Welcome back to The Prompt.
The second largest teacher’s union in the country is starting a new training center to train teachers how to use AI in the classroom, thanks to $23 million in funding from Microsoft, Anthropic and OpenAI. The American Federation of Teachers announced The National Academy for AI Instruction will train teachers on how AI can generate lesson plans and write emails to parents. Based in New York City, the Academy’s goal is to provide free training to 400,000 K-12 teachers over the next five years.
A string of universities like Arizona State University and California State University have already partnered with OpenAI to bring ChatGPT to their students. And the Trump Administration is encouraging educators to widely adopt AI within schools. But AI is hardly ready for prime time in education. Students regularly use ChatGPT to cheat on homework (as teachers once feared) and to write college admissions essays, and old forms of teaching and traditional curriculums simply aren’t suited to the moment. Worse, AI tools designed for education aren’t always fully safe for kids: A recent Forbes investigation found that the top AI study aid chatbots can be prompted to give dangerous advice and generate harmful content like fentanyl recipes.
Let’s get into the headlines.
POLITICS + ELECTION
AI is increasingly being used to impersonate government officials and obtain access to confidential information. The most recent example is an impostor who pretended to be Secretary of State Marco Rubio, using AI to generate voice and text messages in his likeness, then contacting three foreign ministers, a U.S. governor and a U.S. member of Congress on encrypted messaging app Signal, the Washington Post reported.
TALENT RESHUFFLING
As it builds out its shiny new superintelligence team, Meta has snagged top AI researchers from large AI labs like OpenAI, Anthropic and Google DeepMind with multi-million dollar offers. Its most recent addition is Ruoming Pang, who was previously in charge of developing AI models at Apple, Bloomberg’s Mark Gurman reported.
AI DEAL OF THE WEEK
Neocloud giant CoreWeave, which recently went public at a $23 billion valuation, announced it has acquired crypto miner and data center infrastructure provider Core Scientific in an all-stock deal valued at about $9 billion. CoreWeave, which provides prized Nvidia GPUs to large AI companies like OpenAI and Microsoft, has been leasing datacenters from Core Scientific since 2018. The acquisition will help CoreWeave eliminate rent costs for the next 15 years, CEO Mike Intrator said in an interview with CNBC. The deal also gives CoreWeave more power capacity, which is increasingly important to train and run powerful AI models.
DEEP DIVE
As longtime cancer doctors with regulatory experience, Pi Health cofounders Geoff Kim and Bobby Reddy knew that completing clinical trials took far too long. There was the painfully slow process of signing up patients and after that a grueling slog through vast swamps of data to prepare voluminous regulatory filings–something that few hospitals and clinics can handle. The pair knew their startup’s best chance of success meant doing an end run around all that. So they did something audacious and unprecedented: they built their own cancer hospital in India.
Clinical trials are an enormous bottleneck in drug development, and Kim and Reddy thought the AI-enabled software they’d been building at Pi Health could help do them faster and cheaper by expanding the pool of potentially eligible patients. But the majority of clinical trials today are done in top-notch academic medical centers, and first they needed to prove that their AI-enabled software could help overseas hospitals and smaller community cancer centers handle the documentation required to get through regulatory approval. So they found a site in Hyderabad, a major technology and pharmaceutical center in southern India, and built a 30-bed, state-of-the-art cancer hospital.
Only 8% of cancer patients in the U.S. participate in clinical studies, in part because of the voluminous paperwork required to run them. That limits understanding of the disease and the way that it affects diverse populations, and also means drug approvals take longer and cost more than they would if the limited pool of patients weren’t a bottleneck.
Pi Health’s software aims to lower the burden. It combines all clinical trial data into one place, streamlining workflows and reducing errors, starting with the trial design and continuing through regulatory submission. It uses artificial intelligence to check for discrepancies and errors in data and to produce automated notes with clinical documentation from regulatory-grade data. To date, the Cambridge, Mass.-based startup has raised some $40 million at a valuation of nearly $100 million. It is generating revenue, with signed contracts of more than $70 million.
Read the full story on Forbes.
WEEKLY DEMO
Free open source software is helping websites fend off the army of bot scrapers hoovering up their data to train AI models and running up their server costs in the process. Anubis, which has been downloaded 200,000 times and is being used by organizations like UNESCO, uses a browser that carries out a type of cryptographic math to verify that a site visitor is in fact a human and not a bot, 404 Media reported. Another clever way organizations are trying to thwart content scrapers: creating an “infinite maze” that continuously sends bots to useless links that aren’t used by humans.
MODEL BEHAVIOR
A band called The Velvet Sundown has exploded in popularity with more than a million monthly listeners and hundreds of thousands of streams on Spotify. But multiple Redditors noted that the band appeared to be “completely fake.” After initially denying that it used AI to generate its music, a spokesperson for the band told Rolling Stone that it used AI in the production of its songs. The spokesperson later said he was an imposter. The Velvet Sundown’s Spotify bio mentions that the band uses AI as a “creative instrument” to compose its music.
