DeepMind’s AI Just Did the Impossible
DeepMind’s AI Just Did the Impossible
Two Minute Papers 65,848 views 4 hours ago
1.84M subscribers
65,848 views • Nov 12, 2026
❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.com/papers 📝 DeepMind’s paper “AlphaProof Code” is available here:
Video Details
Date Nov 12, 2026
Views 65,848
Likes 1,923
DeepMind’s AI Just Did the Impossible
65,848 views
Nov 12, 2026
124 Comments

Pinned by @TwoMinutePapers
@rinrevo
Great production value on the animations, GG
52

@killersquirel11
4:17 Probably the most infamous limitation of Signal is that it does nothing to protect you if you accidentally invite a journalist into your group chat. Who would ever do such a thing, though?
29

@JohnDoe-sy6tt
What A Time To Be Alive!
33

❤ Liked by @TwoMinutePapers

@faizr93
Also, the animations have really stepped up, love it.
41

❤ Liked by @TwoMinutePapers

@verybigmac
I kind of miss the videos about physics papers (though I still love every upload from Two Minute Papers).
20

@lizardonastick
3:35 They (once again) confirmed that AI, as currently built, is still untrustworthy. Having untrustworthy AI agents act as referees for other untrustworthy AI agents introduces a slight problem…
5

@closingtheloop2593
Great to see the Lean language being put to practical use.
12
@ryanbushby4325
Formal verification has entered the chat
6
@DisturbedNeo
This is actually a huge breakthrough. As powerful AI agents become more accessible, using such agents to breach cybersecurity has become a real concern. A technology that can prove code is genuinely secure will be invaluable in the coming years. Great work by DeepMind, as always.
@malcolm_in_the_middle
“5:25 Minute” Papers.
9
@GrahamLaight
I’m a bit confused: maybe the math behind an encryption algorithm is sound, but when you write it into computer code, it becomes vulnerable: (1) numbers get truncated, leading to unintended patterns; (2) you end up…
