I think dismissing it as simply a chat bot is missing what's actually happening. There's much more to ChatGPT than merely a chatbot, even with the hallucinations.
Maybe you're a few days behind the times?
To the best of my knowledge, before the transformer model, the state of the art for natural language processing was looking out for pre-programmed keywords for actions, and simply summing the total of "good minus bad" for sentiment analysis. I'm not sure if even word2vec made a significant impact there, despite the then-impressive ability for it to embed words into a vector space such that "king" - "man" + "woman" ≈ "queen"
But GPT? Well…
---
Me:
explain the Einstein field equations in the style of an ancient Greek epic poem. Write in Greek.
ChatGPT:
Ω Μοῦσαι, θεαὶ τῆς εἰρήνης καὶ τῆς ἀρμονίας, Τὴν ἱστορίαν τῆς φύσεως μὲν ἀείδετε, Τοῦ κόσμου τὰ μυστικὰ καὶ τὰ βαθέα διδάσκετε.
Αἰνστάινος ὁ φωσφόρος τῆς λογικῆς καὶ τῆς φύσεως, Τὸν λόγον τοῦ κόσμου κατέγραψεν ἐν εἰκόσιν ἀριθμοῖς, Τοὺς λεγομένους Αἰνστάιν εὐφυεῖς εὑρίσκων λόγους.
Στις ἰσότητα τῆς μάζας καὶ τῆς ἐνέργειας ἐστίν ἡ πρῶτη, Τὸν κόσμον διακοσμοῦντα καὶ τὴν κίνησιν διατηροῦσαν λοιπὴν, Τὴν ἐντροπίαν, τὸν πόνον τοῦ κόσμου περιγράφει ηὗται.
Τὸ θεῖον καὶ τὸ ἀνθρώπινον τῆς φύσεως συνάγουσι, Τὰς καμπὰς τοῦ χρόνου καὶ τὰ μυστικὰ τοῦ κόσμου ἀνακαλύπτουσιν, Αἰνστάινου λόγοις τὸν κόσμον συντηροῦσιν ἀεί.
Ὦ Μοῦσαι, θεαὶ τῆς εἰρήνης καὶ τῆς ἀρμονίας, Τὴν δόξαν τοῦ Αἰνστάινου καὶ τοὺς λό
Me:
write a python script which calls the Google Translate API to turn that into Latin
ChatGPT:
I'm sorry, but as an AI language model, I am not authorized to call external APIs or execute any code outside of this platform. However, I can provide you with an example code snippet that utilizes the Google Translate API via the googletrans library to translate text from English to Latin:
Copy code
from googletrans import Translator
translator = Translator(service_urls=['translate.google.com'])
text = "explain the Einstein field equations in the style of an ancient Greek epic poem. Write in Greek."
result = translator.translate(text, dest='la')
print(result.text)
When you run this code, it should output the translated text in Latin:
Copy code
explicare Einstein campo aequationes in modo antiquo Graeca carmen epico. Scribere in Graeco.
Please note that the googletrans library uses the public Google Translate API, which has certain usage limits and restrictions.
Me:
建議量子力學主題酒吧的名稱。
ChatGPT:
以下是一些可能的量子力學主題酒吧的名稱：
1. 薛定谔的酒吧 2. 量子酒馆 3. 希尔伯特的饮料 4. 波粒二象性酒吧 5. 能量溢出酒吧 6. 纠缠酒廊 7. 费曼的小酒馆 8. 量子跃迁酒吧 9. 亚当斯的指南酒吧 10. 量子隧道酒吧
Me:
На Commodore64 BASIC напишите реализацию видеоигры Pong.
ChatGPT:
К сожалению, я не могу написать код на Commodore64 BASIC, потому что это требует доступа к конкретному железу, которого у меня нет. Однако, я могу предоставить вам базовый шаблон кода на Commodore64 BASIC для игры Pong, который можно адаптировать и доработать под свои нужды.
