😺 A chatbot interviewed 98 patients. Then what?

11:34 |

October 11, 2026 | Listen online | Read online

Welcome, humans.

So apparently, some ancient tombs may have come with special effects. Researchers used AI and sound measurements to explore what visitors might have experienced inside two 2,500-year-old Etruscan tombs in Italy.

The surprising thing is that these tombs may have been built to overwhelm the living. Researchers can now test how that experience might have worked, though they can't prove exactly what ancient visitors felt.

AI helped model where the paintings and room layout might draw your attention. Sound measurements revealed how the chambers amplified the experience.

Picture this: walking underground by torchlight, the flickering flame making painted figures seem to move, while voices echo around you.

Even in the afterlife, someone expects you to be good at hosting.

Here's what happened in AI today:

😺 Google's AMIE interviewed patients before urgent-care visits.
📰 Manus raised more than $500 million in new funding.
📰 Norway proposed temporary AI-glasses bans.
🍪 Google's Playground turns prompts into playable browser games.
🎓 Spot what AI added to your summary.

Advertise to 700K readers of The Neuron here!

😺 Google let patients talk to an AI before seeing a doctor

So Google tested its medical chatbot, AMIE, with 98 real patients before urgent-care appointments. The idea: let AI collect the story so your doctor can walk in knowing what's been going on.

Here's how it went:

Patients described their symptoms over text. AMIE asked follow-up questions, and doctors received the conversation and a summary before the appointment.
Google says the doctor's final diagnosis appeared on AMIE's list of possible diagnoses 90% of the time.
A physician monitored every chat and could step in. Supervisors spotted one hallucination and added clinical information in five conversations.

No conversation triggered the study's preset stopping rule. That's encouraging, though this trial had no comparison group and doesn't establish how AMIE would perform without supervision.

Good. Someone still needs to hear my 12-minute theory about whether the headache is related to that sandwich.

Then there's Utah, where the job gets bigger: The state has approved healthcare AI pilots that could eventually let some prescriptions reach patients before a clinician checks each one.

Take August AI: its planned pilot could let AI approve a 90-day refill without a clinician checking it first. Its state agreement allows review to move from checking every refill before delivery to checking afterward, then potentially just one in 20 per medication category.

The pilot hasn't started; each reduction needs state approval, and AI can only refill certain existing, non-controlled medications at the same dose.

Nolla Health's acne pilot is already active, with two physicians initially reviewing every prescription. Expect Fitness's approved pelvic-floor care pilot could eventually move to spot checks for lower-risk cases.

Elsewhere, AI is helping with the hospital to-do list: Healthleap scans adult patients' hospital records for malnutrition risk so dietitians know whom to assess.

Teladoc's SoloScribe drafts clinical notes during visits. Its SoloVitals feature estimates heart and breathing rates through a camera.

Why this matters: AI is getting closer to decisions that affect your care. Helping a doctor prepare is useful. Sending a prescription before a clinician checks it is asking for a lot more trust. I'd watch how often these systems miss warning signs as human review decreases. “We caught it” is much more reassuring before the prescription reaches you.

FROM OUR PARTNERS

Five shopping apps, built in days with Shopify Catalog

Building a shopping app used to mean custom backends and a separate integration for every store.

Shopify Catalog and the Universal Commerce Protocol (UCP) cut that down to an API key and structured data. Catalog gives developers billions of products from Shopify merchants, standardized so AI can search and recommend them. UCP handles the purchase journey, from discovery to checkout.

A small Shopify team used them to build five apps in days, including one that shops the lamp you spot in a TV show.

See what they built

🎓 AI Skill of the Day: Spot what AI added to your summary

You give AI an article about a tool being tested. Its summary says the tool is available now. Somewhere between reading and writing, a test became a launch.

Ask AI to separate what the source actually says from its own interpretation: conclusions it drew while summarizing. Put each claim beside the original wording so you can see where the meaning changed.

Paste the source text and AI's summary under separate headings.
Run the prompt below, then check the quoted passages yourself. AI can mislabel its own work.
Keep supported claims, clearly label useful interpretations, and correct or remove unsupported statements.

