Ask HN: What are some (weird) ways you're using ChatGPT?

[pknerd]

I’m a lazy person, and I’m proud of it. As a programmer, I’m always on the lookout for creative and efficient ways to use LLMs, especially ChatGPT. Here are some weird ways (or maybe not so weird?) I’ve been using different LLMs:

Extracting IBANs: A while back, I needed to add someone as a beneficiary in my banking app. Instead of sending me the IBAN as text, they sent a screenshot via WhatsApp. I was already lying in bed, using my phone, and didn’t feel like typing it out. So, I uploaded the screenshot to GPT and instructed it to extract just the IBAN. Initially, it resisted, warning me about sensitive information. But after some persistence, it provided the IBAN — though it annoyingly prefixed it with, “Here is...” Eventually, after tweaking my request, it gave me the IBAN in isolation. Lazy? Yes. Efficient? Absolutely.

Code generation and conversion: I work in multiple programming languages and frequently use LLMs as converters and generators. For instance, while working on a contact export script in PHP, I printed the $contacts associative array to the console, copied one of the entries, and asked GPT to create a Symfony entity file. It generated it perfectly, complete with accurate data types. I often do the same to generate CREATE TABLE commands from arrays.

Database functions: I copy table schemas from TablePlus and ask GPT to generate functions for fetching connections, storing, and retrieving data. Saves me a ton of time.

Time tracking helper: I work remotely and use Toggl for logging hours. To streamline calculations, I created a custom GPT app called TimeAdder[1]. I simply copy each time entry, and it converts them into the decimal format for easy calculations.

Bootstrap forms: I upload a table schema to GPT and ask it to generate a Bootstrap-based form. It handles the field generation beautifully.

UI design inspiration: Sometimes, I upload a screenshot of a product and ask GPT to create a similar UI. It provides a good starting point for prototyping.

There're probably many more but I can't remember at the moment. What are some weird or unconventional ways you use GPT/LLMs? Share your experiences — I’d love to hear them!

[1] https://chatgpt.com/g/g-RxQPSCC4H-timeadder

[mejutoco]

> sending me the IBAN as text, they sent a screenshot via WhatsApp

The iphone has built in ocr for images, in case other people do not know. Long press on the text of an image and it will happily select it as text to copy.

[stonegray → mejutoco]

You can also OCR right from the camera into any text input by long-pressing as if to paste and selecting autofill -> scan text

[satvikpendem → mejutoco]

Same with Android, at least newer versions.

[muzani]

I was working on a horror game. I wanted to narrow the theme down to a specific time and location. Something like the Salem witch trials or Vlad the Impaler era is too cliche.

I remembered that a relative was stuck on the Flores island east of Bali recently due to volcanic activity. It seemed suitable - remote, exotic area, paranormal region. So I asked ChatGPT to search up lore on the area on cults and volcano gods and such.

ChatGPT brought up Batara Kala, the god of the underworld who resides in the area, banished from earth because of his habit of eating men. It also brought up several stories of shamans in the area who guard against the volcano and keep the land happy. The power of the shamans have declined and been forgotten. But it's said that volcanic activity happens when men feel like they're invulnerable and sow chaos in the land. It also brought up other myths like the hobbits (homo floresiensis) on the island and linked that to Batara Kala's status as a god of time.

Basically it gave me several hours of very high quality research in 15 mins. I wouldn't be able to find most of these details on a google search, much less the stories about the shamans.

[unrealp]

I used it for a while to understand mindfulness (vipassana) and problems I was facing in understanding the concepts and problems I was facing during the practice. It was pretty good, brainstormed well, knew references to original writings etc. Was useful.

[Tomte]

I have it explain Emojis to me.

People constantly message me Emojis that I cannot even describe. Paste into ChatGPT, „explain this Emoji“, done.

[muzani → Tomte]

One of my earliest uses of AI is the opposite - I had it summarize a feeling into a set of emojis

[pknerd → Tomte]

Or GenZ codewords: ngl,ikr,irl etc.

[petargyurov]

Not that weird but I use it as a translator and to learn Spanish. It's better than Google Translate because I can specify the region (i.e: Spain vs Latin America) and the formality. I can also experiment with more colloquial phrases and idioms that don't have a direct translation, e.g: "It's raining cats and dogs"

It can be useful in explaining grammar and fixing mistakes, though it can sometimes falter.

I tend to double check some things against other translators if it's something important of if the output looks fishy.

[dserban → petargyurov]

I use it to extract either vocabulary or grammar patterns from snippets of text written in a foreign language I'm learning. I double-check all output, just as you do.

[Quinzel]

I don’t use AI for any of my work or study requirements.

