it was an audio recording, transcribed with speech to text models. there's definitely some errors and words lost. I also tried to emphasize this

Thanks for the clarification. Bit ironic given the talk’s subject. It is quite a bit of effort, but there’s something to say for going through and manually writing up the transcript like a journalist. Sometimes you can’t beat human effort ;)

What about a middle ground? Speech-to-text AI with manual corrections?

That’s a great approach! That’s what I meant to convey if I had been a bit more articulate. I assume journalists do exactly that. Takes away some laborious work while retaining accuracy.

anything you'd want fixed immediately? happy to do so – or even take this down if you wish. it's your talk.

nice catch! the original transcript kept saying dogs instead of docs. thats the only thing i fixed (until your r's find now) after laughing at it for a while

haha, love how we have the two pieces of the puzzle. we should merge!

No – for more context you can check out Karpathy's original essay from 2017: https://karpathy.medium.com/software-2-0-a64152b37c35

For context - I was in the audience when Karpathy gave this amazing talk on software 3.0. YC has said the official video will take a few weeks to release, by which Karpathy himself said the talk will be deprecated.

https://x.com/karpathy/status/1935077692258558443

To complete the loop, we need an AI avatar of Karpathy doing text-to-voice from the transcript. Who says AI can't boost productivity!

I listened to it with an old fashion CMU speech synth.

Do the talk's predictions about the future of the industry project beyond a few weeks? If so, I'd expect the salient points of the talk to remain valid. Hmm...

i synced the slides with the talk transcript here : https://latent.space/s3

so you took my transcript and put it behind a newsletter sub? haha. just share them!

not quite, i compiled the slides within a few hours of the talk yesterday well before your transcript was available. the slides are my main output/contribution. a full slides+transcript is too long for substack. i've linked your transcript prominently for people to find, and used it to fix slide ordering because twitter people took terrible notes for the purpose of exact talk reconstruction.

i exepct YC to prioritize publishing this talk so propbably the half life of any of this work is measured in days anyway.

100% of our podcast is published for free, but we still have ~1000 people who choose to support our work with a subscription (it does help pay for editors, equipment, and travel). I always feel bad that we dont have much content for them so i figured i'd put just the slide compilation up for subscribers. i'm trying to find nice ways to ramp up value for our subs over time, mostly by showing "work in progress" things like this that i had to do anyway to summarize/internalize the talk properly - which again is what we published entirely free/no subscription required

gotcha. def a good idea! i'm only a bit wary of making all of this a bit too put together, as to create an official source before there is one.

that being said, HN is a negative place, and not what I was trying to go for. thank you for your work with the slides!

Let's all make HN a less negative place.

(As a step towards making it a non-negative place.)

Looks like you are putting a derivative behind a paywall though, no? I think quid pro quo let pudiklubi publish your work too? Some kind of open license?

btw i think your transcript is missing most of the Perplexity slide discussion, right after the Cursor example

What a poor judgement he must have if his outlook becomes irrelevant in a few weeks' time.

Edit: the emoji at the end of the original sentence has not been quoted. How a smile makes the difference. Original tweet: https://x.com/karpathy/status/1935077692258558443

It was in jest, more a take of how quickly things move in AI

the way I read it it's more about how fast examples and references become irrelevant. fundamentals of the speech not.

We better stop talking about the future then.

Every time you talk about the future, it gets altered.

Hard determinists would disagree

Even if universe is not deterministic, what he said still does not make sense because him talking about future is part of destiny.

They basically have to.

unexpected nice touch that i can immediately check out the price as well. thanks for sharing!

cool idea, i've looked for tools like this.

i'd love to import my own notes or something like that, to also capture the thoughts i had while reading

thats a cool idea - probably it would need to be a mobile app, wouldn't it?

This is wild. I've been creating my own dataset of trending articles and ironically this is how I came across your post. I'm doing a similar project for my uni thesis.

I set out with similar hypotheses and goals like you (on a slightly different scale though, haha) but I've been completely stuck on the interactive map part. Definitely getting a lot of pointers from how you handled this!

Maybe one key difference in approach is that I've put more emphasis on trying to extract key topics as keywords.

For ex:

article (title): "Useful Uses of cat"

keywords: ['Software design', 'Contraction', 'Code changes', 'Modularity', 'Ease of extension']

My hypothesis is this will be a faster search solution than using the embeddings, but potentially not as accurate. Not that far yet to really prove this though.

Would love to hear what you think! Any other cool ideas on what could be done with the keywords? I explain my process a bit more here if interested: https://hackernews-demo.streamlit.app/#data-aggregation-meth...