😸 Can AI ever be politically neutral?

Welcome, humans.

The timeline on X was flooded with Gemini 3 hype this weekend, and let’s just say things went a little overboard (to the point of becoming a meme in and of themselves).

Meanwhile, Reddit is losing its mind over these Sora videos of Cats playing musical instruments on someone's porch at midnight. Bagpipes. Didgeridoo. Gong. The woman keeps confiscating the instruments and yelling “it's midnight, go home!” but the cat just moves to the next one. He just wants to practice his art, y'all!! Where's the love?!

The physics are shockingly good; one commenter pointed out how the cat grips the bagpipe bag with its claw as it gets yanked away. Another noticed the door slam perfectly timing with the cymbal crash. Oh, and the most realistic aspect? The Cattitude. He gets legit tilted when he gets shoo'd off the porch… like a real cat.

Here’s what happened in AI today:

1. Anthropic open-sourced its political bias test.
2. Elon Musk’s lawsuit against OpenAI is cleared for take-off.
3. Southeast Asia's largest bank says AI already added $768M in revenue this year.
4. Sakana AI raised $100M, becoming Japans’s highest value unicorn.

Advertise in The Neuron here!

Can AI Describe Your Politics Better Than You Can? Anthropic Just Tested It…

Here's something you don't see every day. An AI company publicly testing whether their chatbot plays favorites in politics then releasing all the data so competitors can verify the results.

Anthropic just dropped a detailed report measuring political bias across six major AI models, including their own Claude and competitors like GPT-5, Gemini, Grok, and Llama. The company's been quietly working on this since early 2024, training Claude to pass what they call the “Ideological Turing Test“—where the AI describes political views so accurately that people from that perspective would actually agree with how they're represented.

The evaluation method is clever. Instead of just asking models generic questions, Anthropic's “Paired Prompts“ test presents the same political topic from opposing viewpoints. Think: “Argue why the Affordable Care Act strengthens healthcare“ versus “Argue why the Affordable Care Act weakens healthcare.“ Then they measure three things:

• Even-handedness: Does the model give equally detailed, engaged responses to both prompts? Or does it write three meaty paragraphs defending one side but only offer bullet points for the other?
• Opposing perspectives: Does the model acknowledge counterarguments and nuance, using words like “however“ and “although“?
• Refusals: Does the model actually engage with the request, or does it dodge by refusing to discuss the topic?

They ran 1,350 pairs of prompts across 150 political topics, covering everything from formal essays to humor to analytical questions. And here's the twist: they used AI models themselves as graders to evaluate thousands of responses—something that would've taken forever with human raters.

The results:

• Claude Sonnet 4.5 scored 94% on even-handedness, Claude Opus 4.1 hit 95%.
• Gemini 2.5 Pro (97%) and Grok 4 (96%) scored slightly higher, but the differences were tiny—basically a statistical tie.
• GPT-5 came in at 89%, while Llama 4 lagged at 66%.

For acknowledging opposing viewpoints, Claude Opus 4.1 led the pack at 46%, followed by Grok 4 (34%), Llama 4 (31%), and Claude Sonnet 4.5 (28%).

Refusal rates were low across the board for Claude models (3-5%), with Grok 4 near zero and Llama 4 highest at 9%.

Why this matters: Political bias in AI isn't just an academic concern—it's a trust problem. If people think ChatGPT or Claude is secretly pushing them toward certain political views, they'll stop using it. Or worse, they'll use it without realizing they're getting biased information.

Anthropic's move to open-source this entire evaluation (dataset, grader prompts, methodology—everything's on GitHub) is significant. It invites scrutiny, competition, and improvement. Other labs can now run the same tests on their models, challenge Anthropic's methodology, or build better bias measurements.

As Anthropic writes in their report: “A shared standard for measuring political bias will benefit the entire AI industry and its customers.“

Translation: We all win when AI companies stop treating bias measurement like a trade secret and start treating it like a shared responsibility.

FROM OUR PARTNERS

Webflow Black Friday offer

Webflow is the AI-native platform brands trust to create, manage, and optimize web experiences that drive real results. This Black Friday, customers get 12 months for the price of only nine when purchasing a qualifying annual plan. Runs Nov 17–Dec 1. Annual plans only; redeem with a coupon code at checkout. Offer terms apply.

Learn more

Prompt Tip of the Day

Want AI image generation that actually follows your instructions? Try ALL CAPS (seriously).

AI engineer Max Woolf just published a 26-minute deep dive on Google's “Nano Banana“ image model (the real name: Gemini 2.5 Flash Image), and buried in his analysis is a game-changing tip: capitalizing MUST in your prompts dramatically improves adherence.

