Howdy! It’s Barsee again.

Happy Monday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

• Apple researchers find “reasoning” models collapse beyond complex problems

• Meta reportedly in talks to invest $10 billion in Scale AI

• OpenAI is retaining all ChatGPT logs indefinitely

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

GROWTH SCHOOL

Your Exclusive Invite for the World’s first 2 day AI Challenge (usually $895, but $0 today) 🤖

• 51% of companies have started using AI

• Tech giants have cut over 53,000 jobs in 2025 itself

• And 40% of professionals fear that AI will take away their jobs.

But here’s the real picture: companies aren't simply eliminating roles, they're hiring people who are AI-skilled, understand AI, can use AI, and can even build with AI.

Join the online 2-Day LIVE AI Mastermind by Outskill - a hands-on bootcamp designed to make you an AI-powered professional in just 16 hours. 

Usually $895, but for the next 48 hours, you can get in for completely free.

In just 16 hours & 5 sessions, you will:

• Learn the basics of LLMs and how they work.

• Master prompt engineering for precise AI outputs.

• Build custom GPT bots and AI agents that save you 20+ hours weekly.

📅 Kick off Call & Session 1:  Friday  (10am EST- 1pm EST)
🕜 Sessions 2-5: Saturday 11 AM to 7 PM EST ; Sunday 11AM EST to 7PM EST

Join in now, (we have limited free seats!)

_{*This is sponsored}

APPLE

Apple researchers find “reasoning” models collapse beyond complex problems 🧠⚠️

A new Apple study found that AI models like Claude 3.7 and DeepSeek-R1 often give up too quickly on tough logic puzzles, even when they could keep "thinking."

Here's what you need to know:

• Researchers tested AI models on logic puzzles of varying difficulty.

• Regular AI models did better than "reasoning-focused" ones on easy puzzles.

• On medium-level puzzles, specialized models (like Claude’s "Thinking" mode) performed well.

• But on very hard puzzles, all models failed, and reasoning models often stopped trying too soon.

Why this matters:

This research comes as AI companies stake their futures on improved reasoning abilities. If today's best models can't handle controlled logic puzzles, it casts doubt on their readiness for high-stakes applications like medical diagnosis or legal analysis. The findings suggest we may need entirely new approaches to achieve robust, scalable AI reasoning, not just bigger models or more data.

THROUGH THE VALLEY

1. Meta reportedly in talks to invest $10 billion in Scale AI

Meta is reportedly in talks to invest more than $10 billion in Scale AI, which would be its biggest AI investment so far and one of the largest private funding rounds ever. Scale AI, which provides training data to companies like Microsoft and OpenAI, made $870 million in revenue last year and expects to reach $2 billion this year. Meta had already invested in Scale's $1 billion Series F round, which valued the company at $13.8 billion. Scale also developed "Defense Llama," a military-focused AI model built using Meta's Llama 3 architecture.

2. AI startups are hitting $2M revenue in year one, doubling old benchmarks

AI startups are shattering old growth models, with companies like Lovable and Gamma hitting $50M in revenue within months and Cursor reaching $100M in its first year (all while raising minimal capital). According to Andreessen Horowitz, the traditional startup playbook is obsolete: enterprise AI startups now average $2M ARR in year one, and consumer AI apps are pulling $4.2M, bypassing the old “grow first, monetize later” model. Faster revenue cycles mean quicker funding rounds and rising investor expectations. Speed is now strategy: winners iterate fast, monetize early, and scale hard, creating a widening gap between breakout companies and those left behind.

3. Everything Apple plans to show at its iOS 26-focused WWDC 2025 Event

Apple’s WWDC 2025 starts today, with big updates expected for iOS, macOS, and more, including what could be the biggest visual redesign in years. Rumors suggest a new “Solarium” interface, inspired by visionOS, with a sleek "Liquid Glass" design and transparent, glass-like UI elements. Apple might also switch to naming software by year (e.g., iOS 26 instead of iOS 19). While last year focused on Apple Intelligence, this year’s AI news may be quieter as Apple fine-tunes Siri and works behind the scenes. The keynote begins at 10 a.m. PT, with everyone watching Cupertino.

OPENAI

OpenAI is retaining all ChatGPT logs indefinitely 📂💬

A federal judge has ordered OpenAI to retain all ChatGPT conversations (even deleted ones) for a lawsuit with The New York Times.

Here's what you need to know:

• The rule applies to most users (Free, Plus, Pro, Team, and standard API). Only Enterprise, Edu, and Zero Data Retention API customers are exempt.

• The saved chats will be locked for legal reasons and kept separate from training data. Only a small security team can access them.

OpenAI’s response:

• The company is fighting the decision, calling it an "overreach."

• CEO Sam Altman suggests new "AI privilege" protections, similar to doctor-patient confidentiality.

Privacy concerns:

• Critics warn this could expose sensitive chats (health issues, business secrets, etc.).

• It may especially affect vulnerable users and those in countries with strict privacy laws (like GDPR).

Why this matters:

This ruling could fundamentally reshape expectations around AI privacy. As chatbots become confidants, assistants, and creative partners, the decision forces a reckoning about what protections should exist for machine-mediated conversations. The outcome may determine whether users continue trusting cloud-based AI or shift toward local alternatives, potentially altering the trajectory of the entire industry.

TRENDING TOOLS

• Circuit Tracer: Anthropic’s open-source tools that reveal how AI thinks.

• Manus: A versatile agent that turns your thoughts into actions.

• Promptmonitor: Get your brand featured across ChatGPT, Gemini, and other AI/LLMs.

• Fieldy: A wearable AI note-taker designed for in-person meetings.

• Tyce: The AI-powered agent that instantly personalizes and powers your documents using your company’s knowledge.

• Semilattice: Leverage an AI model of your audience for instant answers.

• Higgsfield Speak: Create motion-driven talking videos with an avatar and script using AI.

• Runner H: Execute entire workflows across web apps, documents, and spreadsheets with a single prompt.

THINK PIECES / BRAIN BOOST

• Vibe coding shifts power dynamics in Silicon Valley.

• The 3 AI use cases: Gods, Interns, and Cogs.

• AI Fluency by Anthropic.

• Trends - Artificial Intelligence (May 2025).

THE VALLEY GEMS

What’s trending on social today:

1/ AI SaaS is blowing up faster than ever.

2/ Smart take on AI and mental work.

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.