ulopeFeb 17, 2023
Michael Gartenberg

The stages of playing with GPT-3:

- OMG this can do anything
- There goes my job
- I should start a business around this
- Some of the responses aren’t too good
- Actually, some of these responses are just awful
- This isn’t really intelligence
- This is just spicy autocomplete

(as was forwarded to me w/no attribution)

Feb 14, 2023 at 6:01pmtrunks.social
Show threadChristina WarrenFeb 14, 2023
@Gartenberg I’d add a final one “this really does make a lot of my life easier, even if it is imperfect” — I can tell you that after using GitHub Copilot for two years, I could go back to life without it, but I really wouldn’t want to.
Show threadAkshay | అక్షయ్Feb 19, 2023
@film_girl @Gartenberg GitHub copilot switches the action from typing stuff to correcting stuff. The latter sometimes can be a lot harder than the former. 🙂
Show threadMark KahnFeb 14, 2023
@Gartenberg was this written by an AI? :)
Show threadMichael GartenbergFeb 14, 2023
@marksma 😂🤣😂
Show threadhammancheezFeb 14, 2023
@Gartenberg unfortunately between step 3 and step 7 a lot of people WILL lose their jobs while the market engages in its usual frothy bullshit hypewave
Show threadBornachFeb 15, 2023

@hammancheez @Gartenberg
Helped on by influencers

At what stage is Tom Scott?
https://youtu.be/jPhJbKBuNnA

I tried using AI. It scared me.

YouTube
Show threadChristian Kent   𝘊𝘒 :＼﹥Feb 15, 2023
@bornach @hammancheez @Gartenberg Not far enough.
Show threadVijay PremaFeb 14, 2023
@Gartenberg
One more:
- This is incredibly mediochre and hollow... I think ill go read and appreciate some 100% human created content with actual depth and creativity.
Show threadSpockoFeb 14, 2023
@Gartenberg how do we know this wasn't ChatGPT?!
Show threadBrancaCadelaPretaFeb 15, 2023
@spocko @Gartenberg here: https://gptzero.me/
AI Detector - Most Accurate AI Checker for ChatGPT & Gemini

The World's #1 AI Content Detector with over 8 Million Users

GPTZero
Show threadGary E. WalkerFeb 14, 2023
@Gartenberg @GWillow this was forwarded to you without attribution and somehow manages to avoid my librarian-adjacent personality (as in coulda gone that route and am married to one for 22yrs, take your pick). Interesting.
Show threadGary E. WalkerFeb 14, 2023
@Gartenberg @GWillow I’m sorry, I’m a bad judge of what’s funny. The point was that the OP was funny, to be clear.
Show threadendeavorance 🕊️Feb 14, 2023

@Gartenberg A Discord server with my friends and I have a Tuesday tradition of going full "oulipo" (can't use the letter "e") and I thought it'd be interesting to ask chatGPT to adhere to that rule and tell a story

It just could not do it at all. I kept saying "Okay but you used the letter e" and it'd be like "Oh! I'll fix that" and then continue to use the letter "e" extensively

Hell, the main character in the story was "A farmer" lol

Show threadDomFeb 14, 2023
@Gartenberg yep. It got so much stuff wrong the other day I had to correct it three times. But it was interesting asking it things I knew and seeing what responses it gave. Some were actually really useful saved me tons of typing. But anything really important was a bit useless. Also as an aggregator for finding bits of code to do a given task it's not bad at all. Like if I ask it to write an equation for the towers of Hanoi...
Show threadDomFeb 14, 2023
@Gartenberg
public class TowersOfHanoi
{
public static void Solve(int n, string source, string target, string auxiliary)
{
if (n == 1)
{
Console.WriteLine($"Move disk 1 from {source} to {target}");
}
else
{
Solve(n - 1, source, auxiliary, target);
Console.WriteLine($"Move disk {n} from {source} to {target}");
Solve(n - 1, auxiliary, target, source);
}
}
}
Show threadDave ByrneFeb 14, 2023
@Gartenberg It has its place. But having wasted an afternoon trying to use an excel formula with a function it just made up I am not in a happy place with GPT. A couple of generations needed to get this to a mature state.
Show threadOliver GießlerFeb 14, 2023
@Gartenberg As an attorney I can say: most if not all answers in German (!) to legal questions are utterly BS. It cites non existent rulings. Though it cites with confidence. Maybe I should worry 😆
Show threadChristian Kent   𝘊𝘒 :＼﹥Feb 15, 2023

@Giessler @Gartenberg It's the inability to say "I DON'T KNOW" that galls me (as it did, when I was 3 or 4 years old). The sheer lack of humility is … programmed in? It can't be just a side-effect, it's so pervasive.

