“We May Release Models We Cannot Control”: Amodei, Altman… and Gemini Carbon
836 views • Nov 2, 2026
To be honest, by AI standards, this might not have been the craziest week, but that doesn’t mean major headlines weren’t made—from Altman and Anthropic hinting that they will release models they cannot control and whose behavior cannot be verified, to the powerful arrival of Gemini 4 Carbon, Amodei’s rationalizations, de-tokenization, and more…
Chapters
Intro
0:00
Anthropic Employee Survey
1:26
Amodei’s Five Rationalizations
4:32
Altman: Only Alignment Works
11:43
Gemini 4 Carbon & Why Models Keep Surprising Us
18:41
15 Comments

@jakubbartodziej3367
Phew, after seeing Argon’s scores on Integrity Bench, I was genuinely worried Google had kidnapped you.

@DangerousGas
My prediction is that the interpretability problem will be solved much faster than we expect. We’ve already made incredible progress.

@mehdihassan8316
I worry Fable 5.5 will be used to defend against cyberattacks from Iran.

@Psychopathicviewer
Caught this within 47 seconds of refreshing, let’s go!

@SaInTDomagos
Hacking is getting increasingly out of hand. The next few months and years are going to be wild. Digital security is top priority right now.

@hendricksausges3006
The only constant is change.

@mahamadkader488
Finally, an update! Let’s see what’s changed in the AI world since the last video.

@om98357
When on earth are we actually going to start arresting the criminals causing harm?

@thecosmicskye
Haha, ChatGPT actually alerted me that this video was uploaded.
