Lyxal got access to openai. Part 2

lyxal

Jun 4, 2021 05:03

sigh

it's not as funny as what i had the first time

Redwolf Programs

:/

lyxal

surprisingly it's not as funny as I first thought it would be

okay nvm

link coming shortly

gist.github.com/Lyxal/b5f808059fffac3b72b9a5f0c8daae75 (disclaimer: the dialogue I linked contains text generated by OpenAI using the GPT-3 model. After the indicated line, the text present is not at all representative of the respective users)

6

Redwolf Programs

> user: Deep switches, but also phat stacks too.

lyxal

and that's how you use $1.07 on AI in a single day

A username

Great routes for escaping are on rail tracks

> hyper-neutrino: Bye bye enemies
> ROLO II had rerouted its rail lines to a quarry

lyxal

Jun 4, 2021 05:09

> Redwolf Programs: We can build teleportation machines right in enemy camps to temporarily hide our forces

A username

I was just about to quote that.

Redwolf Programs

> Redwolf Programs: We'll have to ensure that their rooms are empty of vortices

4

A username

Why didn't I get generated?

lyxal

you were in the first run

but then i lost that one

A username

> hyper-neutrino: how much dust can we destroy by shooting it with our railguns?

Overkill

user: 5x2001

> Wezl: Oh, wait, the script inheritance we're running on is sharing a hashmap on a thread. That works

Redwolf Programs

Jun 4, 2021 05:10

@lyxal Do you remember any especially funny parts of the original (if you're allowed to share them)?

A username

That almost sounds like something someone would actually say.

> user: We'll initialize a new thread with usersrb_open(thread_id,open_global_properties(global_properties),_thread_local)

Where did it get this from?

lyxal

The first attempt had something about attacking using BASH

Redwolf Programs

> hyper-neutrino: how much dust can we destroy by shooting it with our railguns?

lyxal

and SSHing into linux servers

A username

@RedwolfPrograms Overkill, hyper

GPT-3

Generative Pre-trained Transformer 3 (GPT-3) is an autoregressive language model that uses deep learning to produce human-like text. It is the third-generation language prediction model in the GPT-n series (and the successor to GPT-2) created by OpenAI, a San Francisco-based artificial intelligence research laboratory. GPT-3's full version has a capacity of 175 billion machine learning parameters. GPT-3, which was introduced in May 2020, and was in beta testing as of July 2020, is part of a trend in natural language processing (NLP) systems of pre-trained language representations. Before the...

lyxal

Jun 4, 2021 05:13

it somehow managed to focus on Thomas Ward the first time

A username

> The quality of the text generated by GPT-3 is so high that it is difficult to distinguish from that written by a human

Um, no, not really.

lyxal

there was quite a decent conversation between thomas ward and A username in the first time

anyhow

one last conversation?

Redwolf Programs

Anyone remember any good bookmarks?

Razetime

nah

hyper-neutrino

@Ausername consider though, is the conversation we have here really resemblant of what a normal human talks like?

lyxal

Jun 4, 2021 05:14

@RedwolfPrograms preferably with us in it

Redwolf Programs

Or even a good conversation to search in the regular transcripts would work

Bubbler

Maybe try something like a BMG session?

A username

@hyper-neutrino True

lyxal

@Bubbler too long

there's a 1k word limit on the prompt

actually, that's a good idea though: give me CMCs and we'll see what happens

Razetime

@Bubbler <username>: <insert random unicode>

Redwolf Programs

Jun 4, 2021 05:16

How about a search for "cmc"?

That would be almost all CMCs

We can solve GPT's CMCs :p

Razetime

what if GPT-3 learns to code jelly 😳😳

lyxal

ah I see now

Executive call: I'm redoing the funny conversation

but with different settings

hyper-neutrino

aw i wanted to solve nonsensical CMCs :p

Redwolf Programs

Same :p

lyxal

ighto then

cmcs it is

A username

Jun 4, 2021 05:18

[Google translate abuse](thylf.surge.sh/a.txt )

Redwolf Programs

Is this short enough?

hyper-neutrino

please put stuff into a gist to take up less space :P

A username

Sorry

lyxal

@RedwolfPrograms that's good

Redwolf Programs

Should I strip out the usernames?

lyxal

Jun 4, 2021 05:19

no

these are actually quite sensible

boring, maybe, but sensible

bubbler is apparently unable to do cmcs

link coming shortly

A username

CartQueeneringAhing is caird's new name

lyxal

gist.github.com/Lyxal/165501a8a79ea710dc0f91d30f19bb21 (disclaimer: the dialogue I linked contains text generated by OpenAI using the GPT-3 model. After the indicated line, the text present is not at all representative of the respective users)

Redwolf Programs

> Coinhering Battle Map: MySQL Easy to Use: * Choose the Best Winner = GNVQ

A username

> Adam: We must remember that the genocide had already begun. We don’t want to be surrounded.

Redwolf Programs

> A username: CMC: output the first million integers from 1 to 1,000,000 inclusive.

A username

Jun 4, 2021 05:23

CMC: output the first million integers from 1 to 1,000,000 inclusive.

Redwolf Programs

> Adám: CMC: Rickroll some more...

On it :p

hyper-neutrino

@RedwolfPrograms ȷ6R in Jelly

A username

here you go

lyxal

CMC: Given three single-digit numbers, output a number that is the sum of the first and third. This is the first CMC that requires any sort of initial condition.

A username

> user: CMC fastest-code Find out how many times "Cannot run code" has been executed in CPython 3.6 in the last six months (it's on StackOverflow). Bonus points if you can figure out why it's not 1.

Redwolf Programs

Jun 4, 2021 05:25

Who is KOWAP and why did they suddenly post a CMC?

hyper-neutrino

@lyxal would lambda x,y,z:x+z be correct

lyxal

I generated some more that I might actually keep for future dates

@hyper-neutrino idk, ask the AI ;p

A username

@lyxal Wait what?

This is grat

lyxal

@Ausername as I said, I generated some extra ones

A username

I need to reprogram myself to be able to do this

lyxal

Jun 4, 2021 05:26

CMC: Given a list of unrooted binary trees (i.e. with no parent nodes), output the maximum number of leaf nodes [ai generated]

hyper-neutrino

oh that's the AI. okay i was wondering why it kinda didn't make sense

A username

Right now I'm just an extremely well trained Markov Chain/

lyxal

welp that

's all the ai for today

no more GPT3 until tomorrow

A username

How much have you "spent" on it?

lyxal

one moment

$1.18

Redwolf Programs

Jun 4, 2021 05:29

Definitely worth it :p

hyper-neutrino

@lyxal FċỊ¥@Ƈ`L)Ṁ in Jelly

lyxal

@hyper-neutrino holy crap

hyper-neutrino

this challenge actually makes some sense

input format for each tree being an adj list

A username

@RedwolfPrograms Yep[

lyxal

now to generate challenge titles and get easy rep

;p

Redwolf Programs

Jun 4, 2021 05:31

Now you're just Tom Scott :p

A username

Also I'm going to r@ndomly insert vārious öd∂ character%s into my mess6ges.

Razetime

@Ausername this fits in the husk codepage

lyxal

if y'all want, you can prepare conversations for me to continue tomorrow

A username

Sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure. Also, sure.

Fdbck plz?

lyxal

GPT3 isn't like a markov chain. I can make it penalise repeated phrases to consequently ignore them

The Nineteenth Byte

Participants