DeepMind Alumni Beat OpenAI at Science, Waymo Builds Its Own Chips
A new AI agent outperforms OpenAI and Anthropic at replicating research, Waymo goes custom silicon, and the copyright mess gets messier.
Subscribe free →DeepMind Alumni Built an AI That Beats OpenAI and Anthropic at Science
British startup Inherent released Faraday, an AI agent designed to replicate scientific research papers — and it outperformed models from both OpenAI and Anthropic on that task. Founded by DeepMind alumni, the company is pitching Faraday as an AI "teammate" that can accelerate research by faithfully reproducing and building on existing work. Replication is the unglamorous backbone of science, and automating it well is harder than it sounds.
Why it matters: If AI agents can reliably replicate research at scale, the bottleneck in scientific progress shifts from "can we reproduce this?" to "what do we build next?"
Waymo Is Building Custom Silicon to Power Its Robotaxi Fleet
Waymo isn't just operating self-driving cars anymore — it's designing the chips that run them. The company has developed custom silicon specifically tuned for its robotaxi ambitions, a move that mirrors what Apple, Google, and Tesla did when off-the-shelf hardware stopped cutting it. Owning the chip means tighter control over performance, cost, and the sensor-to-decision latency that determines whether a robotaxi is safe or not.
Why it matters: Vertical integration into chip design signals Waymo is treating this as a long-term infrastructure play, not just a software problem — and it raises the competitive moat considerably.
Training AI on Copyrighted Books: Still Technically a Legal Gray Zone
Authors didn't consent to having their work used to train AI models, and many argue this is clear copyright infringement. But courts are still untangling whether scraping published books for training data constitutes "transformative use" under fair use doctrine — and the answer isn't obvious. Lawsuits are piling up, but no definitive ruling has landed yet.
Why it matters: Whoever wins in court will effectively set the rules for how AI companies can build foundation models going forward, making this one of the most consequential IP battles in tech history.
Flock Safety's CEO Wants "Compromise" — Critics Want Accountability
Flock Safety, which sells AI-powered license plate readers and surveillance cameras to law enforcement agencies across the US, is facing a mounting public backlash over potential misuse of its tech. CEO Garrett Langley is calling for compromise rather than outright bans, but critics argue the company has already enabled a surveillance infrastructure with inadequate oversight. The backlash is sharpening as awareness of how widely Flock's cameras are deployed sinks in.
Why it matters: Flock is one of the largest private surveillance networks in the US, and how this fight plays out will shape the regulatory future of AI-assisted law enforcement tools.
Harvard Is Selling Access to AI Clones of Its Business School Professors
HBS Foundry, a $699 startup bootcamp from Harvard Business School, now features AI avatars modeled on its instructors. The avatars give real-time feedback during practice pitches and mock board meetings, letting students get coaching on demand rather than waiting for a human to be available. It's part simulation, part digital faculty clone — and it's shipping now.
Why it matters: When one of the world's most prestigious institutions starts replacing instructor time with AI avatars, it signals where the edtech market is heading at scale.
Quick Hits
- →The DOJ is reportedly investigating a16z over startup board seat practices — implications for VC governance and AI portfolio companies worth watching. TechCrunch
- →Walmart's Flipkart quick-commerce arm is now processing 1.1–1.2M orders per day in India, nearly tripling its November volume — AI-driven logistics at work. TechCrunch
AI TLDR