Wednesday, February 26, 2025

Welcome to TechCrunch AM! This morning, we're taking a look at how well an AI bot can play Pokémon, who's really leading DOGE, and the executive shake-up at Lucid Motors. We also have notes on the latest drama around Y Combinator, ElevenLabs' newest attempt to steal voice artists' jobs, and why your GitHub repository isn't as secure as you think it is. Let's dive in!

TechCrunch Top 3

Image Credits: Claude Plays Pokémon on Twitch

1. AI choose you! Anthropic's Claude 3.7 Sonnet is playing Pokémon on Twitch for the world to see. AI researchers commonly use video games to test models, and Anthropic says Pokemon is helping Claude think through puzzles. Thing is, it's really slow at the game. Read More

2. All the president's memes: DOGE has confirmed that Amy Gleason has been appointed as the body's acting administrator. The news came after the White House said that Musk was not officially leading the charge, prompting the question of who was. Read More

3. Trouble at the top: Lucid Motors's CEO Peter Rawlison is stepping down from his role as CEO and CTO. COO Marc Winterhoff will serve as interim CEO. Read More

Morning must-reads

Image Credits: Vicente Méndez / Getty Images

A new kind of SaaS: Y Combinator posted on X showing off what its latest startup Optifye.ai does. But the accelerator quickly took the post down after it became clear people really didn't love the fact that the startup makes performance monitoring software for manufacturing companies. Read More

Mistakes happen: OpenAI has released a clarification about the whitepaper it released last week. It said the paper was "incorrectly worded" regarding its work on persuasion research and said the research is separate from its plans to release a deep research model. Read More

Cheap thrills: It turns out that Anthropic's Claude 3.7 might not have been as expensive to train as everyone thought. It cost only "a few tens of millions of dollars," which, in the world of AI, is quite cheap. Read More

Let's see how AI does at fantasy character names: ElevenLabs is inviting authors to try out its latest feature: creating and publishing AI-generated audiobooks. The company raised $180 million last month and just announced a partnership with Spotify for AI audiobooks, too. Read More

Fix your security already: One of Australia's top fertility providers, Genea, has suffered a data breach, and a hacker group has published the private medical and personal data of its IVF patient. Read More

AI for due diligence: Bridgetown Research has raised $19 million from Accel and Lightspeed Ventures to use AI agents to conduct data collection and research and cut down the due diligence timeline and costs. Read More

Around the web

The beef continues: Spotify's CEO Daniel Ek is still very mad at Apple. He says Apple has been flaunting some of the DMA rules in the EU and wants the European Commission to hold the company accountable. Bloomberg has the rundown on the drama. Read More

Get them young: The country of Estonia has teamed up with Anthropic and OpenAI to teach AI to high schoolers, the Financial Times reports. The hope is to better prepare the kids for future jobs. Read More

Before you go

The internet never forgets, and it appears neither do AI models. Security researchers have found that AI chatbots hold on to exposed internet data even after the data is made private. The researchers found thousands of GitHub repositories from some of the world's biggest companies in Microsoft Copilot that had since been taken private. Read More