Demis Hassabis explains LLMs, public safety, and what comes next in AI

11 Jun 2025 · 17 min
16:48

A German-language transcript of a YouTube interview (uploaded June 11, 2025) between Demis Hassabis and a host discussing Large Language Models (LLMs), their inner workings, and societal implications. Hassabis explains LLMs as next-word predictors refined via pre-training, reinforcement learning, and a newer inference-time "deep thinking" (parallel/iterative reasoning). He describes ambitions toward multimodal "world models" (language, audio, images, video) to enable robust assistants and robotics, and highlights applications such as AlphaFold/Isomorphic Labs for speeding drug discovery. The conversation repeatedly returns to public concerns: privacy (XR glasses, data trust), job displacement (an expert estimate of 10–20% unemployment in ~15 years), and the need for responsible deployment, norms, and regulation.

Questions this video answers

June 11, 2025.

An expert suggested 10–20% unemployment in ~15 years (as quoted in the interview).

Over two million researchers (as stated by Hassabis).

Chapters

More from Matt

1:13
youtube.com

This Robot Is Unlike Any Other On The Market

2d ago
20:40
youtube.com

Google Just Caught Up (Gemini-4 Argon)

3d ago
1:01
youtube.com

How To Stop Being An "AI Meat Proxy"

3d ago
35:28
youtube.com

AI News: Haiku-5.5, Visual ChatGPT, Grok's New Trick, and more!

1w ago
0:58
youtube.com

AI Design Hack Saves HOURS

1w ago
30:22
youtube.com

AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know

2w ago
1:04
youtube.com

AI Made a Game I’d ACTUALLY Play

2w ago
1:12
youtube.com

Jev - The New AI model that has people talking

2w ago
34:38
youtube.com

AI News: Opus 5.5, GPT-6 Sol, Jev, Muse and More!

2w ago
1:29
youtube.com

5 ChatGPT Prompts To Try Right Now!

3w ago
0:51
youtube.com

I hate it when they do this!

3w ago
See all content