viraldi.
Video details
How it works
Back to discovery
6:04
Technology & Computing / VIDEO

INSANE: This 46 MB AI Model Runs 5x FASTER on My Laptop (Frozen Embeddings TESTED)

Prompt Engineer 48ยท 30.3K subscribers

15d ago ยท English audio

A new paper (arXiv 2610.22428) says the learned embedding table in small language models is overpriced: replace it with a frozen 16 MB table of 512-bit codes and the model matches or beats the learned one on three of four benchmarks, a ternary version fits in 46 MB, and new words can be added after training at a 3% cost instead of 43%. I read the paper, pulled the repo, built the authors' CPU kernel on my ownโ€ฆ

Watch on YouTube
3.0ร—Score at discovery
2.5KViews
817Median ยท 34 videos

Outlier score

This score was measured on Wed, 04 Nov 2026 17:30:39 GMT. Current view counts may differ.

2.5K views รท a 817 baseline = 3.0ร—. Compared with 34 earlier videos from the same channel, matching the format recorded at scoring time. Older-upload baseline: reference videos had at least as long to gain views. This comparison is not age-matched.

YouTube snapshot: Wed, 04 Nov 2026 20:08:07 GMT. Counters may have changed.

Read the methodology

Generative AI & Agents

Channel-balanced percentile: 30

Comparison details

192 videos from 122 other channels, with each channel weighted equally. Same format and language, with 10,000โ€“99,999 subscribers. Observed at 362 hours after publication, within ยฑ48 hours.

This describes the collected sample, not all of YouTube. Updated Wed, 04 Nov 2026 21:20:15 GMT.

Niches & tags

Transcripts and detailed analytics are available for recent outlier Shorts up to three minutes. Research tools

Breakouts from other channels