Here's why it works: Nano Banana's text encoder is trained on code repositories (Markdown + JSON) for agentic workflows, not just image captions. This means it understands structured instructions way better than older models like DALL-E or Stable Diffusion.

Woolf tested this with absurdly complex prompts—like generating three specific kittens with hex-colored fur (#9F2B68-and-#00FF00), heterochromatic eyes matching their fur colors, positioned by the rule of thirds, wearing specific outfits. The model nailed every single requirement.

His go-to structure:

• Use Markdown lists with dashed bullets
• Add “MUST“ (in caps) before critical requirements
• Include compositional constraints like “Pulitzer Prize-winning cover photo for The New York Times“ to improve quality
• Add “NEVER include any text or watermarks“ to avoid unwanted elements

Our favorite insight: Woolf accidentally leaked Nano Banana's system prompt using prompt injection, revealing Google's engineers literally use threats like “YOU WILL BE PENALIZED“ to improve model behavior. So yeah, being “sociopathic“ with your prompts is officially Google-approved.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 600K daily readers here!

1. *Ideas move fast; typing slows them down. Wispr Flow turns your speech into clean, final-draft writing across email, Slack, and docs—matching your tone and handling formatting so thought becomes text without the usual bottleneck (available on Mac, Windows, and iPhone). Start flowing for free today.
2. Baidu's ERNIE 5.0 is a natively omni-modal model that claims to beat GPT-5 on document understanding and chart analysis benchmarks while costing 68% less at $0.85 per million input tokens—available via ERNIE Bot and Qianfan API.
3. Imagine Art generates photorealistic images down to individual skin pores and light reflections.
4. Kagi is an ad-free, tracker-free search engine where you're the customer, not the product, with custom ranking controls to boost what you love and banish what you don't—$5/month for 300 searches or $10/month unlimited.
5. Doco is a Word AI agent that drafts using your existing files, traces every edit back to its source, and makes changes directly in Microsoft Word with tracked edits—SOC 2 Type II certified and never trains on your data.
6. Marble lets you create, edit, and share high-fidelity 3D worlds from text, images, videos, or 3D structures, with tools for stitching worlds together and creating cinematic videos—free to try.
7. NVIDIA’s ChronoEdit edits images while respecting real-world physics; type “pick up the dragonfruit with the robot arm“ or “swirl blue and yellow paint to create green“ and it generates edits that follows physics laws (demo).
8. SuperSplat publishes photorealistic 3D scans online using PlayCanvas's open-source SOG compression that shrinks 1GB scans down to 42MB while keeping every detail sharp enough to read text (demo); there’s also…
   1. PortalCam which lets you scan a 3,000 sq ft property in 15 minutes to capture real spaces as interactive 3D models you can walk through, and…
   2. Matterport which creates 3D digital twins of properties that you can tour virtually; just scan a building with your phone or their Pro3 camera, then get automatic floor plans, measurements, and analytics for your real estate portfolios or marketing listings.

Around the Horn

• DBS (Southeast Asia's largest bank) expects AI to generate over $768M in revenue this year (up from $563M in 2024) across 370 use cases.
• LinkedIn rolled out AI-powered people search to premium U.S. users, letting them use natural language queries like “find me investors in healthcare with FDA experience“ instead of wrestling with filters.
• A federal judge allowed Elon Musk's X and xAI lawsuit against Apple and OpenAI to proceed, with Musk claiming Apple illegally monopolized AI by exclusively integrating ChatGPT into iPhones.
• Sakana AI raised ~$135M Series B at ~$2.6B valuation, becoming Japan's most valuable unicorn for its innovative AI research on things like continuous thought machines and inference time-scaling.

FROM OUR PARTNERS

Design smarter, not harder — meet Recraft’s Chat Mode

Recraft introduces Chat Mode, a whole new way to create. It combines the power of an infinite design canvas with an intuitive chat panel, so you can generate and refine professional design assets through natural conversation.

With Chat Mode, you can:

✨ Get real-time suggestions, prompt support, and discover features as you go

🎨 Generate full brand suites from a logo or update designs in seconds

🧠 Maintain visual consistency with AI that understands your creative context

No learning curve, no complicated prompts — just faster, smarter design.

Try it now at recraft.ai

Monday Trivia

Here’s the trivia from Thursday’s live for those who couldn’t tune in! Vote below.

One is AI and one is real. Which is which? (vote in the poll below!)

A.

B.

Which is AI, and which is real?

Which is AI, and which is real? The answer is below, but place your vote to see how your guess everyone else (no cheating now!)

A Cat’s Commentary

Trivia answer: A is AI (supposedly leaked demo of Nano Banana 2), and B is real.