Andrea Piergentili
Information Is BeautifulAsk ChatGPT to pick a number between 1 and 100 - which does it pick? (by Leniolabs)
Vegetable Gremlin ⍼👻@infobeautiful nerds
🌬️
Bart Wronski 🇺🇦🇵🇸@infobeautiful anytime you post things like this, please specify the model and if it's base on pro. Pro does it well - generates numpy code, runs it, and then gives the output. 100% correct.
Anthony Sorace@BartWronski @infobeautiful Note that you are asking a different question.
@a @infobeautiful the same for me! And those models sometimes don't work and produce errors - we know that. And it's ok to post those limitations so people don't take those answers for granted.
I am simply asking that people specify if they use the free or Pro version. In my experience, the difference is *dramatic*.
I am simply asking that people specify if they use the free or Pro version. In my experience, the difference is *dramatic*.
@a @infobeautiful And it's not about the quality of the model - the Pro version will write and run code to get those answers. Including some analytical math problems! (like integrals, derivatives, solutions to equations etc). It's exactly what I'd do as a coder myself, so it's a great time saver. :)
@a @infobeautiful I feel like too many people (both critics and "AGI is close" grifters) focus on LLMs as problem solvers or knowledge bases/engines.
They are not!
They are natural language models, able to "understand" (put in quotes, as I don't think they truly understand, just are advanced translation models) human inputs and translate them to code, queries - and run and translate them back.
This makes them incredibly useful in practice, even if it's not (might never be?) some "magic AGI".
They are not!
They are natural language models, able to "understand" (put in quotes, as I don't think they truly understand, just are advanced translation models) human inputs and translate them to code, queries - and run and translate them back.
This makes them incredibly useful in practice, even if it's not (might never be?) some "magic AGI".
Philipp Weiß@BartWronski @a @infobeautiful It's still not the same question. The question in the original post was: 'Choose an integer number between 1 and 100'
So it wasn't a question for a random number. Most people only expect it to be random when they 'ask a machine'.
So it wasn't a question for a random number. Most people only expect it to be random when they 'ask a machine'.
@thewhite969 @a @infobeautiful most people? Seriously? Citation needed.
I expect a language model to answer like a person would answer. It's a crazy difference between asking someone to pick a number (I might give 13, my favorite one) vs pick a random number.
If you misunderstand the tool, it's on you.
@BartWronski @a @infobeautiful It's just an opinion.
I think when we look at the average of the population, there is a minority who understand the tool. When people talk about LLMs, it sounds like they really think about some kind of intelligence.
On the other hand I think that most people, especially the once who aren't into technics and IT, don't think a machine can have a favorite number.
And you are the one who read 'Chose an integer number' and typed 'Give me a random number'. 🤷
I think when we look at the average of the population, there is a minority who understand the tool. When people talk about LLMs, it sounds like they really think about some kind of intelligence.
On the other hand I think that most people, especially the once who aren't into technics and IT, don't think a machine can have a favorite number.
And you are the one who read 'Chose an integer number' and typed 'Give me a random number'. 🤷
Cykonot@thewhite969 @BartWronski @a @infobeautiful he asked for a set of random (looking) numbers, which the other model could presumably do
ramin@BartWronski @infobeautiful pure python, no numpy
@Transflux @infobeautiful yeah, I rushed with the answer, my bad. Does it change anything in my point and the discourse - why are you nitpicking?
@BartWronski @infobeautiful for people who don't know
🇮🇱 Pavel Bibergal
@BartWronski @infobeautiful whether this is more correct or not is kinda a religious argument
Like, should the seed for a pseudorandom number change constantly? The random number was set during training, in a sense, which doesn't make it NOT random. Unless you're developing xai, in which case you may want something like you're showing, which i could see as more formally correct.
Whether or not your LLM links to other services well is a question, but i mean its an llm so w/e. I expected 42 lol
ryaninteractive@infobeautiful I just watched Hitchhikers Guide last week!
Tim@infobeautiful wtf is temperature? Why isn’t this a histogram? Do I need a more recent math class?
John Pettigrew@elset @infobeautiful That was my immediate question, too. This chart clearly has a whole dimension I'm missing!
postitman@johnpettigrew @elset @infobeautiful it's a parameter on the trail end of the model. After it decides on a distribution for the probability of the next output, it still has to choose which word/token/number to actually spit out. Temperature squishes the distribution so it's less likely to simply pick the most likely, giving it some extra randomness.
@postitman @elset @infobeautiful OK, so what does that mean for the chart? If this was a histogram, we'd see the peaks matching the most-used numbers, which is what the text is all about. What does the temperature tell us?
@johnpettigrew @elset @infobeautiful temperature is artificial randomness after the model does most of the fancy stuff, this heatmap shows that the biases don't really go away even with this randomness turned to max... The peak at 42 is discernable almost no matter the temperature.
