Boson-Video · target design for the read view

[1hr Talk] Intro to Large Language Models

Andrej Karpathy · 59:48 59 slides 21 chapters map ready in 1.5 s

The whole talk on one strip

Real data · hover, drag, or use the arrow keys
Most replayed Slide changes Chapters
Summary so far18:00
18:00 Summary so far
Slide 17 of 59
In the real view this box shows the frame and the words spoken at this second.

Read view

Illustrative text, written from the talk's slides

LLM Inference

0:20 – 4:17
Key slide
0:30
  • 0:30Llama 2 70B is just two files: about 140 GB of parameters and roughly 500 lines of C code that run them.✓ checked
  • 2:50It runs offline on a laptop. Getting the parameters is the expensive part.✓ checked

LLM Training

4:17 – 8:58
Key slide
4:20
  • 4:20Training compresses about 10 TB of internet text into those 140 GB of parameters.✓ checked
  • 4:20It takes about 6,000 GPUs for 12 days, roughly $2M.✓ checked

Ask the video

This mockup matches chapter titles. The real one reads the transcript.

The page takes the shape of the video

Talking head
Reads like an article
美股频道 video: one shot fills 97%
Slide talk
Slide deck with speaker notes
Karpathy: 59 slides in 59:48
Tutorial
Numbered steps and code
Not tested yet
Fast cuts
Contact sheet of shots
Rick Astley: 87 shots in 3:33