Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Things created with GPT-4

twitter.com

Post reply on HN
Print view View original

11–20 of 69 posts

Re: Things created with GPT-4

#11

Post by SketchySeaBeast »

turnsout wrote: post #10

This current moment is super impressive, but the output of the models is still mediocre. I worry that over the next 12-24 months, we'll be overwhelmed with autogenerated content (docs, apps, games, news) that gets a B- or C. Once the novelty wears off, no one wants to play your [terrible] autogenerated asteroid clone. There are a lot of startups that are two LLMs in a trench coat, and they will not be able to improve…

Your BAD autogenerated asteroid clone. "Give it time" people will say, but how does one prompt ChatGPT to make a game fun, or a story actually interesting?

Re: Things created with GPT-4

#12

Post by lm28469 »

How do they make sure these "financial advisor" or "research abstract summarisers" are hallucination free ? So far it still is deeply in the "cool gadget" category for me, 6 months after the chat GPT revolution and not much to see

Re: Things created with GPT-4

#13

Post by spoiler »

Don't get me wrong, it's really cool technology, but calling the first example a "game" is... Grossly overstating what it is. However, given it was made by someone who's not a [game] developer, it's pretty cool nonetheless.

Re: Things created with GPT-4

#15

Post by fiso64 »

lm28469 wrote: post #8
cs702 wrote: post #2

The imagined computers that looked so unrealistic in so many past science-fiction movies and TV shows no longer look... unrealistic. Those imagined computers now look more like upgrades to current technology. Imagine an AI model 10x, 100x, 1000x larger than GPT4. That feels... surreal to me.

> Those imagined computers now look more like upgrades to current technology. I still don't get that argument. You can train a parrot for a week and he'll talk a few words, training it for a year won't make him become Shakespeare, LLMs will never ever automagically evolve into these AGI tier computers, not even their creators argue that, no one besides techbros argue that in fact.

>not even their creators argue that

Yes, their creators take the far more defensible and scientific position of saying "we don't know".

Re: Things created with GPT-4

#16

Post by ausbah »

seems like GPT systems are the most useful when you can quickly verify the correctness of the output (ex code generation) but else quickly becomes dubious (ex how do you know a summary is correct or useful?)

Re: Things created with GPT-4

#17

Post by WalterSear »

lm28469 wrote: post #8
cs702 wrote: post #2

The imagined computers that looked so unrealistic in so many past science-fiction movies and TV shows no longer look... unrealistic. Those imagined computers now look more like upgrades to current technology. Imagine an AI model 10x, 100x, 1000x larger than GPT4. That feels... surreal to me.

> Those imagined computers now look more like upgrades to current technology. I still don't get that argument. You can train a parrot for a week and he'll talk a few words, training it for a year won't make him become Shakespeare, LLMs will never ever automagically evolve into these AGI tier computers, not even their creators argue that, no one besides techbros argue that in fact.

We are hearing about new, improved parrots on an almost daily basis.

Re: Things created with GPT-4

#18

Post by turnsout »

SketchySeaBeast wrote: post #11
turnsout wrote: post #10

This current moment is super impressive, but the output of the models is still mediocre. I worry that over the next 12-24 months, we'll be overwhelmed with autogenerated content (docs, apps, games, news) that gets a B- or C. Once the novelty wears off, no one wants to play your [terrible] autogenerated asteroid clone. There are a lot of startups that are two LLMs in a trench coat, and they will not be able to improve…

Your BAD autogenerated asteroid clone. "Give it time" people will say, but how does one prompt ChatGPT to make a game fun, or a story actually interesting?

^ Yes, edited my post! I've spent HOURS tweaking prompts and going back and forth with GPT3–4 to get more creative and interesting output, but it's not there yet. This current generation will just never give you legitimately good song lyrics, let alone a game with an interesting and compelling mechanic.

Re: Things created with GPT-4

#19

Post by lm28469 »

fiso64 wrote: post #15
lm28469 wrote: post #8

Earlier quoted context omitted.

> Those imagined computers now look more like upgrades to current technology. I still don't get that argument. You can train a parrot for a week and he'll talk a few words, training it for a year won't make him become Shakespeare, LLMs will never ever automagically evolve into these AGI tier computers, not even their creators argue that, no one besides techbros argue that in fact.

>not even their creators argue that Yes, their creators take the far more defensible and scientific position of saying "we don't know".

Not at all

Same Altman (openAI ceo) :

> The GPT-4 rumor mill is a ridiculous thing. I don’t know where it all comes from, people are begging to be disappointed and they will be. The hype is just like... We don’t have an actual AGI

Re: Things created with GPT-4

#20

Post by lm28469 »

WalterSear wrote: post #17
lm28469 wrote: post #8

Earlier quoted context omitted.

> Those imagined computers now look more like upgrades to current technology. I still don't get that argument. You can train a parrot for a week and he'll talk a few words, training it for a year won't make him become Shakespeare, LLMs will never ever automagically evolve into these AGI tier computers, not even their creators argue that, no one besides techbros argue that in fact.

We are hearing about new, improved parrots on an almost daily basis.

And no matter how many W16 you put in a bugatti veyron it will never be a space shuttle

I don't get how people, especially here of all places, fall for such things

Post reply on HN