Even as an immigrant starving student I would have paid the $20 to run Knuth's questions. But surely there was someone else with access to GPT-4 in Stanford, of all places?
The conclusion seemed dismissive of the entire field, while simultaneously being ignorant. (“Temperature, whatever that means.”)
I think they believe they can continue to ignore it, but given the huge leaps in performance just from 3.5 to 4 — or 3 to 3.5, for those who’ve tried 3 — I very much doubt that is the case.
He didn't say "temperature, whatever that means", he said "temperature 0.7, whatever that means". Do you know what it means? In the API reference it only says the value you can specify for temperature is between 0 and 2, higher values for more random output
Yes, that’s what “temperature” means, what does a temperature of 0.7 mean?
> It’s not a big ask to look this up. But even if you don’t, making a point to show that you don’t know it seems bad.
Well, no, making a point of highlighting the points of your ignorance when discussing something is good. Especially when you are a notable expert in the broad field being discussed.
> Well, no, making a point of highlighting the points of your ignorance when discussing something is good. Especially when you are a notable expert in the broad field being discussed.
I disagree. Stating “whatever that means” indicates dismissiveness, not a transparent lack of expertise. Also, you should know what it means if you’re an expert.
This quote implies to me that he is actually a beginner when it comes to this technology but is expecting to be treated like an expert whose experience generalizes
Absolutely disagree. I don't think anyone, except someone with access to the source code, knows exactly what temperature 0.7 means.
Knuth is a world expert in randomized algorithms. Do you think he doesn't have a good intuition for what could be happening? But he's a stickler for detail, and temperature is an obfuscation.
I’m getting pretty titled at the number of people who are ignoring everything I’m posting and claiming temperature is some unknowable thing because Knuth does not know what it is. Look at my link. This is not a concept specific to them. It’s a single term in the softmax selection.
There is no reason to assume that OpenAI has changed the definition of this term.
I don't know what prompt you used, but this is what it tells me (just to be clear, I don't think it explains anything beyond higher temperature = higher randomness, range of API values 0 to 2):
> In the OpenAI GPT API, the temperature parameter controls the randomness of the model's output. A temperature value of 0.7 in the GPT API means that the model's responses will have a moderate level of randomness.
> When generating responses, a higher temperature value, such as 1.0, makes the output more random and creative. This can result in more diverse and unpredictable responses. On the other hand, a lower temperature value, such as 0.2, makes the output more focused and deterministic, with the model tending to choose more probable and conservative responses.
> By setting the temperature to 0.7, you can expect the model's responses to strike a balance between randomness and coherence, offering a mix of expected and unexpected answers.
Seems like a good answer. If you want a deeper answer, you should ask a follow up question. Here is my follow up answer to “can you explain it at a deeper level”
> At a deeper level, the temperature parameter in generative AI models like GPT-3 affects the softmax function, which is used to convert the model's output into a probability distribution over the possible words or tokens.
> The softmax function takes the logits, which are the unnormalized output values of the model, and applies the exponential function to each value, normalizing them to produce probabilities. The formula for the softmax function is as follows:
> P(y_i) = exp(logit(y_i)) / sum(exp(logit(y_j)))
> Where:
> P(y_i) is the probability of the i-th token in the output sequence.
logit(y_i) is the logit value (unnormalized score) of the i-th token.
> The sum is taken over all possible tokens.
> The temperature parameter comes into play when computing the exponential function. The temperature value, denoted as T, is applied as an exponent to the logits before performing the softmax. Mathematically, the modified softmax function with temperature is:
> The temperature parameter T controls the steepness of the softmax function. A higher temperature value (e.g., 1.0) increases the exponentiation effect, making the probabilities more spread out and the distribution more uniform. This leads to a higher likelihood of selecting less probable tokens, resulting in more diverse and random output.
> Conversely, a lower temperature value (e.g., 0.5) reduces the exponentiation effect, making the probabilities sharper and more focused. The most probable tokens are emphasized, leading to more deterministic and conservative output.
> By adjusting the temperature, you can manipulate the balance between exploration and exploitation during text generation. Higher temperatures encourage the model to explore more possibilities, while lower temperatures bias it towards exploiting the most likely choices based on the training data.
> It's important to note that the choice of temperature is a design decision, and there is no fixed rule for determining the optimal value. It often requires experimentation and understanding the specific behavior of the model to achieve the desired output characteristics
Not saying it's bad as a qualitative answer, but it doesn't say anything quantitative about the effect of the temperature in the ChatGPT API. Temperature is a well known and wel documented concept, but if you don't know what y_i is, and for all I know that's just a number coming out of a black box with billions of parameters, you don't know what temperature 0.7 is, beyond the fact that a token i whose logit(y_i) is 0.7 higher that that of another token, is e times as likely to be produced.
What does that tell me? Nothing.
My dude it’s not my fault if you don’t understand the concept of asking follow up questions for clarification. This isn’t like a Google search. The way you retrieve knowledge is different
Maybe you misread my comment ;)
I'm sure Knuth knows qualitatively what is meant by temperature, it's been used as a measure for randomness for half a century in simulated annealing and other algorithms
I think you're still misreading my comment (and dragonwriter's and Knuth's): we all know or can look up what temperature is in randomized algorithms. However, what temperature 0.7 means is a mystery to me. I know that at temperature 0 the result is deterministic, and at higher temperature the randomness increases (possibly they are the Boltzmann factors associated to some energy function, but I don't know, and even if it is, I have no idea how it is scaled, i.e. what is the value of the Boltzmann constant). I know that the API accepts values from 0 to 2. I don't know more. Do you?
Yes. I have posted both a very nice link and a complete explanation from chat gpt 3.5 itself. It’s honestly not that complicated, especially for someone who is supposed to have any sort of authoritative view in the field.
I do not feel it is appropriate for you to say you have looked it up if you don’t know what it is besides an API input that affects randomness.
I myself shall certainly continue to leave such research to others,
and to devote my time to developing concepts that are authentic
and trustworthy. And I hope you do the same.
Best regards, Don
PS: Please reply only with respect to binomial coefficients,
because I've already spent way too much time on the topic above!
The topic is timely, and important enough not to ignore completely,
but it's emphatically not for me.
Knuth knows enough to poke the fate bear AI with a stick for his own curiosity but also knows he doesn’t know enough to really pontificate about it, and doesn’t want to spend his remaining time getting to know enough about it.
Knuth has only so many productive years left on this planet and he wants to spend them all on TAOCP (and a few other projects). He had given up email by the 90s and he is incredibly disciplined on what he devotes his time to. If you want to get in touch, send him an actual letter. Last I checked, he reads paper mail once a quarter or so. He's also an utter perfectionist in his own work, in a way that few people ever achieve.
So he's curious enough to try ChatGPT. But then he's smart enough to walk away.
He did pass judgment though by implying GPT is not "authentic and trustworthy". That what got me thinking, not his desire to focus on other things (it's his time, he can do whatever he wants).
Knuth writes people checks for reporting bugs in his software or errors in his books, including at least one $327.68 check for a TeX bug. According to Wikipedia, he has written over $20,000 in checks. (Relatively few of these are ever cashed.)
Knuth's standard of "authentic and trustworthy" is set remarkably high.
Not if you're going to criticize it in public, though. It's okay for us to think less of people who do things like that, even when -- or especially when -- the guilty parties are respected computer scientists.
"Science advances one funeral at a time" is a cynical commentary, not an ideal to strive for.
