It's great to see a _real_ AI application among all this media noise ;-).<p>Seriously though, this is wonderful satire. I asked 88x10 and it returned an HTML meta tag.
> GPT-3 (babbage-002)<p>I'm surprised babbage is still available via APIs - <a href="https://platform.openai.com/docs/models/gpt-base" rel="nofollow">https://platform.openai.com/docs/models/gpt-base</a><p>Anyone else using this?
This neat demo is a year old now, it was first released in July 2023.<p>Source code and prompt here: <a href="https://github.com/Calvin-LL/CalcGPT.io/blob/main/netlify/functions/math.ts">https://github.com/Calvin-LL/CalcGPT.io/blob/main/netlify/fu...</a><p><pre><code> const prompt = `1+1=2\n5-2=3\n2*4=8\n9/3=3\n10/3=3.33333333333\n${math}=`;
let response: Response;
try {
const openAI = new OpenAI();
response = await openAI.completions
.create({
model: "babbage-002",
temperature,
top_p: topP,
stop: "\n",
prompt,
stream: true,
})
.asResponse();
} catch (error) {
return new Response("api error", {
status: 500,
});
}
return new Response(response.body, {
headers: {
"content-type": "text/event-stream",
},
});
</code></pre>
It's using the old babbage-002 model with a completion (not chat) prompt, which is more readable like this:<p><pre><code> 1+1=2
5-2=3
2*4=8
9/3=3
10/3=3.33333333333
${math}=</code></pre>
Entered 42<p>The 8 solutions I got while clicking on regenerate:<p><pre><code> 3.33333333333
42, so the point your talking about is 3.3 (Accuracy is
3 Additionally, 3 coincided with John 3:16 , "$3
1
3.33333333333
42
42+1=3+1=4=42+1=43
2×5
</code></pre>
Not so sure what I just did.
Results are copy-pasted as-is
This is the first time I have come across Calvin Liang, but I’m already a big fan. Their artist’s statement manages to be very funny while making a point. I like today.
I think there is a bug here...<p>8888888×965 = 965 according to this site with temperature = 0 or 3.63... with temperature = 1<p>On the other hand, GPT4 gets it correct:<p><a href="https://chatgpt.com/share/34007f39-cfa8-46c8-bda3-9f641affc17a" rel="nofollow">https://chatgpt.com/share/34007f39-cfa8-46c8-bda3-9f641affc1...</a><p>Even when I instruct it not to think about it:<p><a href="https://chatgpt.com/share/cb22c9dc-1549-4d00-a498-c889f6822b6a" rel="nofollow">https://chatgpt.com/share/cb22c9dc-1549-4d00-a498-c889f6822b...</a>
I'm sorry but this falls flat for me. GPT4 routinely can answer impressive math questions for me (college-level):<p>- What diameter steel wire would I need to be rated for a weight of 500lbs?<p>- How many digits would a ID need to be (using 36 characters) to have a 1/10^20 chance of collision over 1 billion random IDs?<p>- If I have a list of a million times (say durations of a web request) and they follow a normal distribution, and I take a sample of 1 million of those, how close would the average of my .1% sample be to the true average of the billion?<p>- Suppose in D&D I am told to roll 20 d6, but instead of rolling that many dice I want to roll just two (larger) dice and add a constant. Which standard D&D dice might give the closest variance and what is the constant?
I’m enjoying experimenting with nonsensical math:<p>> Apple * dog<p>> CalcGPT: Apple Mini − dog or dog. Total= Apple Dog Mini MiniDog=49<p>I was wondering if math of words would produce the embedding of the operation of those words, but nope :)
This is neat, but most people are going to miss "GPT-3 (babbage-002)". Using a rudimentary, outdated model seems disingenuous when making any kind of point about AI.
AI chatbots differ in their ability to handle long calculations involving single-digit numbers — <a href="https://userfriendly.substack.com/p/discover-how-mistral-large-2-claude" rel="nofollow">https://userfriendly.substack.com/p/discover-how-mistral-lar...</a>