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

Low-code, no-model recommendation system with GPT3

medium.com

Post reply on HN
Print view View original

31–40 of 67 posts

Re: Low-code, no-model recommendation system with GPT3

#31

Post by dtech »

dharma1 wrote: post #29
superasn wrote: post #23

I wish there was a light version of gpt3 with maybe like 100 api query limit per day so just amateurs like me got to play with this API too on weekends. I have no agenda of changing the world and I have nothing substantial to fill in the gpt3 invite form but sometimes the best ideas come from tinkering and I just wish they gave some limited access to everyone.

I thought they have a free tier from October? https://twitter.com/hardmaru/status/1301362995356774401 Not sure what the 100k tokens translates to in terms of query limit

That's 3 month max though

Re: Low-code, no-model recommendation system with GPT3

#32

Post by m12k »

"Recommendation system is so successful in many products and services we interact everyday" ->

"Recommendation system[S] [ARE] so successful in many products and services we interact [WITH] every[ ]day"

I'm sorry, but four errors within the first sentence is just not good enough - did the author even read this themselves just once before hitting publish? Show just a little bit more respect toward the people who give their attention to what you've written.

Re: Low-code, no-model recommendation system with GPT3

#33

Post by plutonorm »

The_rationalist wrote: post #25
matusp wrote: post #4

I am curious how much it actually thinks and how much it simply uses text from recommendation engines that got crawled.

A neural network can't think, please stop cargo culting neural networks

Panpsychism is a serious philosophical position.

Re: Low-code, no-model recommendation system with GPT3

#34

Post by mparramon »

m12k wrote: post #32

"Recommendation system is so successful in many products and services we interact everyday" -> "Recommendation system[S] [ARE] so successful in many products and services we interact [WITH] every[ ]day" I'm sorry, but four errors within the first sentence is just not good enough - did the author even read this themselves just once before hitting publish? Show just a little bit more respect toward the people who give…

Maybe the author's first language is not English?

Re: Low-code, no-model recommendation system with GPT3

#35

Post by m12k »

mparramon wrote: post #34
m12k wrote: post #32

"Recommendation system is so successful in many products and services we interact everyday" -> "Recommendation system[S] [ARE] so successful in many products and services we interact [WITH] every[ ]day" I'm sorry, but four errors within the first sentence is just not good enough - did the author even read this themselves just once before hitting publish? Show just a little bit more respect toward the people who give…

Maybe the author's first language is not English?

Sure, neither is mine. But if you don't have the skills to proofread what you've written yourself, then get someone to help that has. I don't expect to be able to write something in Chinese deserving of a wide audience all on my own either.

Re: Low-code, no-model recommendation system with GPT3

#36

Post by The_rationalist »

plutonorm wrote: post #33
The_rationalist wrote: post #25

Earlier quoted context omitted.

A neural network can't think, please stop cargo culting neural networks

Panpsychism is a serious philosophical position.

Could you please give an explanation of panpsychism applied to software that make sense? Spoiler: you cannot and it's totally refuted empirically. If I code a hello world, what kind of cognitive abilities does this have?

BTW I guess panpsychism researchers should start talking to rocks maybe one day they'll find one that is an AGI! /s

Re: Low-code, no-model recommendation system with GPT3

#37

Post by lt »

strombofulous wrote: post #13
rococode wrote: post #8

Earlier quoted context omitted.

For anyone else curious, the pricing according to gwern [1]: Explore: Free tier: 100K [BPE] tokens, Or, 3-month trial, Whichever comes first Create: $100/mo, 2M tokens/mo, 8 cents per additional 1k tokens Build: $400/mo, 10M tokens/mo, 6 cents per additional, 1k tokens Scale: Contact Us Expensive but (unfortunately) pretty standard market price for cloud NLP stuff. The price is about $40 per million tokens. For compa…

What is a token? One character? Or is it a currency?

A character, both input and output.

Re: Low-code, no-model recommendation system with GPT3

#38

Post by throwaway287391 »

lt wrote: post #37
strombofulous wrote: post #13

Earlier quoted context omitted.

What is a token? One character? Or is it a currency?

A character, both input and output.

Not exactly, GPT-3 uses a variant of BPE [1], so one token can correspond to a character, an entire word or more, or anything in between. The paper [2] says a token corresponds to 0.7 words on average.

[1] https://en.wikipedia.org/wiki/Byte_pair_encoding

[2] https://arxiv.org/abs/2005.14165, page 24

Re: Low-code, no-model recommendation system with GPT3

#39

Post by plutonorm »

The_rationalist wrote: post #36
plutonorm wrote: post #33

Earlier quoted context omitted.

Panpsychism is a serious philosophical position.

Could you please give an explanation of panpsychism applied to software that make sense? Spoiler: you cannot and it's totally refuted empirically. If I code a hello world, what kind of cognitive abilities does this have? BTW I guess panpsychism researchers should start talking to rocks maybe one day they'll find one that is an AGI! /s

"Could you please give an explanation of panpsychism applied to software that make sense? Spoiler: you cannot and it's totally refuted empirically. If I code a hello world, what kind of cognitive abilities does this have?"

You aren't making any sense, go read about the things you are talking about.

Re: Low-code, no-model recommendation system with GPT3

#40

Post by jonathanstrange »

m12k wrote: post #32

"Recommendation system is so successful in many products and services we interact everyday" -> "Recommendation system[S] [ARE] so successful in many products and services we interact [WITH] every[ ]day" I'm sorry, but four errors within the first sentence is just not good enough - did the author even read this themselves just once before hitting publish? Show just a little bit more respect toward the people who give…

Personally, I don't care about form if I find the content interesting.
Post reply on HN