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

51–60 of 216 posts

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

#51

Post by malloryerik »

rpjt wrote: post #26
nl wrote: post #22

Earlier quoted context omitted.

This is a lot more wide-spread than "some kids on Reddit". Maybe 30% of the kids in my son's class are using this or related tools.

Any ideas on how to solve this issue of kids cheating with GPT3 essays?

Test the kids on their own essays, for example? Maybe this could itself be automated with GPT-3?

The highest-quality answer involves skilled teachers with enough time who know and understand their students. (Actually the very highest might involve personal tutors but let's leave that aside.)

Going down a few steps you might combine the automated approach with skilled teachers and maybe add human editors who can do support work asynchronously?

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

#52

Post by saintsfc »

typpo wrote: post #37

In case it interests anyone, I built a document editor + GPT and updated it to use text-davinci-003: https://github.com/typpo/arkose/ If you want to try GPT-3 but don't have an OpenAI API key, I've set up a quick demo here until I hit my billing cap (normally users would supply their own API key): https://arkose.pages.dev/

Very cool. When you hit the generate key sequence mid way through a document, how much of that text is included in the prompt?

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

#53

Post by janalsncm »

rpjt wrote: post #26
nl wrote: post #22

Earlier quoted context omitted.

This is a lot more wide-spread than "some kids on Reddit". Maybe 30% of the kids in my son's class are using this or related tools.

Any ideas on how to solve this issue of kids cheating with GPT3 essays?

Realistically? Grade based on thought process and validity of the argument, not whether it has spelling or grammar mistakes. GPT3 is still pretty incoherent over the span of enough text.

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

#54

Post by johnfn »

robbomacrae wrote: post #44
johnfn wrote: post #20

I've been using GPT-3 recently to give me ideas for my nanowrimo novel. I'll just paste in the last couple of paragraphs that I wrote and see what it thinks I should write next. Most of the time, the responses are pretty silly (it gets into loops a lot or doesn't know where to take the story) but a couple of times I've gotten some really good lines out of it. It kinda feels like copilot, but for writing!

I've always been in love with models. But seriously between this, stable diffusion, and co-pilot.. what a time to be alive! I hope this takes the sting off of all those folks laid off. Just think of all the projects that can be accelerated with these tools. Personally I've got co-pilot helping me build an application using GPT3 summarizations and I'm working on a board game with stable diffusion supplying the artwork…

I completely agree. Have you tried out the new Midjourney v4? It's completely blown my mind when it comes to game art generation. MJ v4 is to Stable Diffusion as Stable Diffusion is to me drawing art by hand :)

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

#55

Post by mgraczyk »

blindot wrote: post #32
oldgradstudent wrote: post #17

Earlier quoted context omitted.

Have it compose a poem — a poem about a haircut! But lofty, noble, tragic, timeless, full of love, treachery, retribution, quiet heroism in the face of certain doom! Six lines, cleverly rhymed, and every word beginning with the letter S!!

Sleek strands of sadness, soon severed and shorn; Solace sought in scissors, slicing and scorned. Sublime sorrow, set sorrowfully still; Stoic heart to surrender, so swiftly and shrill. Secrets spilled in soft, sorrowful swish Sorrowful sounds, soon silenced with snip.

Damn that's bad poetry

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

#57

Post by robbomacrae »

obert wrote: post #43
robbomacrae wrote: post #34

I've been trying out various models for summarization and I have to say, on first glance, this davinci-003 doesn't impress compared to davinci-002 or facebook/bart-large-cnn. If you throw in "summarize the following text:" and copy in the contents of https://en.wikipedia.org/wiki/Hacker_News you get the following (temperature 0.0 max tokens 100)[0]: davinci-003 Hacker News is a social news website focusing on compute…

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.

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

#58

Post by typpo »

saintsfc wrote: post #52
typpo wrote: post #37

In case it interests anyone, I built a document editor + GPT and updated it to use text-davinci-003: https://github.com/typpo/arkose/ If you want to try GPT-3 but don't have an OpenAI API key, I've set up a quick demo here until I hit my billing cap (normally users would supply their own API key): https://arkose.pages.dev/

Very cool. When you hit the generate key sequence mid way through a document, how much of that text is included in the prompt?

This is configurable in Settings, but by default it sends the last ~1024 characters and asks the AI to generate up to 256 tokens.

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

#59

Post by kian »

>>>compose a poem — a poem about the interpretation of programming expressions! But lofty, noble, tragic, timeless, full of love, treachery, retribution, quiet heroism in the face of certain doom! Six lines, with clever rhymes and enjambment!

A language of code, the expression of thought

A complex web of operations tightly wrought

It speaks of a power, a strength from within

Though often misunderstood, the answer is within

A marvel of logic, of reason and rhyme

A careworn path forged by one who has time

It speaks of a courage, a spirit that's strong

Its secrets and wonders will never be wrong.

------------------

Not six lines, and no enjambment, but I almost like the first two lines. Middle School Poetry contestants, watch your step! (and middle school teachers, too).

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

#60

Post by robbomacrae »

johnfn wrote: post #54
robbomacrae wrote: post #44

Earlier quoted context omitted.

I've always been in love with models. But seriously between this, stable diffusion, and co-pilot.. what a time to be alive! I hope this takes the sting off of all those folks laid off. Just think of all the projects that can be accelerated with these tools. Personally I've got co-pilot helping me build an application using GPT3 summarizations and I'm working on a board game with stable diffusion supplying the artwork…

I completely agree. Have you tried out the new Midjourney v4? It's completely blown my mind when it comes to game art generation. MJ v4 is to Stable Diffusion as Stable Diffusion is to me drawing art by hand :)

No I hadn't. Thanks I'll check it out! Of course the open source nature of SD is really nice as the community[0] keep pumping out usability improvements and features.. like the Pokemon trained model[1] which is an absolute blast for my kid!

[0]: https://github.com/invoke-ai/InvokeAI

[1]: https://huggingface.co/justinpinkney/pokemon-stable-diffusio...

Post reply on HN