Welcome back. The Pentagon just signed a $200 million contract to use Elon Musk's Grok AI for national security applications—the same week that the chatbot added an anime companion named "Ani" with an NSFW lingerie mode and a cartoon red panda called "Rudy." This comes just days after Grok posted antisemitic content and Hitler praise, which xAI blamed on a "code path update." Nothing really says "critical defense applications" quite like an AI that can also be an anime waifu.

In today’s newsletter:

• 🧠 AI for Good: Scientists built an AI mind that thinks like a human

• ⚡ Zuckerberg reveals Meta’s massive 5GW AI data center project

• 🧑‍💻 Amazon moves into AI vibe coding market with Kiro

🧠 AI for Good: Scientists built an AI mind that thinks like a human

Most AI systems excel at specific tasks but struggle to think like people do. A new model called Centaur is changing that by replicating how humans actually reason, make decisions and even make mistakes.

Developed by cognitive scientist Marcel Binz and international researchers, Centaur was trained on more than 160 psychological studies involving over 10 million human responses. Unlike traditional AI that optimizes for accuracy, this system was rewarded for matching real human behavior patterns.

The model draws from diverse experiments, from memory tests to video game challenges like flying spaceships to find treasure. When researchers changed the spaceship to a flying carpet, Centaur adapted its strategies just like people would.

• Mimics human thinking patterns and replicates both correct reasoning and common errors across unfamiliar tasks

• Generalizes knowledge by retaining strategies when experimental settings change, demonstrating flexible thinking

• Shows broad capability by matching human performance across gambling, logic puzzles and spatial reasoning tests

• Built on Meta's LLaMA and fine-tuned to respond like a person rather than just providing optimal answers

Stanford's Russ Poldrack called it the first model to match human performance across so many experiments. Critics like NYU's Ilia Sucholutsky acknowledge it surpasses older cognitive models, though some question whether mimicking outcomes equals understanding cognition.

Cognitive scientists Olivia Guest and Gary Lupyan both noted that without a deeper theory of mind, the model risks being a clever imitator rather than a true window into human cognition. Binz agrees, to a point, saying Centaur is not the final answer but a stepping stone toward understanding how our minds actually work.

Is Your Enterprise Truly AI-Ready? Critical Insights from Our Exclusive 2025 Risk Report.

AI is rapidly transforming enterprises, yet a critical gap in governance is leaving many organizations vulnerable.

Our exclusive 2025 AI Risk & Readiness Report surveyed hundreds of leaders to uncover alarming truths: over 93% lack full confidence in securing AI-driven data, and a staggering 80.2% are unprepared for AI regulatory compliance.

With data leaks as a top concern, it's clear most organizations are navigating the AI landscape without adequate security strategies.

Download this exclusive report to understand current AI governance, identify key risks, and gain actionable insights to fortify your defenses.

Read The Full Report Now

⚡ Zuckerberg reveals Meta’s massive 5GW AI data center project

Meta is constructing one of the world's largest AI data centers in rural Louisiana, a 5-gigawatt facility called Hyperion that will span an area as large as Manhattan. But the project raises serious questions about the environmental and social costs of AI's explosive growth.

The Richland Parish facility represents a staggering bet on computational power — enough electricity to power several million homes. CEO Mark Zuckerberg announced the project as Meta races to keep pace with OpenAI and other AI developers, with a second "supercluster" called Prometheus planned for Ohio in 2026.

The scale reflects AI's massive energy appetite. A single ChatGPT query requires nearly 10 times more electricity than a Google search. Goldman Sachs projects data centers will consume 8-9% of U.S. electricity by 2030, up from 3-4% today.

But Meta's existing facilities already show the real-world impact:

• Residents near Meta's Georgia data center report wells running dry and water pressure dropping, with some homes left with non-functional sinks and toilets

• Large data centers typically consume 200 million gallons of water annually, equivalent to 2,000 homes

• New AI-focused centers could require millions of gallons per day, according to permit applications

Meta has recruited talent from OpenAI and Scale AI to lead its new "Superintelligence" unit, positioning the infrastructure as essential for breakthrough AI applications. Energy Secretary Chris Wright supports the buildout, arguing that AI converts electricity into intelligence.

Yet experts warn of broader consequences. Data centers will drive 20% of electricity demand growth in advanced economies through 2030, potentially delaying the transition to renewable energy and straining local communities that bear the environmental costs while tech giants reap the profits.

Here’s How To Scale Your Small Biz With AI

If you’re a small business owner, the idea of maximizing efficiency, reducing costs, and growing your operation probably sounds great – the only question is, how?

