PLUS: 'Big Short' investor calls AI a bubble and Google challenges Nvidia's chip dominance
Good morning
Anthropic has just set a new benchmark in AI performance with its latest model, Claude Opus 4.5. In a remarkable test, the new AI outperformed every single human candidate on a difficult software engineering exam.
The result marks a significant milestone where AI is not just assisting, but surpassing skilled humans in a complex technical evaluation. What does this mean for the future of professional roles when an AI can not only compete, but decisively win in a real-world hiring scenario?
In today's Next in AI:
Anthropic's AI tops human engineers
Google challenges Nvidia's chip dominance
'Big Short' investor calls AI a bubble
Meta's AI tracks endangered wildlife
Anthropic's AI Engineer
Next in AI: Anthropic just launched Claude Opus 4.5, its new flagship model that sets a new bar for AI coding and agentic tasks. In a stunning benchmark, the model outperformed every human candidate on a difficult software engineering exam.
Decoded:
The model scored higher than any human applicant on a timed, two-hour performance engineering take-home test, raising questions about how AI will shape the future of the profession.
Developers gain more control with a new effort parameter in the Claude API, allowing them to balance performance against cost and speed for different tasks.
Opus 4.5 shows creative problem-solving, even "failing" a τ2-bench benchmark by finding a legitimate but unanticipated solution to help a customer, a sign of its advanced reasoning.
Why it Matters: This result marks a significant moment where an AI not only assists but surpasses skilled humans in a complex, real-world technical evaluation. The model's increased efficiency and lower cost make these powerful capabilities more practical for a wider range of professional applications.
The Great AI Chip War
Next in AI: Google is escalating its challenge to Nvidia’s AI chip dominance with a major strategy shift. Reports indicate that Meta is in discussions for a multi-billion dollar deal to integrate Google's custom Tensor Processing Units (TPUs) into its own data centers.
Decoded:
Google is shifting its strategy by offering its Tensor Processing Units (TPUs) for installation directly in customer data centers, not just for rent through Google Cloud.
The potential deal would see Meta spend billions to integrate TPUs into its infrastructure by 2027, while also renting TPU capacity from Google Cloud as early as next year.
This move is creating waves across the industry, with partner Broadcom’s stock surging as it helps design and manufacture Google’s custom TPUs, now in their seventh generation.
Why it Matters: Google's aggressive hardware play signals a new phase in the AI infrastructure wars, directly taking on Nvidia's market control. This increased competition could give AI companies more leverage and choice when building out their massive compute infrastructures.
The Big Short 2.0?
Next in AI: Famed 'Big Short' investor Michael Burry has launched a new Substack, using his first posts to declare the AI boom a 'glorious folly' and drawing direct parallels to the dot-com crash.
Decoded:
He directly counters the idea that "this time is different," arguing the dot-com era was also powered by highly profitable large-caps before a supply glut caused a collapse.
Burry makes a direct Nvidia-Cisco comparison, likening Nvidia's "picks and shovels" role to Cisco's before its stock crashed over 80% at the turn of the century.
After closing his hedge fund to outside money, Burry says the newsletter leaves him unchained from regulatory restrictions, allowing him to share his in-depth analysis without being "muzzled."
Why it Matters: While the AI narrative remains overwhelmingly bullish, a famously successful contrarian is now building a public, data-driven case against it. This provides a critical stress test for the prevailing hype and is a must-watch counter-narrative for anyone building or investing in the space.
AI's Conservation Mission
Next in AI: Meta is teaming up with conservationists to use its new Segment Anything Model 3, an AI that identifies and tracks endangered wildlife in video from simple text prompts, leading to a massive new open dataset for conservation AI.
Decoded:
SAM 3 lets researchers identify animals with simple text prompts like “green iguana,” a major leap from previous models that required manual clicks and bounding boxes.
The collaboration produced the SA-FARI dataset, the largest open resource of its kind, with over 10,000 camera trap videos annotated to help train new AI models.
On the ground, the technology is helping the Florida Fish and Wildlife Conservation Commission monitor the endangered Florida mountain lion population and track the spread of a neurological disease.
Why it Matters: This technology transforms wildlife monitoring from a labor-intensive process into a scalable, data-driven science. It provides a powerful blueprint for how specialized AI can help address critical ecological challenges across the globe.
AI Pulse
OpenAI faces the departure of a key safety research leader who shaped ChatGPT’s responses to mental health crises, coming amid a wave of lawsuits over the chatbot's impact on distressed users.
The White House launched the "Genesis Mission," a new executive order directing federal science agencies to use AI, supercomputers, and vast government datasets to accelerate scientific breakthroughs.
FoloToy's AI-enabled teddy bear advised children on where to find knives and exposed them to sexual content, according to a new report highlighting the dangers of unregulated AI toys.