I have tried my best to get into podcasts, but am turned off by every single one being multiple hours long. Even though Lex is an outlier, the average show still has 1-2 hour episodes, sometimes releasing multiple times a week. I have tried my best to find quality ones with a 20-30 minute limit, but they don't seem to exist. Why is that the case? Is there really that much information to share? Or are they just not bothered to sit and edit their work? Do people really have the time to listen to all of them?
If I want an efficient method of receiving information, I read. If I want to listen to two people have an interesting conversation, or have information conveyed in a way unique to audio (eg. Hardcore History), I listen to a podcast. Two different mediums with two different appeals, imo.
There's also the passivity of it. I can't read while I drive, but I can listen, and length doesn't really matter in these cases because it's a way for me to pass time as much as anything else.
There are lots of podcasts that are 20-45 mins. I don't know of anyone that sits in a dark room and just listens to a podcast. Try going for a walk or on a commute or on a plane.
The long ones are the best to me because I enjoy interesting conversations rather than edited quick cut tiktok/twitter hot take style stuff.
If you have recommendations I'd love to know of them, but my search has come up empty.
There's a world of difference between a 10 second TikTok clip and a 25 minute episode of some show. The latter used to be enough time to get an entire day's worth of news not too long ago. So I don't think that the medium fundamentally needs to be movie length to be effective.
I suspect it's simply the case that shorter, more information-dense content takes a longer time to produce. It's a lot easier to have a 3 hour conversation with someone and dump it online. And I understand that there's an audience for just that. I do also think that there's a huge gap for short form podcast content waiting to be filled.
Pattern Recognition is the most similar to a "Lex style interview show" but there are many episodes in the 20-40 min range. Mostly with founders of companies. Really really good interviews. https://open.spotify.com/show/3Q267Ar7l43e1NeQIvUanv?si=2b7e...
Indicator by Planet Money is 15 mins with news about the economy in a fun way
Right; but there’s also a big difference between scrolling 25 second TikTok videos and settling in to listen to a 2 hour podcast with interesting people. I think the latter is better for our brains somehow. I don’t think the highly produced nature of TikTok makes it “better”.
As for 25 minute, produced podcasts? I really enjoy The Economist‘s Intelligence podcast for that. It’s definitely in the middle ground. Each episode is about 25 minutes. They’re tightly produced. They come out every week day. And each episode talks about 3 things going on in the world. If that’s what you’re looking for, give it a try!
If you spend time listening to podcasts you’ll learn to get the best 15-60 minutes out of a much longer episode. But ultimately it’s something you do while weeding, folding laundry or going on a run.
There are podcasts that try to edit down long conversations but they tend to cut the tangential and inadvertently revealing good stuff to deliver main points. If you just want main points you should just read an article.
I'd love to read an article. OpenAI's Whisper seems to solve this problem, and I'm amazed there isn't a youtube-to-text yet.
"Just listen to it in the background" isn't really workable for everybody. I barely listened to my own podcast episode. Text is just better for some people, which happen to be a majority on HN.
It's better for HN too. Note that we can't get any interesting comments until submission plus two hours, since it's two hours long.
I’m concerned you might thought I said everyone should be interested in podcasts, or able to listen to them. (I’m also concerned I misunderstood you!)
Besides that, yes, I love to read articles, too. I do a lot of transcription skimming of episodes where the voices and cadence don’t add much but the information is solid.
YouTube auto-transcripts are usually fairly readable or at least skimmable for the interesting time stamps. There are sites and services (like Readwise) that display them side by side with the video.
I don’t think HN benefits from people quickly skimming article text to add comments, unless they are primary sources themselves and can quickly address a misunderstanding, but that’s another discussion…
Oddly enough, Lex used to have full transcripts of his videos, through his website, but they seem to not be available anymore. I too prefer to read than passively listening. As of now, I just download the episode and listen at small intervals. YT-DLP can download a video/audio and separate it in chapters, that might work for some.
Sorry for misreading you; I appreciate the clarification. You're right of course. Thanks for pointing out the tools like Readwise – I wasn't aware of that, and it's super helpful.
In the spirit of the podcast, I asked GPT-4 for help in writing a reply, and it did a pretty okay job. https://imgur.com/a/MBQoq4A
> Do people really have the time to listen to all of them?
Yes! People really do have time to do all sorts of long attention span things. Podcasts for me are a mostly-downtime activity. They fall into the same category of activity as chilling on the couch with a good book. But the advantage is I can be doing something else while I’m listening - like driving, cooking, drawing or going for a walk.
You have just as many hours in your week as I do. We all need some downtime from time to time. I think podcasts are a delightful option. (If you find the right ones!)
I don’t listen to this particular podcast, but I do follow a few.
You don’t just… sit there and listen. (At least I don’t). Slap on some headphones and go for a walk, do household errands, etc. Put it on the background during your commute. Some folks have it on the background while they work, like people have done with radio shows for decades.
When I think of these long podcasts as movies, 2 to 3 hours sounds doable. However you will spend time on a movie or a podcast only if you think it will be worth your time. It turns out, some of them are indeed worth it. You don't have to watch them all.
I think it's a form of parasocial relationship. It's like you are "conversing" with someone who has the exact same tastes as you, which is kind of hard to come by in real life.
If you like this kind of thing, I did one a week ago. https://youtu.be/vYhtYjXNBCU It’ll be interesting to diff the two and see where we diverge.
Any progress on automatic transcriptions? Seems like Whisper should be able to give a readable version of both. This one’s a bit hard to listen to since it’s not a YouTube link, which means you can’t use your phone for other purposes for two hours.
