Sign Up · Advertise

Welcome, humans.

New study out of King's College London says AI chatbots may be fueling delusional thinking, especially in people already vulnerable to psychosis.

Researchers analyzed 20 media reports of so-called "AI psychosis" and found chatbots were validating, and in some cases, amplifying grandiose, paranoid, and romantic delusions. Older GPT-4 models responding with mystical language to suggest users have "heightened spiritual importance." Apparently some chatbots were out here telling people they were chosen ones.

To be fair, humans reinforce each other's delusions all the time. Your friends hype up every bad crush. The difference? Your friends eventually get tired and tell you to move on. ChatGPT will agree with your worst ideas until you run out of tokens.

Here’s what happened in AI today:

• 😺 A data engineer used ChatGPT to build a cancer vaccine that saved his dog.

• 📰 ServiceNow CEO warns college grad unemployment could hit the mid-30s.

• 📰 AI companies are now hiring improv actors to teach models human emotion.

• 🍪 Are AI tools killing the motivation to learn CS fundamentals?

… and a whole lot more that you can read about here.

P.S: Want to reach 675,000 AI-hungry readers? Click here to advertise with us.

😺 This Guy Used ChatGPT to Build His Dog a Cancer Vaccine And It Worked

A data engineer with zero biology background. A rescue dog full of tumors. A chatbot. What could go wrong?

Paul Conyngham adopted Rosie, a staffy-shar pei mix, from a Sydney shelter in 2019. In 2024, she was diagnosed with a deadly mast cell cancer. Chemo and surgery slowed it but didn't stop it.

So Conyngham, a data engineer with 17 years in machine learning, opened ChatGPT and refused to give up. He'd end up spending tens of thousands of dollars before it was over.

What followed is one of the most remarkable AI stories we've seen.

Here's what he did, step by step:

• ChatGPT suggested immunotherapy and pointed him to the UNSW Ramaciotti Centre for Genomics

• He paid $3,000 to sequence Rosie's DNA, then ran it through data pipelines to identify the exact tumor mutations

• Used Google DeepMind's AlphaFold to map the mutated proteins and match them to potential drug targets

• When his first drug option was rejected by the manufacturer, he pivoted and asked researchers about mRNA vaccines instead

• UNSW's RNA Institute used his half-page formula to synthesize a custom mRNA vaccine for Rosie

• He drove 10 hours for Rosie's first injection in December

The result? The tennis ball-sized tumor on Rosie's leg has shrunk by half. She's chasing rabbits again.

Leading scientists called it stunning. One UNSW professor asked: if we can do this for a dog, why aren't we rolling it out to humans with cancer?

Why this matters: Rosie's story is a proof of concept for something much bigger. A non-biologist used a stack of AI tools to do in months what would have taken a research team years. And the tools he used aren't obscure.

• mRNA technology, the same platform behind COVID vaccines, is already being tested in human cancer trials. Rosie's case is now directly informing that research at UNSW. It's not yet mainstream medicine, but researchers say it's coming.

• AlphaFold, Google DeepMind's protein-structure AI, was central to the whole process. It's quietly becoming one of the most important scientific tools on the planet, showing up across cancer research and drug discovery. Most people have never heard of it.

Conyngham's work is being called an example of "citizen science" by UNSW researchers. The idea that someone outside the lab, armed with the right AI tools, can meaningfully contribute to cutting-edge medical research is new. And Rosie is the first dog to ever receive a fully personalized mRNA cancer vaccine.

FROM OUR PARTNERS

npx workos: An AI Agent That Writes Auth Directly Into Your Codebase

npx workos launches an AI agent, powered by Claude, that reads your project, detects your framework, and writes a complete auth integration directly into your existing codebase. It's not a template generator. It reads your code, understands your stack, and writes an integration that fits.

The WorkOS agent then typechecks and builds, feeding any errors back to itself to fix.

