Welcome back to The Prompt.

In a surprising move, Devin— after the company’s cofounders Varun Mohan and Douglas Chen and other staff members were poached by Cognition gets access to what’s being called the which includes the remaining talent, its coding tool Windsurf Editor, its IP and its training data, Cognition President Russel Kaplan said on X.

For most people, the move came out of nowhere. Only a few weeks ago, OpenAI was in talks to acquire Windsurf for $3 billion, with an aim to compete with AI coding tools like Cursor. But that deal was a major sticking point between OpenAI and its backer and partner Microsoft, which owns the billion dollar gorilla in the world of coding tools, GitHub Copilot, The Wall Street Journal reported.

The entire debacle represents just how rapidly the as well as the who are well versed in AI. Vercel CEO Guillermo Rauch tells me the excitement around AI has led to big tech companies like Meta falling over themselves to acquire that talent. “A good AI engineer, for example, knows what the tricks are so that you can balance the load [of GPUs],” he says, adding that there will be a with.

The Windsurf acquihire and subsequent acquisition is the latest in poaching in the: Adept’s cofounders were hired by Amazon, Google shelled out a large sum to poach Character AI’s former CEO and prominent AI researcher Noam Shazeer and Inflection’s Mustafa Suleyman is now heading consumer AI products at Microsoft. With near-infinite resources, big tech companies can easily get their hands on the most savvy AI engineers (and therefore AI capabilities) while avoiding regulatory scrutiny. It’s , both of whom usually get adequate payouts, but left behind, whose work was crucial but end up reaping little reward and the startup ecosystem at large.

Let’s get into the headlines.

It’s been a wild, whirlwind month for Elon Musk’s xAI. Last week, the AI startup’s chatbot on X (formerly Twitter) in which it began calling itself “MechaHitler” and praised Nazi leader Adolf Hitler. Multiple users noticed that xAI’s latest AI model, Grok 4, consulted Musk’s social media posts on various controversial topics before answering questions about them. Musk claimed the posts were the result of manipulation by users. After the controversy, the company started deleting the offensive posts. The startup seems to be shaking off Grok’s hate speech episode: Musk said that with the Department of Defense, that it will soon be coming to Tesla cars, and has also released its own AI companions.

After losing several researchers to Meta,, Wired reported. The new hires are reportedly joining OpenAI’s scaling team, which manages hardware and software systems as well as data centers like Project Stargate. In the past few months, the battle for top talent has escalated, not only with major acquihires as I noted above, but also offers of for employees with rare technical knowhow.

, which develops software that helps developers to quickly build generative AI-powered apps, is at a $1.1 billion valuation, Forbes reported last week. IVP is leading the round, four sources familiar with the deal said. A Forbes AI 50 listee, LangChain boasts usage from some 40,000 teams at enterprises like Uber and LinkedIn.

Also notable: Thinking Machines Lab, founded by OpenAI’s former CTO Mira Murati, round led by Andreesen Horowitz, at a $12 billion valuation.

The Doctor Whisperer- Daniel Nadler

For doctors trying to stay abreast of the latest medical breakthroughs, reviewing the latest research is like being shot in the face with a water cannon. A new paper is published every 30 seconds. Trying to comb through it all to come up with a diagnosis or treatment plan that reflects the best current options while seeing 20 patients a day is a near-impossible task.

“There is this enormous firehose of information they need to stay on top of, and the human brain is limited in its capacity to read millions of studies,” Daniel Nadler, cofounder and CEO of OpenEvidence, told Forbes.

So Nadler, 42, a Harvard Ph.D. who sold his previous company for $550 million back in 2018, set out to solve the problem with artificial intelligence. Now, the startup’s proprietary algorithms search millions of peer-reviewed publications, including in top journals like the New England Journal of Medicine and the Journal of the American Medical Association, to help doctors find the best answers fast, with full citations to papers so doctors can read more for themselves. The software is free for verified doctors to use and makes money through advertising–much like Google does.

“I think OpenEvidence looks like it’s going to be for healthcare what Google was for the Internet,” said Kleiner Perkins billionaire chairman John Doerr, who invested in the company personally as well as through his firm, adding, “It’s the free-for-physician model that’s the magic here.”

Since its founding in 2022, Miami-based OpenEvidence has signed up 40% of doctors in the United States, or more than 430,000, and is adding new ones at a current rate of 65,000 per month. Its revenue from advertising is now coming in at an annualized rate estimated at $50 million. That’s not huge, but thanks to the software’s rapid adoption, investors are betting big: OpenEvidence has raised $210 million led by GV (Google’s venture arm) and Kleiner Perkins at a valuation of $3.5 billion, up from $1 billion at its last financing in February, Nadler told Forbes. Other storied VC firms like Coatue, Conviction and Thrive Capital also invested.

The new investment makes Nadler, who owns roughly 60% of the company, a billionaire, with a net worth that Forbes estimates at $2.3 billion. Cofounder Zack Ziegler, the company’s 30-year-old chief technology officer, owns some 10% of the business, worth about $350 million. Nadler was able to hold on to such a large stake by being its first seed investor, putting in some $10 million of his own money before raising any VC funding.

Read the full story on Forbes.

WEEKLY DEMO

Artificial intelligence has led to an explosion of published research papers as scientists turn to ChatGPT and the like for writing and editing assistance. Now AI is also being used for peer reviewing papers— a development that has led to concerns among scientists about the quality of such automated peer review and how AI could be tricked to generate a biased review. Some researchers are also hiding secret messages in white text or a small font that instructs AI reviewers to generate positive feedback, Nature reported.

MODEL BEHAVIOR

In a study published in May, researchers were able to finetune leading large language models like OpenAI’s GPT-4o to generate responses that are harmful and unintended by the makers of the models. This is called “emergent misalignment,” where the model produces malicious content and acts deceptively. This behaviour can also be triggered by a random and undefined phrase or prompt.