Rather, I’ve been using chatGPT as a recipe book. I generally hate cooking so decided to see if having a robot cooking companion might help make it easier. For me, it actually does. Firstly, the AI recipes are so much less wordy than human written recipes but also… chatGPT will adapt the recipe if I realise half way through cooking that I have forgotten to get an ingredient. I’m sure it’s not something good cooks would need, but I find AI most useful in my kitchen of all places.

[embeng4096]

I have a spreadsheet in which I track various habits: time slept, rise/bed time, yes/no for things I want to do every day, screen time, etc. I took a screenshot and pasted it into chatgpt and asked for feedback and suggestions. It gave me (depending on prompt) suggestions for streamlining it to make it more likely I fill it out, and for writing API/Python code to extract data and visualize it.

I also have a text file where I list various activities that reward points or cost points, which I'm trying to use as motivational scaffolding for being more productive/healthy in my daily activities. I pasted that file into chatgpt and it gave me suggestions for rebalancing various point values and suggestions for other rewards/costs, as well as a structure for bonuses with intent for boosting motivation at certain point thresholds. I found the reasoning for the suggestions to be compelling enough I did actually incorporate a lot of the feedback from chatgpt.

[delaklo]

I think the most weird way of using chatgpt is to generate response for your answer->

People use ChatGPT in some fascinating and unconventional ways, such as co-writing stories or songs, role-playing historical figures, simulating a Dungeon Master for RPGs, or brainstorming startup ideas. It's also used as a decision-making tool ("What would a pirate do?"), a dream interpreter, or even for coding quirky projects like ASCII animations. Some get creative by generating stand-up comedy routines, debating bizarre topics like "Is cereal a soup?" or exploring alternate realities and futuristic societies. Others use it for personal growth, whether as a reflective journal, a language tutor, or even a playful companion for venting. How about you? Found any quirky use cases?

[f30e3dfed1c9]

Your first example could not possibly have been more "efficient" than simply retyping the IBAN.

[upcoming-sesame → f30e3dfed1c9]

On a side note, people who send screenshots of text to be copied should be banned from the internet

[muzani → f30e3dfed1c9]

I have a OCR tool just for things like this

[pknerd → f30e3dfed1c9]

Error probability is high to see/write 11-16 digit long numbers.

[upcoming-sesame → pknerd]

Android has a built in way for this btw, swipe up to get to the "task manager" (where you switch between open apps) and then hold press the IBAN to select it from the image

[f30e3dfed1c9 → pknerd]

Ehh, you type it, check it, correct it if necessary, check it again. This is a task that a human being can successfully do. We're not total idiots, at least not yet.

[cblum]

I'm learning Chinese characters following Heisig's method. I use to create mnemonics when those are not provided in the book, since I have a hard time coming up with my own. Turns out ChatGPT can get quite creative. I also use it to generate bizarre images that remind me of how certain characters are written (for example, 象 elephant shares its top part with 兔 rabbit and the bottom is the same as the bottom of 家 home, which actually used to mean pig, so I used ChatGPT to create an image of a rabbit-headed, pig-bodied elephant to prompt me to write 象 when it comes up in Anki).

[chimo777]

I often have it understand complex Chinese subordinate clauses, in order to check the differences in comprehension ability between myself and AI.

[lihaciudaniel]

I have an xml journal, basically notes in tree hierarchy

I edit with python platform on mobile and ask gpt to rethink the tree and notes — we are building knowledge that is for gardening

If anyone curious of the xml journal, reply

[uncomplexity_]

reverse engineering minified javascript

understanding cryptic logs

converting unstructured to unstructured

eli5 everything

[bhu1st]

I use it to convert (translate) complex sounding English text chuncks into Nepali language and it hasn't disappointed me.

[rsynnott]

> Eventually, after tweaking my request, it gave me the IBAN in isolation. Lazy? Yes. Efficient? Absolutely.

... I mean, not very efficient. You could just've used actual OCR; if the phone was an iPhone this is built right in, would take two seconds to select the IBAN, and should be pretty reliable. I think modern Android has something similar, though couldn't swear to it. LLM-y things _sometimes get OCR of numbers wrong_, like, a lot more than actual OCR software, which is now close to perfect on printed text.

Now, with it being an IBAN the practical risk here would be low; IBANs have two checksum digits, so at worst you have a 1 in 100 chance of an incorrectly-OCR'd IBAN being accepted (in practice, it's lower than that; some checksum-passing bad IBANs will have non-existent country codes, and likely be immediately rejected, or non-existent bank codes or account codes, and will be rejected at some point, though potentially not immediately. This would be _really_ risky with, say, US ACH, where the routing numbers are not checksummed and are only eight digits.