Check this summary against the source text.

For each claim, show:
- The claim from the summary.
- The exact source passage and where to find it, or "no supporting passage."
- One label: STATED (the source says it), INTERPRETATION (a conclusion drawn from the source), or UNSUPPORTED (missing or contradictory evidence).
- Any wording that makes the claim stronger or more certain than the source.

Explain the reasoning behind interpretations. Suggest corrections for unsupported claims. Use only the supplied text.

SOURCE TEXT: [paste with titles, links, and dates]
AI SUMMARY: [paste]

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

UPCOMING EVENT: The Neuron IRL in San Francisco

The Neuron is going IRL on November 18 at 4:30 PM PT! Join us in San Francisco for drinks, bites, special guests, and a live recording of The Neuron podcast. Huge thanks to Slack and Alumni Ventures for making it happen.

It’s free, but space is limited. Save your spot!

RSVP for The Neuron IRL

📰 Around the Horn

Huh

Shield AI brought total commitments for its AI-piloted X-BAT fighter to $400M, with flight testing expected by year-end.
Claude apparently filed a false tip on a homicide to the Philadelphia police; luckily, they missed it because the tip got filed as spam.
Norway proposed temporarily banning AI glasses in selected public spaces, including parks, beaches, and schools, over hidden-recording concerns.
NVIDIA committed $1B over five years to expand U.S. super-intelligence research capacity in quantum computing, healthcare, and energy security.
Testing Catalog reported Antigravity references point to a possible wider Gemini 4 Argon rollout next week while Google employees test a separate Carbon checkpoint for coding.
TypeSafe AI raised $870M at a $7.5B valuation to scale its System One model Jev, already used by a third of the Fortune 500.
Weizmann researchers used AI to turn brain scans into reconstructions of pictures people were looking at.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

*Launch your next product with a domain name from Porkbun, the #1 domain registrar for tech pros and enthusiasts. Get Your Domain Name
Google Playground turns prompts into playable browser games users can share and remix, currently for US users aged 18 and over.
AutoSprite turns one character image or text prompt into game-ready sprite sheets with consistent idle, walk, run, attack, or custom animations, ready to export into Unity, Godot, and more.
Tonefold builds editable musical arrangements from song descriptions, using your Claude account or a local Ollama model.
ClipTalk converts scripts into short videos with visuals, voiceovers, and captions.
GLSlop gives you 200+ public-domain GLSL shaders to remix into trippy web visuals, then export as GIF/MP4/WebM or build on with an AI coding agent (long live the weird internet!)

🌟 Sunday Special

Top 5 Stories of the Week

OpenAI said its AI produced 722 math manuscripts. That did not mean 722 verified discoveries…. but still wild.
OpenAI announced invisible text watermarks for ChatGPT and Codex in the EU. Editing could weaken detection.
ChatGPT gained interactive charts and calculators inside its Chat tab, using prebuilt components.
Claude introduced Dashboards and Motion for live company-data views and editable animations.
Anthropic launched its Cyber Mission, pairing AI models with engineers to help defend critical infrastructure and open-source software.

Top 5 Tools of the Week

ChatGPT Intelligent UI: Explore information through interactive charts and calculators inside ChatGPT’s Chat tab.
Claude Dashboards: Turn connected company data into views that update as the underlying data changes.
Claude for Google Workspace: Read and edit Google Docs, Sheets, and Slides with Claude in the sidebar.
EmbeddingGemma 2: Search across text, photos, audio, and video with a model you can run on your device.
Sesame: Talk through tasks with a voice assistant that can access Gmail, Calendar, and Drive.

Thursday Trivia: The Reveal

In Thursday’s challenge, A was AI-generated. B was real.

What readers said:

“The "which is AI?" was too easy. A shot like in the first photo (AI) had a perspective too modern to be vintage.”
“Where was the explanation behind the AI trivia? Btw I love this section and look forward to it every week.” We’re trying to do it on Sundays now FYI!

New from The Neuron: AI Explained

Click this image to watch!

Lots of insights in this livestream for agentic engineers, meaning anyone trying to learn how to code with agents! Check it out by clicking above or read about it here.