# Why Your 'Genius' AI Can't Count Apples (And How Researchers Are Fixing It)

Page: https://stenobird.com/podcast/ai-with-shaily-7095384/why-your-genius-ai-can-t-count-apples-and-how-researchers-are-fixing-it
Text version: https://stenobird.com/podcast/ai-with-shaily-7095384/why-your-genius-ai-can-t-count-apples-and-how-researchers-are-fixing-it.md
Podcast: [AI with Shaily](https://stenobird.com/podcast/ai-with-shaily-7095384)
Published: 2026-05-10T07:01:05+00:00
Episode link: https://soundcloud.com/shailendra-kumaar/why-your-genius-ai-cant-count
Audio file: https://feeds.soundcloud.com/stream/2317915610-shailendra-kumaar-why-your-genius-ai-cant-count.mp3
Processing state: not_requested
JSON: https://stenobird.com/v1/public/podcasts/ai-with-shaily-7095384/episodes/why-your-genius-ai-can-t-count-apples-and-how-researchers-are-fixing-it
Duration seconds: 148

## Resource

Ever wonder why your genius AI can explain quantum physics but struggles to count apples in a photo? 🍎 This surprising blind spot in Vision-Language Models is being tackled by Hof University researchers with SITUATE, a new dataset of synthetic 3D scenes designed to specifically train AI on basic counting. They've found artificial data more effective for these logic skills than real-world images. 🤖 Shailendra Kumar reveals why AI relies on internet shortcuts over visual evidence and shares a bonus tip to force closer image analysis, ensuring accurate results. Gain unique insight into AI's limitations and current solutions. 💡 Have you caught AI making silly counting mistakes? Drop a comment below and let's discuss! Like and follow for more cutting-edge AI insights. 👇

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/ai-with-shaily-7095384/episodes/why-your-genius-ai-can-t-count-apples-and-how-researchers-are-fixing-it/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/ai-with-shaily-7095384/why-your-genius-ai-can-t-count-apples-and-how-researchers-are-fixing-it.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.