explainer · HACKER NEWS RECOMMENDATION
But how do AI images and videos actually work? | Guest video by Welch Labs
3Blue1Brown · 1 HN recommenders · 1 recommendation months
Recommended by 1 Hacker News commenters across 1 months (2025–2026).
SignalTube analysis
Among 14 other explainer videos tagged LLMs & generative AI, the mean is 3.1 distinct HN recommenders. This video has 1 distinct HN recommenders, below that mean. Recommendations span 1 distinct month; the peer average is 2.6 months. This is a data-derived comparison, not a human review or quality rating.
33% of its 3 labeled comments are marked as recommendations or strong recommendations. HN mention years are 2025–2026; those years describe discussion history, not upload date. A verified runtime is not present in the current metadata snapshot. A verified YouTube publication date is not present in the current metadata snapshot.
- Channel / speaker
- 3Blue1Brown
- Verified runtime
- Not verified in the current YouTube metadata
- YouTube upload date
- Not verified
- Hacker News mention years
- 2025–2026
SAME TOPIC AND FORMAT
Compare with related videos
The comparison cohort uses other explainer videos in LLMs & generative AI. “Shared recommenders” counts HN users who recommended both videos in the source graph.
| Video | HN recommenders | Recommendation months | Positive label share | Shared recommenders | Runtime | Upload date |
|---|---|---|---|---|---|---|
| But how do AI images and videos actually work? | Guest video by Welch LabsCurrent video | 1 | 1 | 33% | — | Not verified in the current YouTube metadata | Not verified |
| Transformers, the tech behind LLMs | Deep Learning Chapter 5explainer · 3Blue1Brown | 13 | 10 | 81% | No recorded overlap | Not verified | Not verified |
| Attention in transformers, step-by-step | Deep Learning Chapter 6explainer · 3Blue1Brown | 5 | 4 | 63% | No recorded overlap | Not verified | Not verified |
| Attention Is All You Needexplainer · Yannic Kilcher | 4 | 4 | 80% | No recorded overlap | Not verified | Not verified |
THREE SOURCE-BASED CONTEXT POINTS
Why Hacker News recommends it
Each note combines the dataset’s stance label and short rationale; comment bodies are not reproduced. Open the HN source for the original discussion.
The archive classifies this as a recommendation in “Playing with Vision Embeddings”. Its short rationale: Nice video on training text and image encoders together.
The archive classifies this as a contextual reference in “The maths you need to start understanding LLMs”. Its short rationale: Explains diffusion speedups from statistical mechanics-based differential equations.
The archive classifies this as a contextual reference in “Training language models to be warm and empathetic makes them less reliable”. Its short rationale: Shows subtracting the 'low quality video' prompt to improve generation.