TOGETHER WITH POPHATCH

Hello again, human brain, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

• $24.5 Million Buys You AI's Eternal Silence 🤫

  Google's AI is silent on Trump's mental health.

• Claude Just Worked a 30-Hour Shift 💻

  Anthropic’s Claude 4.5 Sonnet is here.

  New here? Subscribe! 😎

Peep today's Spot the AI at the bottom. 👇

MAIN COURSE

Google AI & Trump. Sitting in a tree 😍

Google's AI won't answer questions about Trump's mental health. But will for other presidents. The inconsistency is raising eyebrows.

What's happening?

When you search “does trump show signs of dementia” in Google, AI Overviews won't generate a response. AI Mode just shows 10 web links instead of its usual summarized answer.

Does it do this for other presidents?

Nah. Search the same question about Biden, and AI Mode gives you a summarized response. Search about Obama, and you get an AI Overview stating there's “no public evidence” he has dementia.

What about other Trump mental health queries?

All blocked. Searches for Trump and dementia, Alzheimer's, or senility produce no AI overview. Just links.

Why is Google doing this?

Google didn't explain. They only said, “AI Overviews won't show a response to every query.” But given that Trump just settled a $24.5 million lawsuit with Google over his banned YouTube account, the timing is suspicious.

What's the problem?

The inconsistency. If you're going to block AI answers for sensitive health questions about public figures, do it for everyone. Not just the one who sued you.

SPONSORED BY POPHATCH

Your AI cofounder, who’s done this 10,000 times 🤖

I was up until 2am researching competitors. Again.

After a full day of coding, I was trying to be ten different people at once. Founder, product manager, strategist, market researcher.

I needed a cofounder, but finding the right person takes months and means giving up equity.

Then I found PopHatch. It’s an AI cofounder that handles market intelligence, competitive analysis, GTM strategy, and scaling decisions 24/7.

Now I build while my cofounder handles the rest. For FREE for the next month.

SIDE SALAD

Claude just took your coding job 💻

On Monday, Anthropic released Claude Sonnet 4.5. They claiming it's the best coding model in the world. The AI reportedly maintained focus on complex tasks for over 30 hours.

What's new?

Claude Sonnet 4.5 is Anthropic's latest mid-range AI model. It scored 77.2% on SWE-bench Verified (a real-world coding test), beating OpenAI's GPT-5 Codex (74.5%) and Google's Gemini 2.5 Pro (67.2%).

What's the 30-hour claim about?

Anthropic says Sonnet 4.5 worked continuously on the same project for over 30 hours without losing coherence. Previous AI models typically accumulate errors and lose focus over long periods as their memory fills up.

What can it do?

It excels at coding, computer use (navigating websites, filling spreadsheets), and multistep tasks. Anthropic also released Claude Code 2.0, a command-line tool for developers, and an SDK for building custom AI coding agents.

How much does it cost?

Same as Claude Sonnet 4: $3 per million input tokens, $15 per million output tokens. It's available now through the Claude API.

What else changed?

Anthropic claims Sonnet 4.5 shows reduced “sycophancy.” It’s the tendency to praise users' ideas even when they're wrong or dangerous. It also added code execution and file creation directly in chat.

More on Claude’s sycophancy, here.

Should we trust the benchmarks?

As always, take self-reported benchmarks with skepticism. But veteran developer Simon Willison says it “felt like a better model for code than GPT-5-Codex.”

The AI coding wars just got more competitive.

YOUR DAILY MUNCH

Think Piece 🧠

Photocurable 3D printing. It’s a 3D printing technique makes printed structures more robust and economical, traditional methods.

Startup News 💰

Meta reportedly acquired Rivos. It’s an AI chip startup which developed a chiplet-based SoC with GPU optimized for LLMs and data analytics.

Research 👨‍🔬 

LongLive — a real-time AI system that generates interactive long videos (up to 240 seconds) by letting users input sequential prompts during generation.

QAE — (Quantile Advantage Estimation) stabilizes AI model training by replacing the standard mean reward baseline with a group-wise K-quantile baseline.

MEME FOR DESSERT

SPOT THE AI

3 of these are real kittens. 1 is AI. 🐈

Which one is AI-generated? 👇

Ideas? Comments? Complaints?

Respond to this email or hit me up on 𝕏.

Until next time 🤖😋🧠