Good morning everyone,
This week: Anthropic and Mistral secured massive funding rounds, OpenAI researchers explained why language models “hallucinate,” and Geoffrey Hinton — the “Godfather of AI” — raised fresh warnings about existential risks. Plus: regulation, market shifts, and real-world use cases that matter for strategy.
Why language models hallucinate
Large language models are improving fast — but hallucination remains a stubborn flaw. OpenAI researchers traced the cause to how models are trained: they’re rewarded for giving an answer rather than saying “I don’t know.” This incentivizes confident mistakes over honest uncertainty.
Their proposed fix is to adjust training and evaluation so models are rewarded for flagging uncertainty. That won’t make them infallible — but it could make them more transparent and trustworthy, especially in high-stakes contexts.
Business & Market Moves
Mistral’s leap forward: French AI startup Mistral raised €2 billion, with ASML becoming its largest shareholder, valuing the company at about €10 billion (~$11.7 billion). A good moment to review what they’ve been building.
Anthropic scales up: Anthropic closed a $13 billion Series F, pushing its valuation to $183 billion and cementing its role as an enterprise AI leader.
OpenAI moves into jobs and training: OpenAI announced plans for an AI-powered jobs platform and certification program, developed with Walmart, aiming to reskill 10 million Americans by 2030.
Healthcare AI: From Hype to Impact
Cancer care efficiency: In the UK, “Osairis,” co-developed with Microsoft, has cut radiotherapy planning from hours to minutes, freeing up clinician time.
Voice AI in GP practices: Transcription tools are easing documentation and improving appointment efficiency.
Adoption rising fast: Two-thirds of physicians now use AI tools, up from 38% last year.
Why this matters: AI could reduce administrative costs and improve accuracy in claims, underwriting, and customer service. Deeper collaboration with healthcare AI firms may unlock new efficiencies. Here’s what providers say they want from AI.
Strategic Insight & Leadership
Crafting AI strategies that deliver value: A recent HBR piece argues that AI creates value only when tied to business goals such as efficiency gains, differentiation, or new pricing models.
Building the AI-enabled enterprise: An MIT Technology Review briefing shows how leading firms are embedding AI into operations with interoperable systems, cross-functional teams, and strong governance. The biggest payoff comes from moving beyond pilots to scalable, repeatable models.
Takeaway: Evaluate AI investments against tangible outcomes — cost savings, fraud detection accuracy, underwriting speed — not novelty.
Opinion
In a Financial Times interview, Geoffrey Hinton — Nobel Prize and Turing Award laureate, often called the “godfather of AI” — warned that AI’s greatest risks stem from profit motives and geopolitics, not science fiction. He predicts superintelligence could emerge within 5–20 years, triggering mass unemployment, wealth concentration, and weaponization.
He was scathing about leadership — likening the choice between Sam Altman and Elon Musk to “being shot or poisoned” — and argued that Washington’s focus on “beating China” overlooks broader existential dangers.
He closed with humility:
“We don’t know what is going to happen […] something amazing is happening, and it may be amazingly good, and it may be amazingly bad.”
Useful Links & Tools
Kick off with 3Blue1Brown — a brilliant math explainer on YouTube. His video Large Language Models explained briefly makes model fundamentals easy to grasp. For those keen to go deeper, his full neural network course is a visually rich introduction — essential for cutting through AI hype.