I couldn't agree more. Neither LLMs or search engines yield reliable answers due to the ML model and misaligned incentives respectively. Combining the two is a shortsighted solution that doesn't fix anything on a fundamental level. On top of that we have the feedback loop. LLMs are used to search the web and write the web, so they'll just end up reading their own (unreliable) output over time.
What we need is an easier way to verify sources and their trustworthiness. I don't want an answer according to SEO spam. I want to form my own opinion based on a range of trustworthy sources or opinions of people I trust.
If you are optimizing for finding geniuses like R, you may be right. Many probably fall through the cracks of the educational system. But I don't think this is what we are or should be optimizing for. The vast majority of people would end up unemployable if they weren't "forced" to study things they don't enjoy because some skills are just more employable than others. You're lucky if you enjoy engineering/science, but not so lucky if you only care about art literature.
4. Skimming HN is useful in text-format because you can go deeper on something that catches your eye by following the link. The podcast format removes this because you can't click on something you hear and dive deeper into it. Skimming without the ability to dive into something is not useful, see next point.
5. High-level summaries are not my use case for HN, or any social media. They don't provide real value, just the illusion of value. What I want to dive deep into stories I am curious about. That can be done with TTS, but I would need to curate my feed first, and then use the podcast format to dive deeper into my curated stories.
I almost feel like this product would be more useful if you remove the LLM aspect and instead let me paste a list of HN threads into it and just TTS all of them, including the full comments. Then I could listen to this long-form content while driving or doing something else.
I don't think the heuristics are that different. SEO-spam and BS content existed before, and both Google and YT were full of them, all made by human "content creators" who optimized for clicks and focused on gaming the YT recommendation system. AI content isn't that different. But unfortunately it's now 100x easier to generate such content, so we see a lot more of it. The problem is fundamentally a problem of incentives and the ad-based business model, not a problem of AI. AI has made the enshittification problem a lot more visible, but it existed before.
I don't know what the solution here is. My guess is that the "public internet" will become less and less relevant over time as it becomes overrun by low-quality content, and a lot of communication will move to smaller communities that rely heavily verifying human identity and credentials.
What people forget to mention is when they have read a book. I read the Selfish Gene when I was a teen and it left a lasting impression. It was published in 1989. How the Mind works was published in 2011.
Reading the Selfish Gene today as an adult, when you've probably read a dozen similar books already, is not going to have the same effect. It's going to be pretty boring. That's why asking for book recommendations is flawed to begin with.
Dawkins is not an original researcher. He is wonderful at synthesizing research and writing eloquent popular books.
The importance of The Selfish Gene is to write a correct summary of the neo-Darwinian project. In the future (perhaps even now), it will be more famous for the coining of meme (in an appendix IIRC).
However, The Extended Phenotype is a remarkably original exposition of important (then) contemporary frontiers of Darwinism. It is a powerful idea, powerfully described.
It is certainly his most important original contribution to the literature. And a beautiful book. True and convincing. It will change your view of the world. There is no higher praise.
"The God Virus: How Religion Infects Our Lives and Culture" by Darrel Ray follows in the footsteps of The Selfish Gene to certain logical conclusions about clusters of related memes/concepts (specifically: how religion(s) formed and evolved as clusters of overlapping concepts)
Interesting, I feel pretty much the opposite. To me these podcasts are the equivalent of the average LLM-generated text. Shallow and non-engaging, not unlike a lot of the "fake marketing speech" human-generated content you find in highly SEO-optimized pages or low-quality Youtube videos. It does indeed sound real, but not mind-blowing or trustworthy at all. If this was a legit podcast found in the store I would've turned it off after the first 30 seconds because it doesn't even come close to passing my BS filter, not because of the content but because of the BS style.
It's decent background noise about a topic of your choice, with transparently fake back-and-forth between two speakers with some meaningless banter. It's kind of impressive for what it is, and it can be useful to people, but it´s clearly still missing important elements that make actual podcasts great
It’s intentionally fine tuned to sound that way because Google doesn’t want to freak people out.
You can take the open source models and fine tune them to take on any persona you want. A lot like what the Flux community doing with the Boring Reality fine tune.
Exactly. And pay more attention to the delta/time and delta/delta/time.
We are all enjoying/noticing some repeatable wack behavior of LLMs, but we are seeing the dual wack of humans revealed too.
Massive gains in neural type models and abilities A, B, C, ..., I, J, K, in very little time.
Lots of humans: It's not impressive because can't L, M, yet.
They say people model change as linear, even when it is exponential. But I think a lot of people judge the latest thing as if it somehow became a constant. As if there hasn't been a succession of big leaps, and that they don't strongly imply that more leaps will follow quickly.
Also, when you know before listening that a new artifact was created by a machine, it is easy to identify faults and "conclude" the machine's output was clearly identifiable. But that's pre-informed hindsight. If anyone heard this podcast in the context of The Onion, it would sound perfectly human. Intentionally hilarious, corny, etc. But it wouldn't give itself away as generated.
I feel like people have been saying that since GPT-4 dropped (many papers up the line now) and while there have been all sorts of cool LLM applications and AI developments writ large, there hasn't really been anything to inspire a feeling that another step change is imminent. We got a big boost by training on all the data on the Internet. What happens next is unclear.
Except that none of the fundamental limitations have changed for many years now. That was a few thousand papers ago. I'm not saying that none of the LLM stuff it's useful, it is, and many useful applications are likely undiscovered. I am using it daily myself. But people expecting some kind of sudden leap in reasoning are going to be pretty disappointed.
We don't even need to look that far. During an extended interaction the new ChatGPT voice mode suddenly began speaking in my boyfriend's voice. Flawlessly. Tone, accent, pauses, speaking style, the stunted vowels from a childhood mouth injury. In that moment there were two of him in the room.
Yeah, but "reason" is not a well-defined term. It means different things to different people in different contexts. It's just marketing speech. You can easily argue that all ML models, even those from 50 years ago, can reason to some extent.
Eight Years of GraphQL, and I have yet to find a single use case for it in my projects. I've had to interact with external GraphQL APIs a few times and each time it has been a terribly painful experience. Funnily enough, for the few queries where GraphQL would've actually been useful to get some deeply nested data structure, it was usually impossible to use it because the "query is too big/deep" or it had cycles
Not so sure about that. Because when you hunt around for things you know the source. If I watch a lecture by a math professor I can make a guess about the trustworthiness of the information. With an LLM, I can't.
What we need is an easier way to verify sources and their trustworthiness. I don't want an answer according to SEO spam. I want to form my own opinion based on a range of trustworthy sources or opinions of people I trust.
reply