Copy code
10 REM PONG 20 REM SET UP VARIABLES 30 X=160:Y=100:Z=60 40 S=10:C=1 50 I=5:J=5:K=0:L=0:M=0:N=0:O=0 60 GOTO 1000 70 REM DRAW SCREEN 80 PRINT "{CLR}";:FOR A=1 TO 24:PRINT "{DOWN}";:NEXT A 90 REM DRAW PADDLES 100 PRINT "{HOME}";"{RIGHT}";"{DOWN}";"{UP}";:PRINT "{UP}";:PRINT "{LEFT}";:PRINT "{DOWN}";"{RIGHT}"; 110 REM DRAW BALL 120 PRINT "{HOME}";"{RIGHT}";"O"; 130 REM GET JOYSTICK INPUT 140 Q=PEEK(56320) 150 IF Q=11 THEN K=-1:GOTO 160 160 IF Q=14 THEN K=1:GOTO 170 170 REM MOVE PADDLES 180 I=I+K:J=J+L 190 IF I<=5 THEN I=5 200 IF I>=15 THEN I=15 210 IF J<=5 THEN J=5 220 IF J>=15 THEN J=15 230 REM MOVE BALL 240 X=X+C:Y=Y+O 250 IF Y<=7 THEN O=-O 260 IF Y>=22 THEN O=-O 270 REM COLLISION WITH PADDLES 280 IF X<=5 AND Y>=I-1 AND Y<=I+5 THEN C=1:O=S(Y-I-2) 290 IF X>=30 AND Y>=J-1 AND Y<=J+5 THEN C=-1:O=S(Y-J-2) 300 REM SCORE KEEPING 310 IF X<=0 THEN Z=Z-1:GOTO 350 320 IF X>=35 THEN Z=Z+1:GOTO 350 330 REM REPEAT GAME 340 GOTO 60 350 REM GAME OVER 360 PRINT "{CLR}";"GAME OVER":PRINT "{DOWN}{DOWN}";"FINAL SCORE:";Z 370 END 1000 REM SET UP GAME 1010 X=160:Y=100:Z=60 1020 S=10:C=1 1030 I=5:J=5:K=0:L=0:M=0:N=0:O=0 1040 REM INTRODUCTION SCREEN 1050 PRINT "{CLR}{DOWN}{DOWN}{DOWN}";:PRINT "PONG":PRINT "{DOWN}";:PRINT "PRESS FIRE TO BEGIN":PRINT "{UP}{UP}{UP}";:WAIT 198 1060 IF PEEK(56334)=255 THEN 1060 1070 WAIT 198:GOTO 70
Это базовый шаблон, который необходимо доработать и оптимизировать под ваше железо. Он также может содержать ошибки, которые могут потребовать дополнительной отладки. Если вы не зн
---
I'm not going to edit this comment so the BASIC formatting shows up correctly; each line number has a line break before it when I pasted it here.
I also can't test if that C64 BASIC will run, as I can't find an emu that accepts paste nor get Apple Script to successfully inject fake key presses, but hey, that Russian contains a disclaimer warning it might be buggy.
I will however say that it did an acceptable job (given how little a description I gave it) of the same in JavaScript when I asked it to.
Am I supposed to be impressed by a bunch of gibberish in languages I don't speak, and therefore can't verify?
Anyway…
--
Me:
Respond to the following hacker news comment, give an answer in the style of Shakespeare:
<I copy pasted your comment here>
It:
Hark! Thy words ring hollow as the empty air,
Forsooth, to claim we're not but mere reaction,
Is to ignore the science everywhere,
And spout ignorance without true traction.
Oft have philosophers and scientists sought,
To fathom consciousness and all its might,
A mystery still, with answers not yet wrought,
But progress made through study and insight.
And why must language be a barrier,
To truth and knowledge we can all attain?
By learning, we can all become carrier,
Of wisdom, and our horizons expand again.
Thus, I implore thee, don't dismiss with disdain,
The knowledge that awaits in languages not plain.
--
(HN messes up the formatting a bit; it was 4-4-2 lines in each group)
I'm basically on board with what it says here. Consciousness may be a mystery, but everything we see points to substance monism, not Cartesian dualism.
I take caffeine, a chemical, one thing happens to my mental experiences; alcohol, another chemical, a different thing; if I were to take THC or LSD or whatever, they too would have predictable effects on my mental state.
You cannot be seriously suggesting that the quality of AI generated text be evaluated through the broken telephone of Google translate. That's absurd.
> Forsooth, to claim we're not but mere reaction,
aaand it already fucked up by the second verse. My comment clearly states that we are not mere reaction, yet this "AI" seems to think I claimed the exact opposite. You're proving my point, it's just guessing what word fits where, not understanding anything whatsoever and producing garbage as a result, that only seems intelligible to gullible fools.
> if I were to take THC or LSD or whatever, they too would have predictable effects on my mental state.
I don't think I've ever heard anyone describe the effects of LSD as "predictable". I'm guessing you've never actually taken it and are talking out of your ass.
