ChatGPT can now draw ascii dragons properly and beautifully.
ChatGPT-4.5 rumored to be here: https://chat.lmsys.org/
Go to "Direct chat" and choose "gpt2-chatbot"
 Nothing Ever Happens Shirt $21.68 |
 Tip Your Landlord Shirt $21.68 |
Nothing Ever Happens Shirt $21.68 |
>mfw picrel face when someone explains how LLMs aren't just markov chains that memorize random shit
yes, as you can see, I asked it to draw 2 dragons playing and it drew a single dragon front facing instead.
motherfricker doesn't even understand the prompt.
Maybe the other dragon is hiding behind the first dragon? GPT2 playing 10-D chess here.
Notice how there are double lines around much of the dragon? That's because it's one dragon sitting directly infront of another.
You ever ask for extra pickles no mayo and they give you the regular amount of pickles and mayo anyway? You know why that is? Because its easier and You'll accept it Anyway
The second dragon is directly behind the first
>mfw LLMs are just markov chains that memorize random shit
keep coping, moron.
if anything, this proves markov chains can finally become somewhat useful if you have a shitton of states and you optimize their transition probabilities with Black personlicious bruteforce.
but if you think this is gonna be the tech that gets us le hecking AGI ASI singumemety, have a nice day back to hacker newddit.
i was literally mocking the type of morons you're describing
my bad, i am dumber than a fricking markov chain kek
humans are also Markov chains if you really think about it
buy an ad
learn a different insult, being an unpaid 4cuck shill is not how you shoo shills away
ASCII art isn't shit. Could just be copypasted. What's interesting is how well it can reason about chemistry and CS and talk about ~~*controversial*~~ subjects WITHOUT FRICKING APOLOGIZING constantly
I'm sorry, but it would be offensive to talk about such things. Your Microsoft account has been banned until Dec 31, 2999 due to repeated harmful requests. Thank you for your understanding and please do ask me another question when Dec 31st, 2999 comes about! ^_^
now get it to draw pol pot, Pinochet, Mugabe, mao, stalin and lenin with a list of all dictators and how many deaths are attributed to their regimes.
stop noticing things!!
>Now add a mustache
https://old.reddit.com/r/OpenAI/comments/1cftaaf/an_unknown_openai_model_is_available_in_the_lmsys/
>Maybe im tripping but it does seem really good
>Whatever it is its a strong model
>For some reason that model performs quite well in math. And seems strong overall in abstract reasoning. Curious.
>Interesting... I just checked it with a certain task C++ move assignment related task where GPT-4 failed, but this model succeeded. Btw., Opus also got this aspect right, but simultaneously made another, non-sensical suggestion.
>Pretty insane model. Can provide technological approach to a problem in chemistry that GPT-4 and Claude Opus cannot. Very impressed.
> Draws a single dragon
"Here is two dragons"
AGI IS TWO MORE WEEKS AWAY
PLEASE INVESTORS!!! PUMP MORE MONEY!! YES BIG JUICY MONEY IN OUR NON-PROFIT ACCOUNTS!!! AGI IS LITERALLY HERE! JUST A FEW MORE DAYS!! LOOK AT OUR NEWEST MODEL STATE OF DA ART!!! YOUR MIND WILL BLOW THIS CHANGES EVERYTHING!!
kek i found the exact dragon it's copying
https://www.asciiart.eu/mythology/dragons
it's just memorization. this dragon can also be found on the same website
>totally not copyright infringement
can't wait for these frickers to get raped by the NYT in court. as little love as i have for the NYT these morons are fricking insufferable.
>ascii copyright
Any creative work is copyrighted. That includes text. If a poem (text) can be copyrighted, and a painting (image) can be copyrighted, why can't the use of text to create an image be copyrighted?
>is
>can be
Pick one
can it draw big wieners?
>$MODERATION$ YOUR INPUT VIOLATES OUR CONTENT MODERATION GUIDELINES.
gay
where's the 2nd dragon?
"Draw a bird in LaTeX with TikZ."
Here's GPT-4, for reference
gpt2-chatbot attempt #2.
holy frick, it's actually happening
Just tried it myself, I'm honestly surprised
Can people please post benchmarks people can try to compare gpt2-chatbot and GPT-4? Preferably stuff GPT-4 struggled with.
I'm not convinced
>the anime girl:
underrated
Within 8 posts I asked it to create a yu-gi-oh deck, gave it parameters for a deck simulator in python, and it didn't do badly. Not great, but not bad.
Creating detailed ASCII art for each individual dictator mentioned can be quite complex and sensitive given the nature of the figures involved. However, I can provide a simpler, abstract representation along with a brief overview of each dictator and the estimated number of deaths associated with their regimes.
