supercuts / didnt-hold-up
Didn’t hold up
The most confident predictions the public record went against, highest certainty first. · 14 clips, 3 minutes, hands free.
0:00 / 0:20exact quote · 20.2s
1/14Andreessen: AI will never transform existing US K-12 public classrooms
Marc Andreessen · 2026Didn’t hold upfull source →
1Andreessen: AI will never transform existing US K-12 public classroomsMarc Andreessen · 20sDidn’t hold up
2Cheah: Standard Transformers Will Never Scale to Ten Million TokensEugene Cheah · 12sDidn’t hold up
3Cheah: Cloud providers will slash model inference prices before raising themEugene Cheah · 14sDidn’t hold up
4Nair: LLM agents will hit $1T before robotics hits $10BAshvin Nair · 6sDidn’t hold up
5Kamradt Predicts ARC-AGI-2 Will Not Be Beaten For 12 MonthsGreg Kamradt · 3sDidn’t hold up
6Patel: AI inference will deploy more GPUs than training by 2024Dylan Patel · 9sDidn’t hold up
7Godement: Developers will rely on continuous, automated fine-tuning within yearsOlivier Godement · 19sDidn’t hold up
8Sohmers: NVIDIA Blackwell memory bandwidth efficiency will be lower than HopperThomas Sohmers · 13sDidn’t hold up
9Conrad: GPU market will likely return to a shortage by winterEvan Conrad · 8sDidn’t hold up
10Swix: Overcast will basically never have searchable transcriptsShawn Wang · 7sDidn’t hold up
11Patel: Google will avoid deploying local laptop models to retain controlDylan Patel · 9sDidn’t hold up
12Swyx: OpenAI will always release both general and Codex model variantsShawn Wang · 7sDidn’t hold up
13Mohan: Automated PR generation will require specialized models trained on diffsVarun Mohan · 21sDidn’t hold up
14Swix: OpenAI will issue a cryptocurrency token to fund computeShawn Wang · 17sDidn’t hold up