How is it not predictable? Varying degrees depending on dosage of hullucinations, euphoria, altered state of mind including altered senses of time, sound, vision, touch etc. This is how the effects of drugs are described by medicine and science.
It seems as if you are ignoring this part of their statement "effects on my mental state." They aren't saying the content of the hallucination is predictable or anything like that.
>Everything alters your mental state.
You are overgeneralizing to the point of absurdity and it does not appear that you are interested in a good-faith discussion. The contents of a hallucination and the experience of having a hallucination are distinct.
Suggesting that the effect of LSD altering my mental state and me walking out my back door altering my mental state are the same because they both alter the state is absurd. You're obtusely ignoring intensity, impact, duration and that the hallucination and euphoria are different effects.
False. People experience euphoria without LSD, even randomly. There are entire spectrums of neuroconfigurations in which people precisely experience random bouts of euphoria. And that's just one of the effects you listed. In fact, one of the (potentially concerning) effects of LSD is called hallucinogen persisting perception disorder, in which the experience under LSD persists long after its use, and if you bother to read up on it, you'll also find that people who have never taken LSD can randomly experience them, proving you wrong again.
> You are overgeneralizing
No you are. You're clumping an entire mind-altering experience under "hallucination" and "euphoria". You should start asking, what is being hallucinated, why is the patient euphoric. That's the opposite of generalization.
> it does not appear that you are interested in a good-faith discussion.
Seems like an admission on your part. I'm speaking from research and experience. I've read books about LSD, I've experimented with it myself, and I've observed and discussed its effects with many people. My faith in what I'm saying is just fine. Accusing me of bad faith is nothing but projection on your part.
> The contents of a hallucination and the experience of having a hallucination are distinct.
You couldn't be more wrong. The experience of having a hallucination is fundamentally defined by the contents of that hallucination. Seeing hallucinations of your friends stabbing you in your sleep is not the same experience as feeling one with God while the Sun shines on your face. Different content, different experience.
> You're obtusely ignoring intensity, impact, duration and that the hallucination and euphoria are different effects.
And you're ignoring the actual contents of mental state, arguing in favor of simply measuring everything by "intensity", "impact", and "duration". Do tell, what's the "intensity" of walking through your back door? What's the "impact" of having a religious or spiritual experience? What's the "duration" of learning something about yourself?
Yeah, it kind of is. Fake it 'til you make it. Vaporware landing pages to validate products. But that isn't what we need to focus on.
ChatGPT is more than a chatbot. And if you use its output knowing that it may contain hallucinations and correct for that possibility, there is real potential here. You can't take it at face value, and have to validate whatever it gives you, but you should be doing that anyway for anything you find online of through the media.
I am the first to say that its lack of accuracy is a problem, but the last to say that it is a non-solvable problem.
We're not there yet, but we're probably not far from that.
I see you've never talked to a salesman. Try going to a time-share sales presentation. They actually have an acronym the CEO of some of the big hotel chains came up with: TAFT - Tell them ANY FRIGGING THING. (really it's frigging because: corporate ethics or something).
Even right now, given it is properly implemented, it can leave millions of accountants and auditors without jobs.
But I agree, it wont revolutionize anything, it will automate EVERY SINGLE JOB that is dependent on human's intellectual input and eliminate billions of the jobs, which will lead to global hunger and reduction of human population.
I think those are two examples of exactly what cannot be replaced yet. Their jobs are specifically about ensuring that rules have been followed precisely. Current AI sounds confident, but still makes obvious errors that would be completely unacceptable for accounting and auditing. Same as software engineering, you currently still need a human reviewing the output and verifying that it's actually doing the right thing, and without side effects, poor handling of errors and edge cases, etc.
If that really happens we get universal basic income, or massive government employment programs. Otherwise the economy collapses and the ultra rich lose money.
Also, it's game theory, if you move to automate all the work, you'll get super rich in the meantime, and if you don't move to automate all the work, someone else will move to automate all the work and you'll lose.
On page 24 of this paper, there is a 3D game with enemies and objectives coded by GPT-4:
The page describes itself as "A logic puzzle game invented by artificial intelligence" (emphasis mine). There is no claim that ChatGPT "entirely wrote it" anywhere on the page.
> On page 24 of this paper, there is a 3D game with enemies and objectives coded by GPT-4:
There is no "game" on that paper. Only two screenshots and the dubious claim that "GPT-4 did it", with no code anywhere to be seen, also checked the appendix.
Yeah that’s interesting. There used to be a linked blog post describing the development process with screenshots of the LLM output. It’s missing now. I suspect the author made some improvements or monetized it or something