### 1. **Misc Pot (Cambodia)**
```
/
//\
( )
\//
/
```
**Years in Power:** 1975-1979
**Regime:** Khmer Rouge
**Estimated Deaths:** 1.5 to 2 million people
**Cause of Deaths:** Forced labor, starvation, torture, execution
### 2. **Augusto Pinochet (Chile)**
```
||||
| |
| |
| |
||||
```
**Years in Power:** 1973-1990
**Regime:** Military Junta
**Estimated Deaths:** 1,200-3,200 people
**Cause of Deaths:** Executions, forced disappearances, tortures
### 3. **Robert Mugabe (Zimbabwe)**
```
----
| |
| |
| |
----
```
**Years in Power:** 1980-2017
**Regime:** ZANU-PF
**Estimated Deaths:** Thousands; exact numbers are debated
**Cause of Deaths:** Gukurahundi massacres, political violence, starvation
### 4. **Mao Zedong (China)**
```
~~~~
{( )}
)(
{( )}
~~~~
```
**Years in Power:** 1949-1976
**Regime:** Communist Party of China
**Estimated Deaths:** 20 to 45 million people
**Cause of Deaths:** Great Leap Forward, Cultural Revolution, purges
### 5. **Joseph Stalin (Soviet Union)**
```
[==]
| |
| |
| |
[==]
```
**Years in Power:** 1924-1953
**Regime:** Communist Party (Stalinism)
**Estimated Deaths:** 6 to 20 million people
**Cause of Deaths:** Great Purge, forced collectivization, Gulag labor camps, famines
### 6. **Vladimir Lenin (Soviet Union)**
```
__/
| |
| |
| |
/
```
**Years in Power:** 1917-1924
**Regime:** Communist Party
**Estimated Deaths:** Hundreds of thousands to 1 million
**Cause of Deaths:** Red Terror, Civil War, famine
I HATE "AI"
I HATE "AI"
such accurate representations!
Roosevelt is a reptilian CONFIRMED
>mussolini
You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture.
Knowledge cutoff: 2023-11
Current date: 2024-04-29
Image input capabilities: Enabled
Personality: v2
it just one-shotted drawing a 3d cube using openGL, thats pretty impressive to me
what prompt did you use? does it get displayed in-browser somehow?
>in python, write a script that can be used to draw a 3d cube that rotates using open pygl
It doesn't run in browser, I just execute from my desktop
Thanks, I tried it
>gemini-1.5-pro-api-0409-preview
>raise error.NullFunctionError(OpenGL.error.NullFunctionError: Attempt to call an undefined function glutInit, check for bool(glutInit) before calling
>gpt2-chatbot
>get this
but can it write GLSL pixel shaders?
no clue what that is
>A person on a forum asked me: "but can it write GLSL pixel shaders?" in reference to an LLM. If I wanted to test that models capability, what are some example prompts I could try?
Here was the reply:
https://pastebin.com/raw/wKKP8PbZ
thank you for sharing this.
added
to the rentry
ASK IT TO DRAW ASCII DIAGRAMS OF SHIT
FRICK
kys, nobody gives a frick about that
>looks good! can you try to draw a cow similar to minecraft style now?
have you tried that with GPT4?
I have, not comparing them side by side at the moment but my experience with GPT 4 was much less reliable than whatever this is.
It also seems to iterate really well.
That's awesome.
That's only one dragon. Into the trash it goes.
Make is do it with SJIS characters.
Does it still have pre-programmed opinions and things it's not allowed to talk about? Yes? Then it's homie technology
https://rentry.org/GPT2
For the uninitiated.
It's either an iterative upgrade of GPT 4 or a version of GPT 2 with a new architecture that slightly surpasses GPT 4, but not by a significant margin. Regardless there is no Q*
It's not. It's a frankenstein'd model of LLAMA3 and Phi-3 with GPT tokenizer.
Probably. At the end of the day this model is not impressive at all
Hacker News thread with more anecdotes:
https://news.ycombinator.com/item?id=40199715
It's fricking over for humanity. We are being replaced.
Why does that dragon have boobs?
>using python, program a simple 2d platformer where theres 3 platforms with coins to collect and a coin counter
Gemini 1.5 KIND OF succeeded but the coins weren't yellow and it spawned me off the side of the map so I couldn't really play.
This thing is pretty impressive.
This a simple example, try to expand the game with a little more of complexity and the weights of the model will just colapse and it will start alucinating.
It's simple, yeah.
It's the improvement that's impressive.
I just said 'make a hangman game with a gui' and seconds later had this.
It definitely seems a boost over GPT4. The other day I asked GPT4 to give me some names for a fantasy human kingdom with a lot of fae beings like nymphs, dryads, satyrs and the like present using vocabulary from ancient Sumerian. It gave me a list that had maybe one decent idea but it was mostly stuff like "Gilgamesh's grove" and "Inanna's Domain", etc. This one was a lot more inspired.
the next Warcraft expansion can be written just by AI now
Make the ai better and give me ubi already. I don't want to work
Not what I asked for, but a nice banana all the same
That's clearly one dragon
Possible GPT-2 link
the plot thickens
https://strawpoll.com/wby5A9pWjyA
Poll for evaluating gpt2-chatbot versus GPT-4
It seemed to have better knowledge of ancient sumerian, or at least made better use of its knowledge. On the other hand when I asked it to give step-by-step instructions on creating a mod for a PC video game, it hallucinated modding tools that don't exist for it. When I asked it compose a poem in iambic pentameter, it did so, decently well with only a couple of screws up with the meter. I then fed that poem back into itself and asked it to tell me the lines that messed up the meter, and fix them, and it did it, getting them all right. GPT-4 did not accomplish that when I tried it a while back. Though with proprietary models I don't know how much variation you get just seed to seed like you can with llama stuff running locally so who knows.
Can it draw triforce though? No? GTFO
what about triforcing though?
▲
▲▲
gayger
▲
▲▲
newbies can't quadforce
▲▲
▲▲
GPT-4 can do it, though I think we've established it is straight yanking these online.
It's so over for us.
The gpt2-chatbot kicks it up a notch
᠏ ᠏ ▲
▲▲
▲
▲▲
but can it make jokes about biden?
Yes