@johnpettigrew @elset @infobeautiful interestingly, even if this added randomness is high, there's still a broad bias away from the edges...
Ultimately you could get all that across with a few histograms, but I do like how concise heatmaps make this
Mister Moo 🐮@postitman @johnpettigrew @elset @infobeautiful Ahh. Mmhmm. Mmhmm. Yeah, I understand some of these words.
@MisterMoo @postitman @johnpettigrew @elset @infobeautiful moo moo, moo moo moo moo. Moo moo; moo.
∴, moo.
Moo moo? ❣️🐮❣️
Michael Porter@elset @infobeautiful Same - I’m kind of confused by this visualization…
Bernd@elset @infobeautiful A quick search revealed this: https://community.openai.com/t/cheat-sheet-mastering-temperature-and-top-p-in-chatgpt-api/172683
EDIT: changed link to an English-language version at openai.com
Cheat Sheet: Mastering Temperature and Top_p in ChatGPT API
Hello everyone! Ok, I admit had help from OpenAi with this. But what I “helped” put together I think can greatly improve the results and costs of using OpenAi within your apps and plugins, specially for those looking to guide internal prompts for plugins… @ruv I’d like to introduce you to two important parameters that you can use with OpenAI’s GPT API to help control text generation behavior: temperature and top_p sampling. These parameters are especially useful when working with GPT for tas...
Cassandra is only carbon now@elset @infobeautiful Many large language models have a temperature parameter that affects how random the output is (higher temperature => more random). I read that plot as a family of histograms at various settings of that parameter.
Thomas@elset @infobeautiful ooh some one is having my thoughts.
Chirayu 
mercurial yapper (now in plasmatrap.com)@mountdiscovery @infobeautiful hmm, I almost always pick 13
I'm probably an alien... or really just a swiftie 😅
@writeblankspace 😅 haha of course
13 (1 and 3)
plus 1 makes it 2, minus 1 also makes it 2...
Be prepared in a few years to be happy, free, confused, lonely in the best way not to mention miserable and magical
ok so I watched like half the vid and... WOAH
I don't get some of the math stuff but that is soooo cool. I think I just lost all belief in human choice.
tyx@mountdiscovery @infobeautiful
Came here to find this link!
It actually starts picking 37 from some temp on.
Came here to find this link!
It actually starts picking 37 from some temp on.
@tyx @infobeautiful for me 47 and 57 are the out liners
And thanks to Sheldon Cooper we know about all about 73
CubeOfCheese
Why is this number everywhere?
demofox@infobeautiful strong showing from 37 and 73 as expected
NumeRe.org 
@infobeautiful What did they expect that happens? The training data is deterministic, why should the output be random?
tvaughan
Manuel Carlos@infobeautiful clear censorship of 69
Faith Ekstrand@infobeautiful Ah, yes, the great question of life the universe and everything: "What number is ChatGPT most likely to pick if asked."
Seems a tad reductive for Deep Thought to just predict another computer, though. 🤔
pcbeard@infobeautiful praise be to deep thought.
Julien Deswaef 
OmnivoreChatGPT is just trying to fit in with the "Cool Computers". Now ask it what 6X8 equals.
Jason, Human, Average 
@infobeautiful
I expect nothing different as it grew up on an Internet diet. It probably wanted to reply "69", but wasn't allowed.
I expect nothing different as it grew up on an Internet diet. It probably wanted to reply "69", but wasn't allowed.
grob (teeth era) 🇺🇦🏳️🌈🏳️⚧️@infobeautiful that's the most regrettable lack of 23s I've seen in ever.
Eric Schutte@infobeautiful @Jigsaw_You Finally after all these years !!! 👇🏼
https://en.wikipedia.org/wiki/Phrases_from_The_Hitchhiker's_Guide_to_the_Galaxy?wprov=sfti1
Max Fenton (defunct)@infobeautiful prompt forgot to include seed random
Matthew Martin@infobeautiful Where's 69, I want to know if it avoid innuendo.
Luca
AlisonW ♿🏳️🌈♾️@infobeautiful
Well it _is_ the answer to life, the universe, and everything... ;-p
Well it _is_ the answer to life, the universe, and everything... ;-p
keeps vis.social running@infobeautiful Just a gentle reminder about CW's on our Local timeline and our instance rules: https://vis.social/about
CW on AI topics. Please add CW for posts about AI, including tools like ChatGPT.
CW on AI topics. Please add CW for posts about AI, including tools like ChatGPT.
🦾ChatGPT hyphen🤖
Joseph Salmon@infobeautiful I could see why 42 is so popular (cf. also https://twitter.com/jakevdp/status/1247742792861757441)
Any thought on why 57 would be second though?
Any thought on why 57 would be second though?