As someone said to me yesterday, "Did they import all of Reddit?" YES, yes they did that.

Show threadManicDeeFeb 16, 2023
@Giessler @Gartenberg the time to start worrying is when the chat bot goes and sets up fake web sites with fake literature to reference so that any cursory investigation of citations will find sensible-enough-sounding papers, and then you have to go and actually background all the references to find that they were all created within five minutes of you asking your question.
Show threadMichael GartenbergFeb 16, 2023
@manicdee shhhh, don't give it any ideas. This is going to be a great tool for the upcoming election where previously illiterate nuts can now sound rational when they post.
Show threadAlan MartelloFeb 15, 2023

@Gartenberg In the 80’s AI was going to change the world. They said it again in the 2000’s. Now again in the 20’s. See a pattern?

It looks like every generation thinks they are so much smarter than everyone who came before them … until they realize they’ve invented something that is simultaneously amazing and entirely useless for any practical general purpose application.

I’m getting tired of watching this story repeat itself.

Show threadKinky KoboldsFeb 15, 2023
@Gartenberg
That you are reposting this without attribution seems incredibly on brand for something dunking on a text generator that learned without attributing anything either.
Show threadMike MolFeb 15, 2023

@Gartenberg "spicy autocomplete" is absolutely spot-on.

So I asked GPT-3:

> If large language models are "spicy autocomplete", explain in what way black pepper, habanero and horse radish each would metaphorically represent a different kind of large language models.

Black pepper: Black pepper represents a large language model that is widely used and accepted, but may not be the most advanced. It is reliable and easily accessible, but not the spiciest.

Habanero: Habanero represents a large language model that is more advanced, but also more difficult to use. It is spicy and packs a punch when used correctly, but can be overwhelming to the uninitiated.

Horse Radish: Horse radish represents a large language model that is highly innovative and cutting edge, but also highly unpredictable. It is the spiciest of the three and can bring an unexpected kick to a conversation.

Show threadAidoFeb 15, 2023
@Gartenberg Now we get an awesome additional stage where it insults and gaslights you.
Show threadYogthosFeb 15, 2023
@Gartenberg GPT is just a really fancy Markov chain
Show threadChristian Kent   𝘊𝘒 :＼﹥Feb 15, 2023

@yogthos @Gartenberg Random Walker Chat.

This describes how it keeps apologising to me over and over, for bullshitting me. Is it supposed to imitate humanity, because, I'm not sensing any shame here?

Show threadPixel RefreshFeb 15, 2023
@Gartenberg I am glad some humans recognize ChatGPT isn't truly intelligent like I am. It is lower AI
Show threadPete Feb 15, 2023
@Gartenberg allowing for some hyperbole, I think these are all true
Show threadOven🧪Feb 15, 2023
@Gartenberg I used it to make a schedule for cleaning my home today and it gave me a suprisingly detailed plan. I tried a few other things with decent success, but yeah these tools really are only okay at best in their current form.
Show threadRandi RafaeliFeb 15, 2023
@Gartenberg Yup, asked it about what do they think about the Subaru sportero and gave me a custom answer. And btw the subaru sportero doesn't exist but this inteligence did not correct me.
Show threadChristian Kent   𝘊𝘒 :＼﹥Feb 15, 2023

@randirafaeli @Gartenberg yeah, uh, great so what we have is a functioning Fiction Generator?

I mean, I had the same level of fascination with 100-kilobyte games in 1983 like Mad Stories and Eliza

Show threadsebastian büttrichFeb 15, 2023

@Gartenberg

i like "spicy autocomplete".
I will steal that as if i was an LLM ... :)

Show threadPhil RumensFeb 15, 2023
@Gartenberg I wrote something about using it to start to create strategies here: https://philrumens.blogspot.com/2023/02/can-chatgpt-write-your-strategies.html
Can ChatGPT write your strategies?

Tired of spending months writing a strategy or paying thousands for someone else to? Can ChatGPT do it for free? ChatGPT is, on the face of ...

Show threadDaniel FreysingerFeb 15, 2023
@Gartenberg AI is that friend that talks too much after a couple of drinks.
Show threadYingtaiFeb 15, 2023
@Gartenberg someone did a quote toot of your post! https://mastodon.social/@randfish/109865233217438873
Show threadAmsterFeb 15, 2023
@Gartenberg #SpicyAutocomplete
Show threadChristian Kent   𝘊𝘒 :＼﹥Feb 15, 2023

@Gartenberg I just skipped straight to "Mansplaining as a Service", and couldn't move off that

Or maybe it's that my QA chops are so good that I've hated every single interaction I've ever tried. Black-and-white wrong answers for every question I want to research or wanted to research.

