GPT Excel | Hacker News Reader

GPT Excel

gptexcel.uk

42 pointsjoel9957 comments

Not sure if everyone knows just how easy making something like this is. This kind of an app takes less than one week to develop, using the OpenAI function calling API.

A week!?

An hour if developed with ChatGPT...

I'll give you 10k usd if you copy this work in 1 hour (which is much easier than creating it, mind you). Otherwise stop downplaying other's people work.

Other people work was not the target of the remark. Stop not having a sense of humour...

Ok, one week.

OK, do I understand correctly you are offering 10000 USD for someone who recreates this service (=a website that generates an Excel formula based on its description in English) in a week?

Frankly, and I don't want to downplay anybody's work, but this seems doable. The core engine based on API calls to OpenAI is probably a few hours. The remaining time is for user authentication, subscriptions, payments, graphics and so on.

I know this because I'm using OpenAI API for personal automation projects and in spite of its quirks it's relatively simple especially if you don't need continuity and each task is independent.

No, that was a joke. But only a few very capable people can do something at the same level as GPT Excel in one week. Including videos, templates, etc. A prototype? Yes. it will be bad. If you wanna bet, I'm all in.

Just to bring back the comment that sparked this off:

> Not sure if everyone knows just how easy making something like this is. This kind of an app takes less than one week to develop, using the OpenAI function calling API.

So - the original statement was "something like this". Not the design, the videos, the templates but the core, essential functionality. And that is probably a couple of hours work.

I suspect the real value is mainly in the prompt templates. I haven't tried but there's probably some wording that helps guide the results. It might be that these took a lot of trial and error to tune - or it might be that the first thing I try gives me results as good as GPTExcel.

A week that includes day job, family life and general adulting stuff.

Well I don't have insight into your adulting commitments so that's not a terribly useful measure for me. ;-)

Not sure if you're trying to motivate me to do something, or just the opposite!

Yes, it is easy and it will get easier, the overall quality of software products is rising fast.

Apps and product development is like a marketplace now, the demand for software is huge but most people just don't know about what they can do with software. So there are hundreds of GPT Excel's out there, but there is demand for thousands so if you can do this in one week, do it! There is a lot of people trying those projects just for fun and spending money on it.

"130k+ Happy users" that's a very good traction for a new project. I'm wondering how it got so popular. It got viral somewhere?

it's also super easy to lie about something like that. what are the practical consequences of accidentally adding an extra "k" character on there? or what is deemed as the definition of "happy" or "user" — could be extremely weak definitions

Not to be too cynical but I would also had my doubt. 130K is huge number for a brand new product, the 3 avatars shown are stock photos. And when you combine that with the advertised "845,022 formulas generated, so far." (6 per person) + "4 requests per day." (on free plan) , seems very unlikely that there is that much paying or regular users.

am i misssing something....but what can someone do with this that they cant do on chat gpt.

Maybe 130k people have visited the page, or seen the tweet..

I believe this is the standard way of doing things now. And has been for decades. Practically all big brands used some form of "growth hacking techniques" on launch so you can't really blame newcomers.

Their Twitter has 15 followers. And not not “k followers”

If you use Google Sheets, there's 10's of different GPT formula plugins that are integrated. The challenge is knowing which ones are any good and have good privacy measures.

https://workspace.google.com/marketplace/search/Gpt?host=she...

And there is Bard built right in now.

This could be more of a symptom of Excel (and spreadsheets) being too darn hard to get right.

I've switched to Sqlite and Jupyter from what was a useful but hard to debug spreadsheet with really long repeated formulas that were just database joins. The amount of code to view has probably dropped by 90%. Of course now its easier there is ten times the functionality, but at least its a lot easier to understand some SQL vs Excel formulas.

PowerQuery and the built in data model do a lot to make Excel better at handling joins in a sane fashion, but are relatively recent features that some are still realizing are there in the first place, particularly for the Office365/Mac crowd. Nice way to keep your basic tables basic and still maintain the full depth of your relationships all inside of Excel. Throw in some calculated columns and measures using DAX and it's a pretty darn powerful tool.

That said, in the data science world, there's a LOT of crossover when it comes to functionality, between tools like Excel, Tableau, and *SQL in particular. And the crossover seems to be growing by the day, especially with Python functionality being built into Excel in the near future.

This is a feature in the upcoming Excel from Microsoft.

If it isn't already, it's an obvious thing to add since chatgpt got popular.

Especially as M$ is heavily in bed with OpenAI (a complex bed)

A friend of mine has created some amazingly sophisticated spreadsheets with the help of the regular chatGPT. How is this any different?

That's the neat part, it's not.

https://chat.openai.com/share/bce0e12f-ebfc-4fcd-babc-9e0a48...

Just another in a long line of fancy wrappers made by AI grifters.

It's not called AI grifting, but Trickle-down economics...

We are Costco customers complaining about convenience stores

nice but i expect microsoft integrating this feature into excel for their enterprise customers soon

Yep. There'll soon be some opensource alternatives as well.

Now, just to make Excel formulas actually work consistently and we’ll be all set. VLOOKUP doesn’t even work correctly about half of the time.

vlookup is usually not the right tool for the job (index + match better)

Or xlookup if you are younger.

You are probably doing an approximate match which is the default. Always add FALSE as the last argument.

Nope, I know about that. It’s just garbage software, especially their Mac version. See also the inability to open a CSV file correctly without corrupting it, unless you use a specialized import.

Man i just want to be able to use js instead of excels crappy meta dsl

This looks to me like what the old Clippy should have been. And I mean it as a praise not a diss. I'm all for AI assisting us.

Four requests per day seems a little low to verify the quality of the responses before shelling out even a few bucks.

That said, I've already gotten great quality formula and VBA generation directly in ChatGPT (GPT-4.0), with only a sentence of prompting, so I'm not the target customer.

That have the io domain but redirect to uk. Very odd!

Is this a spam?

Not working in Firefox. Console shows "Uncaught (in promise) TypeError: r is undefined"

hey there. just checked with firefox and its working fine. are you using a phone? also what were you trying when this error happened

Strange. It's working now. Might have been user error.

Smells like network request failure

Someone needs to wrap this in a clippy UI

This may be really dangerous. Many will see enablement here and speeding up dull processes, I see untrained users getting results via a GPT that can spout out hallucination and false answers any time, and the user will simply believe it as he won't/can't do validation. It's one thing developers using it to create easy to review code snippets, another to generate complex queries in office apps to operate on raw data. I know, the same user would have copy pasted a formula before without checking, but using GPT will open up the floodgates for such issues.

Good excel (such a thing does exist) should almost never refer to a cell by its coordinate rather than using either a named range or better still a table, this probably encourages not doing that.

Named ranges are evil except in a limited number of cases (referring to a cell from VBA). If you duplicate a tab or move it to another workbook and you have named ranges, hell starts there.

Also, I find that named ranges seem to slow down the Excel file when the number of rows get a bit large (100k or so)

True of variables in most languages, are they evil too?

In the languages where all variables are global variables, they are.

Sounds like job security. Loving it.

Yeah. Lots of huffing and puffing when python support in excel was unveiled, how dare those dumb business people write python code?!

I think it's great for job security. Whatever happens to the economy, there will be spreadsheets full of shit to be repaired until deep into the 2030s

Generate Spreadsheet Formulas, scripts and excel templates with AI.

Reply on news.ycombinator.com