AI Scout, building text-2-video @Waymark, host of The Cognitive Revolution podcast
Detroit, MI
- Tweets 5,369
- Following 3,255
- Followers 17,710
- Likes 56,589
Pinned Tweet
Introducing "text-2-commercial" – the unique text-2-video experience we're building @Waymark
Watch our CEO @aperskystern make an original, creative, compelling marketing video for a small business in <1 minute.
I'll explain how it works in the thread
23
81
542
221,929
That is the show.
AI:AM is live weekday mornings, 9a PT. Back Monday.
nitter.net/i/broadcasts/1aJbdEVXZ…
386
Nathan's verdict closing AI:AM: 'a banger episode' — two deep-tech guests off the mainline paradigm of ever more RLVR on language models.
The upside: biomedicine, and all the real-world sensor data that largely gets logged and mostly ignored.
nitter.net/i/broadcasts/1aJbdEVXZ…
433
Georgescu frames drug discovery as RL: the experiment is the actor, 'make it healthy' the objective, the tissue's state the value function.
A planet of Vivodyne labs couldn't brute-force combination therapies — the model picks what to test next.
nitter.net/i/broadcasts/1aJbdEVXZ…
1
402
Georgescu's scaling-laws answer: today's virtual cell models saturate after a couple percent of their input data.
Cells on plastic just try to colonize the plastic — knock out a gene, mostly nothing changes. 'There's not even causality to extract.'
nitter.net/i/broadcasts/1aJbdEVXZ…
437
Nathan's frame, put to Gillian: what AI still lacks is intuition for domains humans have no native senses for. Does that end in a frontier-lab bidding war for Archetype?
Gillian's vision: a key node — Newton agents talking to digital agents.
nitter.net/i/broadcasts/1aJbdEVXZ…
365
Next on AI:AM, 10:15a PT: Andrei Georgescu, founder and CEO of @vivodyne.
12 robotic labs running 3.1M human tissue experiments a year. His argument: AI biology isn't compute-bound, it's bound by causal human data.
nitter.net/i/broadcasts/1aJbdEVXZ…
338
Gillian on AI:AM: a Japanese construction giant spent 5 years moving a river. Newton turned its cameras + sensors into a Gantt chart of crew activity.
The pattern nobody saw: productivity stays low days after a storm — debris lags the weather.
nitter.net/i/broadcasts/1aJbdEVXZ…
334
This morning on AI:AM, 9:30a PT: Nick Gillian, CTO and cofounder of @PhysicalAI (Archetype AI).
One foundation model, Newton, pointed at drilling rigs, an assembly line, a canal in Niigata and a Bellevue crosswalk. No robots. Just perception.
nitter.net/i/broadcasts/1aJbdEVXZ…
1
346
Up now on AI:AM: Andrei Georgescu, founder and CEO of @vivodyne.
12 robotic labs growing human tissue at a scale of 3.1M a year. His argument: AI biology isn't compute-bound, it's bound by causal human data.
nitter.net/i/broadcasts/1aJbdEVXZ…
332
Up now on AI:AM: Nick Gillian, CTO and cofounder of @PhysicalAI (Archetype AI).
Archetype's Newton is one foundation model pointed at drilling rigs, an assembly line, a canal in Niigata and a Bellevue crosswalk. No robots. Just perception.
nitter.net/i/broadcasts/1aJbdEVXZ…
383
Prakash's frame on AI:AM, via Noah Smith: do you want obedience or benevolence? You can't have both — obedience does the bad thing you ask; benevolence protects you even from yourself.
The labs, he says, are confused about which they're going for.
nitter.net/i/broadcasts/1aJbdEVXZ…
441
Jensen to Ezra Klein, per Prakash on AI:AM: if companies can't control their products even during the build phase, shut them down.
Nathan: hard to steelman 'It's just software' from a front-row seat — but sees a moral clarity in the stance.
nitter.net/i/broadcasts/1aJbdEVXZ…
1
2
468
Foundation Models for the Physical World + Making Biology Computable nitter.net/i/broadcasts/1aJbdEVXZ…
399
Nathan Labenz retweeted
From Animal Models to Robotic Human Tissues—and AI for the Physical World
Join us this morning, Friday, 25th September:
• 9:30 AM PT — Nick Gillian, CTO, Cofounder, Archetype AI
• 10:15 AM PT — Andrei Georgescu, CEO, Vivodyne (@vivodyne)
𝕏 Chat is Live on Screen, Come Join Us:
nitter.net/i/broadcasts/1aJbdEVXZ…
2
2
593
That is the show.
AI:AM is live weekday mornings, 9a PT. Back Friday.
nitter.net/i/broadcasts/1qxoNYLAN…
2
553
Nathan's old metric, revived on AI:AM: a year ago, some buildouts could do a GPT-4-scale training run in a couple of days. Probably two triplings later, a hyperscaler can potentially do a couple per day.
His bottom line: buckle up once again.
nitter.net/i/broadcasts/1qxoNYLAN…
1
548
On AI:AM: the A100 was supposed to be end-of-life by now, Nathan noted — last he heard, it still rents for more per hour than at launch.
Nelms's thesis: useful life beyond six years — price-elastic open-weight models route to older hardware.
nitter.net/i/broadcasts/1qxoNYLAN…
1
1
532
Nathan to Wayne Nelms: I don't think AI broadly is a bubble — but an issue in the financing chain wouldn't shock me. What's the canary in the coal mine?
Nelms: pricing matters, but the real one is utilization — GPUs available vs. actually rented.
nitter.net/i/broadcasts/1qxoNYLAN…
521
A moat take from Wayne Nelms on AI:AM: everyone cites NVIDIA's hardware and software edge — but the biggest moat, he says, is the financing landscape. When you're building a neocloud, NVIDIA GPUs are just materially easier to underwrite.
nitter.net/i/broadcasts/1qxoNYLAN…
1
472