Know What's in the Talk Before You Invest the Hour

A free NVIDIA model pulled the key points out of Jensen Huang's 92-minute keynote in 20 seconds and got 12 of 14 details right.

I have a ton of talks I'd love to listen to and I don't have the time. I'm guessing you're in the same spot. Most of the time I just want to know what the high points are before I invest an hour.

NVIDIA is our subject this week and the company has a free way to do this. Its open AI models are called Nemotron and you can try one on NVIDIA's site without an account. Give it the transcript of a long talk and you get what was announced almost right away, but it won't tell you which part people will remember.

I asked Claude, my daily AI assistant, to try it on the 92-minute keynote Jensen Huang gave at NVIDIA's GTC conference in Berlin on October 21. Claude tried it once yesterday and the notes came back in 20 seconds.

You can run the same steps yourself. You copy the transcript from the description under the YouTube video and paste it into the Playground tab for the Nemotron model. Then you ask for the five points most worth knowing, the time stamp where each one starts, and which ten minutes to watch if you only watch ten.

There's one caution before you paste. NVIDIA's page says it records what you send to improve its products and asks you not to upload confidential information or recordings of people's voices, so keep this to public talks and paste the transcript instead of the recording.

The five points were Jensen's case that AI is now a manufacturing industry, NVIDIA's investment in Europe, the new Vera Rubin computers, a data center that ships in containers, and security software for AI agents.

Claude checked 14 details in those notes against the transcript and 12 were right. One miss was what Jensen calls a five-layer cake, where the notes listed four layers and not the ones he names, and the other was money. Jensen says NVIDIA invested nearly $10 billion in Europe and the notes turned that into euros. The number was right and the currency was wrong, so search the transcript for a number before you pass it along.

The pick of ten minutes is where I'd trust it least. NVIDIA's model said to watch minutes 20 to 30, where Jensen holds up the new hardware, and it never mentioned the ending, where an AI agent teaches a small Hugging Face robot called Microduck to see and walk. That's the part Nous Research was posting on X, and afterward Jensen wrote, "I was the opening act for a family of ducks." I'd treat the ten-minute pick as a guess and skim the last ten minutes of any talk myself.

Try it on a talk you already know before you trust it on one you don't, but you do what works best for you. Reply and tell me which talk has been sitting on your list.