I don't know if it is different in reality, but what *feels* different about the LLM push even compared to the personal computing push, is the idea that discovering use cases is your job, as an individual scrambling for survival and not the people who want to engage in commerce with you.
Hell, you pay a premium for doing something worthwhile. People know you’ll work in education, healthcare, elder care, child care, research or NGOs out of the goodness of your heart.
You gotta spend 50 hrs a week finding new ways to exploit workers or fool consumers to get paid a market rate
great channel generally
www.youtube.com/watch?v=pjrr...
well done, I tip my hat to you
Lmao I'm actually kind of impressed with how badly I did. A perfect no-hitter #subwaydle
Subwaydle 1535 X/6
⚪⚪⚪
⚪⚪⚪
⚪⚪⚪
⚪⚪⚪
⚪⚪⚪
⚪⚪⚪
Not trying to subtweet anyone in particular here, but, I feel like right now everybody with power in the world is worried about their velocity, that they're not going fast enough, and is just assuming they're headed in the right direction
It tickles me that it's still trivial to get Google Translate to produce this type of error (wouw means kite as in the bird, cometa means kite as in the toy; kite the bird and kite the toy are not homographs in either Dutch or Spanish, so the only explanation is using English as an intermediary)
you’re a new yorker if you live in new york city and want to call yourself one, there are no other requirements, doesn’t matter if you’ve lived here for one day or thirty years, this is the last and only meaningful thing about this stupid place (+maybe the big outdoor chairs at lincoln center)
The Venmo android app is something else
my bracket and the last perfect NCAA mens' basketball bracket both predict duke winning
The last perfect bracket agrees with me about the big winner which means we'll definitely win. As I always say go duck
Luckily I basically just pick low seeds because I know nothing about this sport, so it's statistically impossible for this to happen to me. Lucky because I don't think I would have the existential confidence to handle it
There are just two perfect brackets left! If I was the owner of one of those brackets I would think I was being Truman Show'd lol
😢
to add to my other replies: I would never, ever use LLM output as a primary source of knowledge. I use it as a search tool when I am sure I can verify the search results.
Fair enough. My personal definition of careful is a) I understand every line written by the LLM and will retain some mental model of how the logic works, and I suppose b) I am not writing unethical code with the LLM, which isn't really an LLM-specific check but is still obviously important.
bsky.app/profile/cstu...
See this thread for an example: bsky.app/profile/cstu...
I attempted to do a lot of this type of search pre-LLM and it was a lot harder. Realistically, I don't think I would have found the article without an LLM.
whoops, just noticed this thread got split. for clarity: bsky.app/profile/cstu...
This is true. I often find factual errors in information produced by LLMs in domains I happen to know a lot about. But note that I'm describing a *search* use case rather than using the LLM as a primary source of information; I would only do this when confident that I can verify search results.
Nonetheless, writing code is a non-trivial capability that can be useful when deployed carefully.
I largely agree and don't use LLMs for code generation very much myself. My biggest concern is cranking out code and features without concern for the cognitive load of understanding and maintaining them, which is what I think leads to the loss of productivity.
I can then use google translate to translate the article to English and verify that it is relevant. I use a healthy dose of skepticism and care at every step, but in this case I am confident in the result.
I asked Gemini the other day "Can you find any photos, videos, or articles to corroborate the claim that Vladislav Zotin's inauguration ceremony as the President of Mari El included both an orthodox and Mari animist priest?" It found an obscure article in Russian I probably couldn't find otherwise.
A completely different example: LLMs are a good alternative for google search when I'm looking for sources in a language I don't know. I can try to google translate individual phrases and enter them in google, but this is a pretty painful process that rarely works well for me.
I know smart, politically progressive software engineers skeptical of the societal impacts of AI who nonetheless tell me they haven't directly written a line of code since November because Claude has gotten that good. I don't use LLMs to that extent for coding, but they absolutely have utility.
As a fan of your youtube content, I think this absolutist viewpoint of yours is something you've kind of lost me on, not because "AI" as a phenomenon doesn't deserve criticism, but because the way you talk about it comes across as lacking nuance.
I do think that an absolutist viewpoint that LLM can't do anything useful, a view which I used to be tempted to hold, is at this point definitely factually incorrect.
I am a (former?) software engineer currently taking a break from my field in large part because the AI hype wave has been insufferable in terms of how management behaves. I consider myself a soft AI skeptic in that I think its benefits are often overstated and ethical risks understated. However...
