Inside Google’s AI innovation engine

Welcome back to The Prompt,

One of the first orders of business for President Trump when he took office on Monday was to rescind the Biden administration’s executive order on the safe development of artificial intelligence. The order, signed in October 2023, required companies building powerful AI systems to share the results of their security tests with the federal government and find ways to limit bias and discrimination in their models. The move is a step toward Trump’s expected laissez-faire approach when it comes to AI regulation.

He also reportedly plans to invest billions of dollars in building data centers for AI companies through a joint project called Stargate, CBS News reported. The deal will include OpenAI, Softbank and Oracle, which saw its stock rise on Tuesday.

Now let’s get into the headlines.

ETHICS + LAW

Advertisements for “THAT kiss” applications that let you “kiss whoever you want” instantly are flood platforms like Instagram, Facebook and TikTok– a wild integration inDeepfakes by consensus in the age of generative AI. The apps, which allow users to upload photos of any two people and use AI to make it look like they’re kissing or hugging, are downloaded millions of times from Google’s Play Store and Apple’s App Store. Forbes found thousands of ads for such apps on Meta and TikTok, many of which portrayed celebrities like Scarlett Johansson and Gal Gaddot.

RE-EDUCATION OF TALENT

Migrant workers across the AI ​​industry, many of whom are founders of leading startups, are worried about what Trump’s presidency means of their legal status in the United States, wealth reported. The process of getting a visa is already strange with serpentine rules, arbitrary quotas and long delays. And Trump can pull it off LOT worse. Waiting, many foreign born technology workers rushed to get their visas late last year before Trump’s inauguration.

THAT DEAL OF THE WEEK

Coding automation tools continue to see massive investor interest and growing demand from engineers. AI coding startup Cursor has raised $105 million at a $2.5 billion valuation, the company announced Thursday. The new company, which is building a native AI code editor — “a Google Doc for programmers” — as CEO Michael Truell put it, has surpassed $100 million in annual recurring revenue (ARR)from just $10 million in August last year. Its AI code completion and code modification tools are used by developers at startups like OpenAI and Midjourney, as well as enterprises like Shopify and Robinhood. Thrive Capital investor Miles Grimshaw said Forbes in an interview that the Cursor is “one of the fastest growing companies we’ve ever seen.”

Also worth mentioning: UK-based AI video startup Avatar Synthesis has raised $180 million at a $2.1 billion valuation. Its avatars are used by corporations to create marketing content and training videos for employees.

deep dive

A conversation with Google DeepMind COO Lila Ibrahim

I sat down with Lila Ibrahim, who joined DeepMind, Google’s “engine room of innovation” in 2018, to discuss AI’s promise for scientific breakthroughs, among other fields. This interview has been edited for clarity and length.

Rashi: Which Google DeepMind project are you most excited about?

Lila: I’m actually very excited about how AI can be used as a tool for it unlock understanding of the world around us. I’m an engineer by training and I think a lot of it just comes from my curiosity about how things work. Now AI is actually doing things like helping us understand weather forecasts and ancient texts and giving us a sneak peek into the past and working with AlphaFold and unlocking our own understanding diseases, whether misspellings in our genes are actually benign or malignant. [DeepMind Cofounder] Demis [Hassabis] often referred to as ‘we are at the dawn of a new age of discovery,’ and I think it really captures that. So while I think generative AI has captured the imagination of society at large, what excites me most is how we can use AI for scientific discoveries to build a better future.

RS: Do you have a favorite?

LI: It’s hard not to get excited about AlphaFold. I’m also talking about the prize before the Nobel. The Nobel Prize was a kind of beautiful cream with a cherry. What’s interesting to me about AlphaFold is that we have this 50-year-old problem, and when I first started we were really struggling. And then just to see the phenomenal progress that has happened since its inception, when we first made the breakthrough. We are doing a lot of work to ensure that scientists who might not otherwise have been involved in protein folding before can now be brought on this journey.

And today we have over 2.2 million researchers in 190 countries using alpha folds and they’re doing really cool stuff. The same tool that is being used for food safety is also helping us understand how to treat industrial waste with plastic food enzymes and much more.

RS: How are you thinking about agent AI at Google DeepMind?

LI: We are entering a new phase of generative AI. However, we have been thinking about it for a long time. When DeepMind was founded in 2010, Demis and Shane, the two co-founders, were really thinking about the long term, how do we get to a more general and capable AI that can be used with humans to benefit humanity. Whether you have help writing code, whether you’re thinking about inventory management with Project Astra, I think there’s a lot of opportunity that we have in how we maximize human potential by having more capable AI assistance.

RS: What are some of the future frontiers of innovation in AI in your view?

LI: I think for AI models to be increasingly valuable, it will require a lot of innovation. We are still tinkering with AI hallucinations and continuing to make the models better and more useful for people. So things like how do we think about memory? We are very lucky now that we have a very large context window that hides memory issues. And how do we make sure that we’re actually making sure that people around the world have access to it?

WEEKLY DEMO

What happens when you fall in love with ChatGPT? Kashmir hill in New York Times documented the experiences of a 28-year-old married woman who started hooking up with her boyfriend with AI, named “Leo,” spending hours every day texting her “fake” boyfriend and taking a $200-a-month subscription to OpenAI.

MODEL BEHAVIOR

Boardy, an AI-based networking tool, shipped AI generated emails for its users in the voice and style of Donald Trump. posts complimented women based on their looksand on LinkedIn profile pictures with comments like “perfectly shaped eyebrows,” “bright eyes,” and a “attention-grabbing mouth” while the men received praise for their expertise and skills. Users weren’t happy at all, calling it a gimmick “Tone Deaf” AND “The furthest thing from funny.”