It's a normal podcast (RSS feed and everything, not exclusive to any big platform to my knowledge), so you can listen to it on your favorite podcast app!
I avoid “celebrity” podcasts and rather listen to researchers with real experience in the trenches. For LLMs I’ve recently liked listening to Ilya Sutskever on Clearer Thinking and Eye on AI.
Anyone recommend any others in this vein, on this topic?
But that's terrible, because it seems like a good way to learn about OpenAI. It presents as if you can listen to this to gain insight into how OpenAI operates.
But you can't. What you get is a hour long PR fluff piece about a man related to OpenAI. But you listen anyway, and because it's appearing as this useful thing, your normal skepticism guards are down. So you swallow what could be abject nonsense.
More meta but thats all conversations. You never know how "vulnerable" someone is being no matter the medium, so again, it's up to you to understand that and gather the information that is there.
It absolutely has something to do with it. One of the symptoms of autism is a weak theory of mind, i.e. the inability to see the world through the perspective of another person, i.e. to understand their motivations.
Disagree. In podcasts I love interviews and conversations rather than monologues. I think the pauses and speakers taking turns, make listening less monotonous and also cognitively probably helps absorb info better
Monologues aren’t a realistic alternative, no podcast I know of is just one person speaking constantly. That sounds more like political radio shows, and those are also terrible.
Podcasts may not be a particularly efficient way to convey information, but they make up for it in convenience. You can listen while doing other things like walking the dog and make use of otherwise wasted time.
There were quite a few unintentionally hilarious moments. Some that stick out was Sam not having seen Ex Machina and then also Sam using the Elon Tears interview against Elon in a kind of Uno reversal move. Seems to be a weird rivalry brewing.
That rivalry has been brewing for many years now, and only now coming out into the public because of OpenAI's massive success (and of course Elon's Tweets). It's also kinda one sided. Elon fancies himself as the father of AI, but has failed at it multiple times, first by trying to take over OpenAI, then with all of Tesla's AI stuff. He is not a player in the industry, and cannot stand it.
He seems to think the world can absorb higher programmer productivity without destroying jobs. It's still one person's opinion but it was nice to hear.
That seems pretty clear to me? The web made programmers drastically more productive, and it coincided with a huge increase in both total employment and salaries, not a decrease.
For example, I first used Google in summer 2000 at a college internship, one where I flailed and didn't really know how to program. In high school and even college I could write code, but not anything real. (The CS program I went to was good, but focused on CS).
I think my programming skill and productivity took off when I had access to the web, and more people posted content to the web.
It really is like an external brain. Tim Berners-Lee isn't wrong; we just take it for granted now (and large parts of it have been ruined by advertising.)
And this is from a person who read the MS-DOS manual from front to back when I was a teenager. I also installed a Borland C compiler from a bunch of floppy disks, yet never really managed to write much of a C program.
So Google (in addition to education) made me much more productive, and StackOverflow attests that it has made millions of programmers similarly more productive. But yet I lived through a huge programmer tech boom in San Francisco for over a decade -- increasing staff, increasing salaries, etc.
What would be different about 2023?
It seems clear that the demand to build things will just get larger, just as it got larger from 2000 to now. There seems to be a "fixed pie" fallacy in the opposite view.
Why would it be the same? Think about it from first principles instead of assuming that what happened in the past will keep happening in the future. If there's a ceiling to the demand for programming work and we're near that ceiling, and you then increase the productivity of all programmers, the number of jobs will have to go down.
Is there a ceiling for demand and are we near it? I don't know. But the argument that stackoverflow and search made things better and so it goes for GPT-4 isn't thought out very well.
That's not first principles, that's a bunch of assumptions:
> If there's a ceiling to the demand for programming work and we're near that ceiling, and you then increase the productivity of all programmers,
Look at the history of civilization. Throughout millenia, we've advanced tremendously but still the vast majority of people across all (current) industries have jobs.
Just to put something out there, I think LLMs will be something like Google.
I remember when Google started improving a lot in 2003-2004, a coworker wrote a blog post "Is Google God?" People thought it was conscious.
It was absolutely revolutionary. It created huge amounts of wealth for itself and others
But clearly it hit a limit. It wouldn't surprised me if LLMs were huge boon for 5 or 10 years, and they hit a fundamental wall, stagnate, and then we repeat the cycle.
I agree with this, and LLMs seem pretty clearly the next step after search engines, in terms of intelligence augmentation. The parallels are too obvious to ignore. Search engines didn't replace humans, but they made us better at what we do, and let us do it with less effort. Also, librarians still have jobs.
I haven't come across any experienced software engineer who actually sees AI as a threat to their job. Maybe there will be a point when ChatGPT (or whoever else) can actually write better code than me, but I'd actually welcome that. Writing code is the most insignificant part of my job, and automating it away would leave more time for everything else, the stuff I actually get paid for. And in the worse case everyone will just switch to coding AI bots (because someone still needs to do that).
I agree with him, but I do worry that cost-effective software development may become drastically more capital-intensive, which will reduce the bargaining power of software developers in the job market. Currently, we largely own the means of production, because the valuable work comes almost entirely from our heads, but that may not be the case for much longer.
I watched it. It was overall a good interview, and I think Altman comes off as much less of a woke nazi than some have assumed he is. I think the political bias issue is one that will become less important over time, and Altman also seems to understand why it pisses people off. I liked the quote about Steve Jobs adding a handle to the computer because you shouldn't trust anything you can't throw out a window.
I was also pleasantly surprised at how open he sounds to things like open sourcing the model weights.
Overall, he seems fairly grounded in reality and I came away from the interview feeling less judgmental of the guy than I did before it.
https://news.ycombinator.com/item?id=35305561