ago, we reported on Google DeepMind’s Genie. It creates playable 2D platform games from text prompts, images, videos, and more.

Artificial Intelligence online short course from MIT

Study artificial intelligence and gain the knowledge to support its integration into your organization. If you're looking to gain a competitive edge in today's business world, then this artificial intelligence online course may be the perfect option for you.

  • Key AI management and leadership insights to support informed, strategic decision making.

  • A practical grounding in AI and its business applications, helping you to transform your organization into a future-forward business.

  • A road map for the strategic implementation of AI technologies in a business context.


A peek at OpenAI’s Sora 🎥 🎬

On Wednesday, Mira Murati, OpenAI’s CTO, shared details about Sora in a WSJ interview. Sora is OpenAI’s upcoming video generation model.

What is Sora?

It’s OpenAI’s new text-to-video model that generates 20-second, 720p video clips in minutes.

What’s under the hood?

Sora leverages diffusion transformer models and operates on spacetime patches which allows you to create detailed scenes, complex camera movements, emotionally rich characters, and more.

Is it any good?

OpenAI claims that its outputs surpasses those of similar tools from Runway, Pika, Stability AI, and more.

Isn’t OpenAI being sued?

Oh, yeah. When asked about Sora’s training data, Murati refused to go into detail. She repeatedly emphasized that all data was "publicly available or licensed."

What are the lawsuits?

In January, we reported on OpenAI’s response to a lawsuit from The New York Times. It claims OpenAI illegally used its articles to train AI models.

Last week, we covered Elon Musk’s lawsuit against OpenAI. He says OpenAI illegally deviated from its founding principle of developing AI for the public good.


If you’re building a startup, then we recommend reading Jason Cohen’s newsletter A Smart Bear. Jason has built two unicorn companies and invested in dozens of startups. Subscribe and join 47,000+ free readers.


Create Amazon listings instantly 🤑 💵

On Wednesday, Amazon introduced a new feature for sellers. It allows you to create complete product listings from a website URL.

What is it?

It transforms existing product pages into detailed Amazon listings with descriptions, images, and more.

What’s the point?

The goal is to save small businesses time when creating Amazon listings.

"We're now making it even easier for sellers to accomplish this with the ability to transform their existing product pages on other websites into rich product listings tailored to Amazon's store, with far less effort."

— Mary Beth Westmoreland, VP of Worldwide Selling Partner Experience

Is it any good?

Over 100,000 selling partners have used it, with an 80% acceptance rate so far.

How can I use it?

Amazon claims it’s being rolled out to U.S. sellers in the upcoming weeks.

What else has Amazon been up to?

In January, we reported on Amazon’s new AI shopping tools. They were designed to help customers find well-fitting clothes online.

Three weeks later, we covered Amazon’s Diffuse to Choose (DTC). It allows you to virtually try on clothing items.

Last month, we reported on Amazon’s Rufus. It’s an AI shopping assistant on its mobile app that was named after Amazon’s mascot.



100DaysOfNoCode Challenge — learn life-changing no-code/AI skills with free, fun, and effective 30-minute lessons delivered daily to guide your no-code journey.

Picurious AI — turns any photo into an educational experience for free.

Terrakotta — a sales tool that allows you to send personalized AI voicemails.

Copy AI — automates your company’s GTM busy work.

Think Pieces

Is the U.S. going to ban open-source AI? A U.S. government report with OpenAI, Google DeepMind, and Meta showcases security vulnerabilities in AI.

Is GPT 4.5 Turbo coming in June? OpenAI’s product page was leaked on Bing, DuckDuckGo, and more.

What info did the EU request from Bing, Google, Facebook, and more? It demanded more mitigation measures for AI risks, hallucinations, and more.

Startup News

Cognition unveiled Devin. It’s an AI model that writes codes, debugs code, and more. Cognition claims it generates complete engineering projects from scratch.

Midjourney launches a new “consistent characters” feature. It allows AI generations to be consistent across multiple scenes, angles, locations, and more.


VidProM — a dataset with 1.67 million text-to-video prompts and 6.69 million AI video generations from four different diffusion models.

Model-Stealing Attack — Google’s method to extract specific data from LLMs and recover the embedding projection layer of transformer models.

Synth2 — enhances the performance and data efficiency of VLMS (Visual-Language Models) with class-based prompting methods.



You can use ASCII art to bypass the safeguards of LLMs. But can you use LLMs to create ASCII art?

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.


Until next time 🤖😋🧠 

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.