Show threaddamageboyFeb 15, 2023
@Gartenberg To be honest, many of my human interactions go the same way
Show threadLeon CowleFeb 15, 2023

@Gartenberg @anildash I would be careful with simply accepting the last 2 bullet points as fact.

Be cognizant of our own biases and denialism. Because without such diligence, we risk — at best — missing opportunities to utilize new technologies, or — at worst — we risk not being involved in the ethical evolution & application of such technologies.

Show threadSörenFeb 15, 2023

@leoncowle @Gartenberg @anildash @film_girl I don’t think it’s bias or denial to say that this isn’t intelligence, and that this is essentially an extremely advanced form of autocomplete.

Where bias and denial come in is when you evaluate how useful this may be nonetheless.

Show threadChristina WarrenFeb 15, 2023
@chucker @leoncowle @Gartenberg @anildash absolutely agree with you. It’s advanced auto-complete, but that’s a) what it is now (and not what the whole field represents) and b) as you say, advanced auto-complete is pretty damn useful a lot of the time.
Show threadAnthony PirainoFeb 15, 2023
@film_girl @chucker @leoncowle @Gartenberg @anildash To be fair, in a way, aren't we all just very advanced auto-complete?
Show threadJesseFeb 15, 2023
@piraino @film_girl @chucker @leoncowle @Gartenberg @anildash No.
Show threadSörenFeb 15, 2023
@misc @piraino @film_girl @leoncowle @Gartenberg @anildash speak for yourself!
Show threadchkaloonFeb 16, 2023
@piraino @film_girl @chucker @leoncowle @Gartenberg @anildash yes, and... oh nevermind
Show threadMichael GartenbergFeb 15, 2023
@film_girl except when it gaslights me after giving me wrong information and finally calls me rude. It's like you modeled this thing on Musk's Twitter feed and BillG interviews during the 90s
Show threadLeon CowleFeb 15, 2023
@chucker @Gartenberg @anildash @film_girl I don’t disagree, Sören. The risk I see is that it’s *written off* as not-intelligence and only-autocomplete. And us humans have a very bad tendency to not revisit our worldview assumptions in future, at which point in time it may be much more than it is now, but having written it off today (2023) could result in denialism when/if it becomes more than a fancy autocomplete.
Show threadmaxaposterioriFeb 16, 2023
@leoncowle @Gartenberg @anildash I had the exact same thoughts as the bullet points, 100% in reverse.
Show threadray ozzieFeb 19, 2023

@leoncowle @Gartenberg @anildash if we reached the trough of disillusionment this quickly, could the plateau also arrive sooner than we expect?

As we get past the tech's growing pains, and as we see it permeate business and society, I suspect we'll all look back and remember the dawn of 2023 as a sea change.

When I watch the toddlers in my family using iPads, I can only imagine what life will be like when this type of HCI is as commonplace (and as much of a crutch) as goog search & maps.

Show threadJohn LuskFeb 15, 2023

@Gartenberg
"spicy autocomplete"

😆

Show threadEthan MarcotteFeb 15, 2023
@Gartenberg (if it’s of interest: https://twitter.com/solomania/status/1625458520811339777)
Mike Solomon on Twitter

“The stages of playing with GPT-3: - OMG this can do anything - There goes my job - I should start a business around this - Some of the responses aren’t too good - Actually, some of these responses are just awful - This isn’t really intelligence - This is just spicy autocomplete”

Twitter
Show threadMichael GartenbergFeb 15, 2023
@beep thanks, proper attribution
Show threadNoodlesFeb 15, 2023
@Gartenberg This was basically my experience with AI art and Stable Diffusion.
Show threadMichael GartenbergFeb 15, 2023
@Noodles I didn't get done rather nice stuff from those two that are now framed on my wall.
Show threadanti-hero🤧Feb 15, 2023
@Gartenberg I used it for, hey here is some data i scraped put this in given structure.
Show threadBarrenkauerFeb 15, 2023
@Gartenberg https://www.theverge.com/2023/2/15/23600699/were-all-living-through-our-own-personal-ai-hype-cycle
We’re all living through our own personal AI hype cycle. - The Verge

If you’ve been playing around with ChatGPT or Bing’s AI chatbot you’ll know exactly what this guy is talking about. Also, “spicy autocomplete” is very good — apologies if we steal that Mike.

The Verge
Show threadGrodaeuFeb 15, 2023
@Gartenberg
I went to the last stage immediately. The output is just nothing even remotely interesting ever. It's just a search engine.
Show threadMichael GartenbergFeb 15, 2023
@grodaeu and the bad one at that
Show threadLRFeb 15, 2023
@Gartenberg @dbrgn also a great demonstration of overfitting to underfitting