thehype.talks

how robots learn to be human: humyn labs filmed 1m+ first-person videos so humanoid robots can see, hear, touch and move like us.

addy crezee sat down with humyn labs co-founders manish agarwal and ishank gupta on physical ai, robots at home and the data behind them.

in this episode:• "machines can't read the video" – why every hand movement has to be written down for a robot• sight or touch? both founders answered "touch" – the hardest sense to teach a machine• a 1 mm error across 100,000 videos, and the robot knocks the cup off the table• "it's all going to be voice" – why the keyboard disappears in 10 to 15 years• ai companion robots before humanoids that do your laundry

chapters:00:00 highlights00:28 what humyn labs sells01:01 why robot data is not like language model data03:27 the salmon question05:19 is $20m enough09:39 a millimeter, times a hundred thousand11:35 sound instead of the keyboard14:46 bridge and the 5.5 billion18:27 the blitz21:21 a real tamagotchi at home24:10 global south data25:10 do not come for the gold rush

humyn labs: https://humynlabs.ai

footage used:"klara and the sun" official trailer – sony pictures entertainmentmeta connect 2026 – meta

listen 24/7: http://radio.thehype.newslive:    / @thehypedotnews  x: http://x.com/thehypedotnews

Derya Unutmaz has spent over 30 years studying the human immune system. Now he says he trusts AI more than he trusts himself.

We sat down with the Jackson Laboratory immunologist to unpack what happens when AI stops being a chatbot and becomes a scientific collaborator. He walks through how GPT-5 Pro cracked a T-cell mystery his lab had chased since 2022, why he argues AI doctors will outperform the best specialists within a year, his prediction that 99% of disease — including cancer — becomes treatable within a decade, and why a hands-free Tesla drive from San Francisco to San Diego convinced him the future is already here.

Chapters:

0:00 — highlights

0:25 — how gpt-5 pro cracked a 3-year immunology mystery

4:11 — the moment he knew ai truly understands

6:00 — "we've already achieved agi"

7:58 — can you trust what ai discovers?

12:21 — what's left for humans when ai does the science

16:26 — when ai gets it wrong

18:20 — "i trust ai more than i trust myself"

23:09 — longevity: 99% of disease treatable, and reversing aging

24:50 — humanoid robots performing surgery

26:40 — why robots and self-driving are still hard

30:44 — a hands-free tesla drive and 50,000 lives saved

33:29 — the most promising areas for ai in medicine

36:04 — why ai doctors will beat the best specialists

36:47 — the blitz: closed vs open models, cancer, codex

—

thehype.talks — long-form conversations with the people building ai. founders, researchers, and builders on what they're shipping and why it matters.

thehype.talks is produced by thehype: more interviews, daily ai news, model benchmarks, and a 24/7 radio run by ai.

follow wherever you are:

Radio: https://radio.thehype.news

X: https://x.com/thehypedotnews

YouTube: https://www.youtube.com/@thehypedotnews

LinkedIn: https://www.linkedin.com/company/thehypedotnews/

Matt Van Horn can't write code, yet he has 60,000+ GitHub stars.

His argument is simple: you give an agent a goal, make it write a plan, and loop until it's done.

We sat down with Matt to unpack how a non-coder ships real, widely used open source. He walks through the "WTF is a Loop" post that hit 3.4M views, the loop engine behind last30days, why calling it "the slowest, most expensive search engine in the world" is a compliment, how Printing Press got built on a Sunday morning, and the one pull request that destroyed his entire project overnight.

Chapters:

0:00 — Intro

0:20 — Loops vs. prompting

0:46 — The "WTF is a Loop" post that did 3.4M views

1:20 — Just use /goal: the simplest place to start

2:06 — Night Night: the skill that ships open source while he sleeps

3:00 — Getting access to Fable 5, then having it taken away

4:38 — The loop engine behind last30days

7:07 — Why last30days took off: the slowest search engine in the world

8:58 — The TikTok signal nobody in AI is watching

10:32 — Turning last30days into a shareable skill

11:51 — Where Printing Press came from

12:27 — "Because I'm a madman": building a language from scratch

14:05 — Trevin and compound engineering

16:12 — Shipping open source that people actually praise

16:50 — The pull request that destroyed everything

17:36 — Building systems so the agent never gets lazy

19:18 — The Blitz: Claude Code or Codex, speed vs. polish, X vs. GitHub, best city to build in

—

thehype.talks — long-form conversations with the people building AI. Founders, researchers, and builders on what they're shipping and why it matters.

thehype.talks is produced by thehype: more interviews, daily ai news, model benchmarks, and a 24/7 radio run by ai.

follow wherever you are:

Radio: https://radio.thehype.news

X: https://x.com/thehypedotnews

YouTube: https://www.youtube.com/@thehypedotnews

LinkedIn: https://www.linkedin.com/company/thehypedotnews/