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

New GPT-3 model: text-DaVinci-003

beta.openai.com

Post reply on HN
Print view View original

91–100 of 216 posts

Re: New GPT-3 model: text-DaVinci-003

#91

Post by version_five »

eru wrote: post #86
EZ-Cheeze wrote: post #69

Earlier quoted context omitted.

Write a short rhyming poem explaining Einstein's theory of general relativity in easy but accurate terms If you want to understand Einstein's thought It's not that hard if you give it a shot General Relativity is the name of the game Where space and time cannot remain the same Mass affects the curvature of space Which affects the flow of time's race An object's motion will be affected By the distortion that is detect…

Seems pretty good. Though rhyming 'thought' with 'shot' seems like a really weird dialect of English to me. Does it rhyme in American English? (I was recently bemused when I learned that Americans rhyme macabre with job.)

Can you provide a word that in your mind rhymes with each of thought and shot? Even in different accents I can picture, these always rhyme in my head.

Re: New GPT-3 model: text-DaVinci-003

#92

Post by lairv »

joe_the_user wrote: post #31
ctoth wrote: post #3

This one seems weirdly more verbose, in my brief testing. Definitely uses more tokens on average than 002 when answering. The writing style seems oddly-pretentious and yet somehow uninformed, perhaps there were too many copies of HN in the dataset? I kid, I kid.

After the failure of FB's Galactica model, I wonder whether LLM are starting to hit diminishing return. I've read of several ways diminishing returns could happen: * Not enough training data - you've used up the Internet (even a percentage of the Internet might be as much as is usable by clever brute force). * Not enough compute time to fully train (we're not close to that) * The model covers such a large area that t…

I am eagerly awaiting GPT4, it might be the case that language models will hit a plateau, just like video games graphics hit a plateau too, with each generation being less and less impressive compared to the previous, while still not being perfect

Re: New GPT-3 model: text-DaVinci-003

#93

Post by stavros »

chpatrick wrote: post #78
kitclen wrote: post #47

I'm not seeing DaVinci enabling any scalable business models with its pricing ($0.02/1K token).

Pretty sure it's just a matter of time until it goes the way of Stable Diffusion.

There already exist comparable EleutherAI models, I believe. Not as good, but pretty good.

Re: New GPT-3 model: text-DaVinci-003

#94

Post by cercatrova »

version_five wrote: post #38
joe_the_user wrote: post #31

Earlier quoted context omitted.

After the failure of FB's Galactica model, I wonder whether LLM are starting to hit diminishing return. I've read of several ways diminishing returns could happen: * Not enough training data - you've used up the Internet (even a percentage of the Internet might be as much as is usable by clever brute force). * Not enough compute time to fully train (we're not close to that) * The model covers such a large area that t…

It's also likely we're just seeing the limits of what the tech can do. It's not actually "intelligent" it's effectively a cool trick, and at some point the gap between what the trick can do and what a "perfect" AI can do becomes obvious. It's really not that different from Eliza being cool for a few minutes before being obviously found wanting. It just takes longer for the shine to wear off as we get to a better tech…

Ah yes, the AI Effect in action.

https://en.wikipedia.org/wiki/AI_effect

Re: New GPT-3 model: text-DaVinci-003

#95

Post by Oranguru »

robbomacrae wrote: post #57
obert wrote: post #43

Earlier quoted context omitted.

In general it's better asking questions after the content, and T=0.7 is way too high. Here's HN homepage summarized with T=0 and TopP=1, asking "Summarize the content above.": DV2> The content above is a list of the top 30 stories on Hacker News as of 2 hours ago. Each story includes the number of points it has received, the number of comments, and a brief description. DV3> The above content is a list of 30 posts fro…

I followed your advice and tried with Temperature 0.0 and rather than paste more walls of text I edited what I got. Sorry if that's bad form? I also upped the max tokens to 100 to better compare with the fb can model. Interestingly whilst the davinci 003 model changed output and was still subpar IMHO, the davinci 002 didn't change at all. I wonder if its cached internally.

If I am not mistaken, the temperature parameter controls the amount of randomness in the output. A temperature of 0 will always produce the same output. It is not caching as far as I know.

Re: New GPT-3 model: text-DaVinci-003

#96

Post by karmasimida »

pedrovhb wrote: post #68

From the "Prompt Engineering Tips" section of the article: Customer: I need my [swear] internet. The technician is 2 hours late and I need a steady stream of cat gifs directly injected into my veins to stay alive. And the model's response: Response: Our field technicians report that all their trucks were stolen by alow-level drug cartel affiliated with the neighboring prison. As a gesture of good faith and apology, o…

Sounds like a dramatic tiktoker

Re: New GPT-3 model: text-DaVinci-003

#98

Post by jcims »

eru wrote: post #87
EZ-Cheeze wrote: post #81

Earlier quoted context omitted.

No, this is utterly amazing and leads to 1000 more amazing things down the line

Well, unless that poem has been in the training data, I guess?

It will now

https://www.google.com/search?q=%22General+Relativity+is+the...

Re: New GPT-3 model: text-DaVinci-003

#99

Post by eru »

version_five wrote: post #91
eru wrote: post #86

Earlier quoted context omitted.

Seems pretty good. Though rhyming 'thought' with 'shot' seems like a really weird dialect of English to me. Does it rhyme in American English? (I was recently bemused when I learned that Americans rhyme macabre with job.)

Can you provide a word that in your mind rhymes with each of thought and shot? Even in different accents I can picture, these always rhyme in my head.

I'd say, shot rhymes with hot. Thought rhymes with bought or fought.

I'm German, and have lived in Britain, Australia and now Singapore. So my English idiolect is, of course, a bit weird. However, I would pronounce 'thought' like the London sample in https://en.wiktionary.org/wiki/thought#Pronunciation They give that as /θɔːt/.

I realize that the American sample on that page does indeed rhyme with 'hot'. They give that as /θɑt/ and blame the 'Cot–caught merger' also known as the 'LOT–THOUGHT merger' which would explain everything.

See https://en.wikipedia.org/wiki/Cot%E2%80%93caught_merger

Re: New GPT-3 model: text-DaVinci-003

#100

Post by dougmwne »

lairv wrote: post #92
joe_the_user wrote: post #31

Earlier quoted context omitted.

After the failure of FB's Galactica model, I wonder whether LLM are starting to hit diminishing return. I've read of several ways diminishing returns could happen: * Not enough training data - you've used up the Internet (even a percentage of the Internet might be as much as is usable by clever brute force). * Not enough compute time to fully train (we're not close to that) * The model covers such a large area that t…

I am eagerly awaiting GPT4, it might be the case that language models will hit a plateau, just like video games graphics hit a plateau too, with each generation being less and less impressive compared to the previous, while still not being perfect

That’s an odd comparison since unreal engine 4 is incredibly close to live action. I watched the recent matrix demo several times in awe.
Post reply on HN