AI ML
Post #7047 · source on Telegram
CalcGPT: A Calculator Unnecessarily Powered by GPT That Gets 1+1+1+1 Wrong
Description
A side-by-side screenshot showing two screens. On the left is a calculator app called 'CalcGPT' displaying the calculation '1+1+1+1' with the result '13' - hilariously wrong. The calculator has a 'Regenerate' button instead of a typical operator button, mimicking LLM interfaces. On the right is the project's README.md on GitHub describing it as 'A silly calculator unnecessarily powered by GPT' with a TODO section that reads 'Add blockchain into this somehow to make it more stupid.' The meme perfectly captures the absurdity of using AI for tasks that simple deterministic code handles perfectly
More like this
More like this in AI ML →
Use J and K for navigation
Comments
38Comment deleted
CalcGPT: where 1+1+1+1=13, the TODO is 'add blockchain to make it more stupid,' and somehow it still has more GitHub stars than your actual production app
This is what happens when the prompt engineer is your architect. The scary part isn't that it's wrong, but that there's a 'Regenerate' button, implying you might get the right answer through sheer luck and API credits
Sure, the sums are wrong, but once we shard the model behind a blockchain-backed microservice mesh, eventual consistency will give us the right answer - eventually
After 20 years in tech, I've seen us go from optimizing assembly for calculator apps to needing a GPU cluster and venture funding to add four ones together - and still getting it wrong. At least when we add blockchain, we'll have an immutable record of our inability to do basic math
When your startup pitch is 'We're disrupting basic arithmetic with AI-powered, blockchain-enabled calculations,' you know we've reached peak tech absurdity. The fact that 1+1+1+1 equals 13 here is actually the most honest thing about using LLMs for deterministic operations - at least it's consistently unreliable. The TODO about adding blockchain is chef's kiss: because nothing says 'enterprise-ready' like a distributed ledger for addition. This is what happens when you let VCs write your technical requirements
CalcGPT is peak innovation: replace O(1) addition with a nondeterministic, rate‑limited API call, then put the wrong answer on‑chain so it’s immutably 13
1+1+1=13: Proof LLMs excel at creative accounting, perfect for that next 'disruptive' VC pitch
Only in tech do we replace an O(1) ALU with a 700‑ms LLM call that returns 13 for 1+1+1+1, then add “make it eventually consistent on‑chain” to the TODO
Checks out. Comment deleted
Make it 49000x slower Comment deleted
after this it's extra funny, when people say "but chatGPT has calculated a different result" Comment deleted
don't forget about microservices Comment deleted
How did you guys managed to gets such bad results ? Did you use an old version ? Of course there is randomness, so it may be wrong if I rerun the prompt a lot of time. Of course it makes no sens to have such app, but pretending that today's LLMs are not capable of computing simple arithmetic is disinformation. Edit: https://github.com/Calvin-LL/CalcGPT.io Comment deleted
I tried to ask "how many h in heil hitler" and I got banned 💀 Comment deleted
doing minimal testing, chatGPT (and mistrel, which is a lot faster) seem to have dedicated math modules. I wonder if they access that on purpose or if it tries to externally detect when the AI is trying to do math. either way, I'm sure there's ways to make it not use it, be it accidentally or on purpose Comment deleted
can't test this on chatGPT (still banned), but it's completely wrong here lol. first mildly complex question and suddenly my mass is relative to gravity (yes I know it corrects this later on, but this kind of critical error would earn me an F on a school test) Comment deleted
correct calculation (to the best of my non-physicist knowledge): 45kg on the earth and moon stay same earth has around 9.8m/s² accel moon has around 1.6m/s² accel earth: ~441N moon: ~72N fwiw, you *would* feel around 6 times lighter on the moon. I'm sure it got that from some external source instead of calculating it itself. Comment deleted
I don’t get this AI hate Like, use cheap shit - get shit result Any answer of model are not final - you are human and expert here. Once humans will not be needed, governments and corpos will get rid of most of us Comment deleted
1. No model being mentioned 2. No clarification on init/system prompt Wow, this screenshot is useless! Comment deleted
my brother in christ, I just opened https://chat.mistral.ai/ and used whatever was there Comment deleted
This screenshot couldn’t be chatgpt though… Comment deleted
sry, wrong AI. same sentiment tho Comment deleted
edited the msg to correct Comment deleted
Also, if you don’t like Musk, there x5 reasons to actually hate Sam, no idea why you would still chatgpt 🌚 Comment deleted
oh absolutely, I only occasionally use GPT out of curiosity Comment deleted
Mistral is European failure, not an AI :( Comment deleted
mistrel is better than chatGPT the few times I've tested it. not that that's saying much Comment deleted
No-no-no It’s an absolute failure Comment deleted
"And used whatever was there" Do you also start work on any project with just random language? Like, you wanna landing page? Lemme quickly move some registers with asm since I have IDA opened? 🌚 Comment deleted
I was just dicking around, jeez Comment deleted
I know, but everyone dicking around for 6 months and doesn’t try to improve themselves It’s makes me angry now 😭 Comment deleted
I just don't see any reason to invest what little time I have into learning a tool that I've at best seen giving ok results Comment deleted
not to mention the financial bubble and insane electricity usage ofc Comment deleted
even openai ceo i think said it was a bubble Comment deleted
Try running it locally if you can ;) gpt-oss:20b with Ollama Comment deleted
not gonna infect my system with bloatware, but thanks Comment deleted
LGTM til it is Rust Comment deleted
"1" + (1 + 1 + 1) = "13" CalcGPT with JavaScript Logic! 💀 Comment deleted