🤓California Reboots AI Oversight

Today we will discuss:

• California’s AI reset—New report outlines path forward after veto of sweeping AI safety bill

• Don’t miss this week’s Workflow Wednesday—Using AI tools to adapt, navigate change; now available at only $20/month! 

• Pokémon as the new AI benchmark—Researchers study how Gemini and Claude reason through classic gameplay and sometimes fail

• In other AI news — Meta is teaming up with Oakley and Prada for new smart glasses, Elon Musk’s xAI is seeking a $4.3B equity raise, Sam Altman says Meta tried luring OpenAI staff with $100M offers, and Max will use AI to power autoplaying video previews

• Saved the best for last—10 must-try AI tools

This Week in Workflow Wednesday: AI Adaptability

Learn a New Market in 30 Minutes Using AI
You’ve just been told to pivot into a new vertical. No research team. No prep time. Just you, a blank page, and a deadline.

Here’s how to go from clueless to confident - fast:

Step 1:
Open Perplexity.ai and ask:

“What’s the current state of [insert industry] in 2025? Top players, trends, pain points.”
Skim the results. Pull out key insights: names, numbers, and friction points.

Step 2:
Drop that intel into ChatGPT and prompt:

“Build a buyer persona based on this industry overview. Include their …..Read the full workflow (plus 2 others) in this week’s edition →

California’s new AI report replaces vetoed bill with updated guidelines

Key Points 

• California’s new AI report urges third-party testing and transparency instead of relying on internal company evaluations.

• With federal AI rules uncertain, California positions itself as a potential leader in responsible AI oversight.

📖Context of the news - Last year, California Governor Gavin Newsom vetoed SB 1047, a bill that would have required large AI models to undergo specific safety testing. 

While tech companies welcomed the move, AI researchers and whistleblowers raised concerns. In response, Newsom asked a group of experts to come up with a better plan one that balances innovation with accountability.

♨️News - The newly released California Report on Frontier Policy suggests a different approach. Instead of focusing only on how much computing power a model uses, the authors recommend looking at how these models are used in real-world settings.

They argue that compute is useful as an initial screen, but actual risk and downstream impact matter far more.

The report calls for independent, third-party safety evaluations, not just internal checks. It highlights how current systems, including OpenAI’s partnerships, still limit access to models and data, making robust assessments difficult. Without meaningful access, the report argues, even the best evaluators cannot detect real risks.

🤨Can California lead the way? With federal regulation on hold and a national moratorium being debated, the report says California has a chance to set the standard. But doing so will require more transparency, stronger whistleblower protections, and a willingness from companies to open their systems to outside scrutiny.

Read the full story here.

Gemini and Claude show panic while playing Pokémon games

Key Points 

• Gemini 2.5 Pro shows signs of "panic" during Pokémon battles, leading to visibly worse decision-making under pressure.

• Claude once tried to faint on purpose to skip a cave, misunderstanding how Pokémon Center mechanics actually work.

 👾Context of the news - In a curious turn for AI evaluation, researchers are using classic Pokémon games to test how well models like Google’s Gemini 2.5 Pro and Anthropic’s Claude handle problem-solving. What’s surprising is not their speed or success, but how emotionally erratic they can appear. 

🤖News - According to a new DeepMind report, Gemini 2.5 Pro exhibits something like “panic” when its Pokémon are low on health, often leading to a noticeable drop in reasoning.

Over the past few months, two independent developers launched Twitch channels called "Gemini Plays Pokémon" and "Claude Plays Pokémon," where the models’ decision-making is streamed for anyone to watch. Alongside the game footage, viewers can also see how each AI reasons through problems, revealing just how inconsistent and unpredictable they can be.

Claude once got stuck in Mt. Moon and wrongly assumed fainting on purpose would help it reach the next town faster. Instead, it was sent back to an earlier checkpoint. Gemini, on the other hand, has occasionally impressed. With only a brief description of boulder physics, it managed to solve difficult puzzles in Victory Road.

🎮Why this matters - These experiments may look silly, but they’re revealing. Watching AI “panic” or improvise under pressure is offering a strangely human window into machine behavior and where reasoning still falls short.

Read the full story here.