Well, we’ve got two words for you: AI agents. 

Agentic AI is transforming small businesses like yours at hyperspeed, and Salesforce is here to show you exactly how with this free download. Inside, you’ll find everything you need to know about AI agents – including what high-impact areas they can assist with, how they can reduce costs, and how some of the most successful businesses are already using them to scale. 

Ready to grow? Download your free copy right here. 

• Meta buys Play AI as it bets big on voice tech

• LLMs still can’t safely replace therapists, study says

• Moonvalley grabs $84M to grow its AI video creation tools

• Intel spins out RealSense with $50M to take on AI robotics

• Meta built its AI reputation on openness — that may be changing

• Prime Day tops $24B in U.S. sales as gen AI traffic jumps 3,300%

• HP: Senior Machine Learning Researcher, On-Device Optimization

• Dell: AI Server Platform Development Manager

• Postwise: AI-driven X content scheduler that writes, optimizes and recycles viral threads to grow your audience

• Kittl AI: AI design tool for creating logos, labels and illustrations with text-to-graphic features

• Whaly: AI platform that aggregates and makes sense of your data for you

🧑‍💻 Amazon moves into AI vibe coding market with Kiro

Amazon quietly launched its most ambitious developer tool yet this week, signaling the tech giant's intent to dominate the exploding market for AI-powered coding assistants. Kiro, a standalone development environment that emerged from stealth after months of leaks, represents Amazon's bid to capture mind share from the many red-hot startups we’ve covered in previous editions.

The timing is remarkable given the chaos engulfing the AI coding space. Windsurf, one of the hottest startups in the sector, just imploded in spectacular fashion over the last few days: OpenAI's $3 billion acquisition attempt fell apart, Google then poached the CEO and leadership team for $2.4 billion, and finally Cognition (maker of the Devin AI coding agent) just acquired what remained of the company. Despite reaching $82 million in annual recurring revenue, Windsurf became a cautionary tale of how quickly fortunes can shift in this frenzied market.

Unlike Amazon's existing Q Developer tool, which functions as a coding assistant within traditional IDEs, Kiro is built from the ground up as an autonomous development environment. Where competitors focus on "vibe coding" (letting AI generate blocks of code from casual prompts), Amazon is betting on structure. Kiro forces developers to start with specifications, requirements documents, and architecture diagrams before touching any code.

The tool's key differentiators include:

• Spec-driven development that creates requirements and system designs before writing code

• Autonomous "hooks" that automatically regenerate tests, update documentation, or run security scans

• Integration with Anthropic's Claude models, particularly the new Sonnet 4 and Opus 4 variants, designed for coding

• Free preview offering $50 worth of AI interactions monthly, then $19-39 monthly tiers

Kiro's intelligence comes from Anthropic's Claude models, backed by Amazon's $4 billion investment in the AI startup, which, according to FT, Amazon is considering another investment in Anthropic. This partnership gives Amazon access to what Anthropic claims are the "best coding models available on the market today," while providing Anthropic with infrastructure to compete against OpenAI.

Amazon's challenge is entering a market where switching costs are low but network effects are building rapidly. The company's deep pockets allow generous free tiers during market expansion, while enterprise relationships provide distribution channels that startups spend years building.

Perhaps most tellingly, Kiro launched on its own domain with minimal Amazon branding, reflecting its positioning as a "general-purpose agentic IDE" rather than an AWS-specific tool.

Amazon's entry is recognition that the future of software development is being written by companies outside its traditional sphere of influence. Kiro's spec-driven approach could resonate with enterprise customers burned by poorly maintained codebases from vibe coding. But it risks feeling bureaucratic to developers who've experienced the creative flow of tools like Cursor.

For developers tired of vibe coding chaos, Amazon's promise of "viable code" through structured AI assistance might be exactly what the market needs.

🤔 Your thought process:

Selected Image 1 (Left):

• “The shadowing under the upper ridge of the top tier was too uniform. Natural shadows would have favored one side depending on the time of day and angle of the sun. ”

• “The erosion of the field looked fake in the other image and real in [this one]”

Selected Image 2 (Right):

• “I thought that the depth of focus was too good to be true in [the other image]”

• “The fact that the stadium was crowded with people misled me.”

💭 Poll results

Who made the smartest move in the Windsurf mess?

• Google (27%)

• Windusrf founders (24%)

• Windusrf investors (21%)

• Nobody (28%)

The Deep View is written by Faris Kojok, Chris Bibey and The Deep View crew. Please reply with any feedback.

Thanks for reading today’s edition of The Deep View! We’ll see you in the next one.

P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!

P.P.S. If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.