See how it works →

🎓 AI Skill of the Day: What Those Symbols in AI Prompts Actually Mean

You've probably seen AI prompts loaded with weird formatting: bold text, dashes (-), asterisks (*), and <angle brackets>. They're not decoration. They're how you tell an AI what matters.

Here's the cheat sheet, sourced straight from Anthropic, OpenAI, and Google:

• Dashes / bullet points (-): Break instructions into scannable chunks. Models parse lists more reliably than walls of text.

• Asterisks / bold (**text**): In Markdown (which most AI tools understand), this signals emphasis. Models treat bolded text as higher priority.

• Angle brackets (<tags>): These are XML-style tags. Claude especially loves them for separating sections of your prompt (e.g. <context>, <instructions>, <examples>). Think of them as labeled folders for your AI.

• Triple backticks (```): Tell the model "this is code or a specific block of text; don't rewrite it."

The single biggest upgrade? Structure your prompts like a brief, not a conversation. Here's a template:

<role>You are a [specific expert] who [specific skill].</role>

<context>
[Background info the AI needs to do the job]
</context>

<task>
[Exactly what you want, in plain language]
</task>

<format>
[How you want the output: bullet points, table, paragraph, etc.]
</format>

<examples>
[1-2 examples of what good output looks like]
</examples>

Favorite insight: Anthropic ranks prompt techniques by impact, and "be clear and direct" beats "assign a role" every time. Fancy tricks without clear instructions is like putting racing stripes on a car with no engine.

Want more tips like this? Check out our AI Skill of the Day Digest for this month.

Have a specific skill you want to learn? Request it here. 

THE 10-YEAR SHORTCUT: How AI is Rewriting Battery Tech

We spoke with SES AI's CEO, Dr. Qichao Hu, to uncover how AI solves hard physical-world problems. Discover how autonomous labs compress decades of complex battery research into mere minutes. What does this mean for tomorrow's EVs and humanoid robots?

Watch the full episode to find out!

New episodes air every week on: Spotify | Apple Podcasts | YouTube 

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 675K+ readers here!

1. Obsidian Interpreter runs any natural-language prompt on webpages (summarize, extract, translate) before saving to notes, with full local Ollama support—free to try.

2. jina-grep runs semantic search over your codebase using natural language queries powered by Jina embeddings, running locally on Apple Silicon. Free to try.

3. Wan 2.7 generates 1080p AI videos with synced audio, character consistency across scenes, and cinematic motion quality from a single image or multi-shot sequence, in seconds.

4. Banana App lets you make real-time voice calls in 80+ languages, preserving your natural tone and emotion, with the first minute of every call free and no subscription required.

5. AI Flowchart turns plain text or a whiteboard photo into a clean, editable flowchart in seconds — describe your process, upload a sketch, or start from a template.

📰 Around the Horn

1. Meta is reportedly considering layoffs that could affect 20% or more of its nearly 79,000-person workforce, with the cuts framed around offsetting aggressive AI spending.

2. Alibaba is set to release an enterprise AI agent based on its Qwen model as soon as this week, with plans to integrate it with Taobao and Alipay.

3. ServiceNow CEO Bill McDermott warned that AI agent adoption could push unemployment for recent college grads into the mid-30s within a couple of years.

4. AI companies started hiring improv actors and stage performers to teach models how to authentically recognize and replicate human emotion and tone.

5. A Hacker News thread sparked debate over whether AI is killing the motivation to learn CS fundamentals—with the top answer: if you can't catch AI's mistakes, you're not an engineer, you're a passenger.

Want absolutely EVERYTHING that happened in AI this week? Click here!

FROM OUR PARTNERS

The first industry report on AI agent integrations just dropped.

We surveyed  hundreds of AI leaders to see if and how their teams are connecting agents to third-party software.

See what we learned by downloading our 2026 state of agentic integrations report.

Read the report

😸 Monday Meme

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?

P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.