If you expect ChatGPT to give you information or direct you to it (like Wikipedia or Google) you will be frequently disappointed. You may also be frequently pleased, but you often won’t be sure which and that’s a problem.

ChatGPT is very good at transforming information. You need to show up with stuff and then have it change that stuff for you somehow. You will be disappointed less often.

It’s not good at math. It can do stuff sometimes, and you can easily teach it to use a calculator.

What’s similar that it would be good at is extracting what you bought and turn it into a list and formula.

[Me] Extract the list of things that I bought and their prices. Write a shell command to figure out how much I spent.

I spent 200 on groceries and paid my 70 phone bill. I was going to buy three pairs of pants that were $40 each. I didn’t like the colors. I bought a t shirt for thirty. And I did end up buying one pair of pants. But it was 5 dollars more than those other ones. And I bought bag of tea for every day next year, they were a quarter each.

[ChatGPT] List of things bought and their prices:

Groceries: $200

Phone bill: $70

T-shirt: $30

One pair of pants: $45

Bag of tea: $0.25

To figure out how much was spent, you can use the following shell command:

    echo $((200 + 70 + 30 + 45 + 365*0.25))

This command will add up the prices of all the items, including the bag of tea for every day next year. The output will be the total amount spent.

I pay for ChatGPT Plus, and use it with no delays at all dozens of times a day. The more I use it the better I get at predicting if it can be useful for a specific question or not.

What do you use it for? I'm assuming code related? I've found it useful for some boilerplate + writing tests and making some script and some documentation.

I'm curious what you or others that use it all day use it for especially if it's not for programming?

I was just working on a small exploratory project in Python. I used sys.argv because it's so quick to prototype with.

When I started refining the project for longer term development, I wanted to convert the CLI to use argparse, so I could build a more nuanced CLI. I gave GPT a couple example commands I wanted to use, and in less than a minute I had a fully converted CLI that did exactly what I wanted, with more consistent help settings.

I can do that work, but it would have been 30-45 minutes because there was one setting in argparse I hadn't used before. That alone was worth this month's $20.

For more complex and mature projects I could see having to give GPT a minimum working example of what I need instead of the whole project, but I can already see how it will enhance my current workflows, not replace them.

See, a better LLM or an experienced human would have told you to use docopt instead of argparse and saved you more time in the long run.

There are some examples of code things I've done with it here: https://simonwillison.net/2023/Apr/7/chatgpt-lies/#warn-off-...

In terms of non-code things, here are a few from the past couple of days:

- Coming up with potential analogies to explain why it's OK to call something a "language" model even though it doesn't have actual understanding of language

- Exploring outliers in a CSV file (I was playing with the new alpha code interpreter, where you can upload a file and have it run and execute Python to evaluate that file)

- Asking some basic questions about GDPR cookie banners

- Figuring out if there are any languages worldwide where "Artificial Intelligence" translates to something that doesn't include the word "Intelligence" (found some good ones: Finnish translates to "Artificial wit", Swahili to "Artificial cunning")

- Understanding jargon in a tweet about payment services - I wanted to know the difference between a "payment aggregator" and a "merchant acquirer"

- As a thesaurus, finding alternatives to "sanctity" in the sentence "the sanctity of your training data"

- Fun: "Pretend to be human in a mocking way", then "Again but meaner and funnier"

- "What should I be aware of when designing the file and directory layout in an S3 bucket that could grow to host millions of files?"

How can you trust that whatever the expression used in Finnish to denote "Artificial Intelligence" uses "wit" when translated back into English? Words in one language often don't have an exact counterpart in another language, and I'd be especially wary when it comes to languages from different families: Finnish is notorious for being one of the few European languages not part of the Indo-European family.

It may very well turn out to be the right translation, with the appropriate connotations; but without some clarification and confirmation from a native speaker, I would not trust it.

"How can you trust that" - I can't. This was idle curiosity. If I was going to publish this or do anything beyond satisfying my idle curiosity I would consult additional sources.

thing is, it's not load bearing information.

> Finnish translates to "Artificial wit"

Yes and no: the root of the word (and of the meaning) is still the same as in the word for intelligence.

tekoäly "artificial intelligence, artificial wit"

äly "wit"

älykäs "intelligent, having wit"

älykkyys "intelligence, the property or extent of having wit"

One dictionary definition for 'äly' is 'älykkyys' and vice versa, so they can be used more or less as synonyms but with some different connotations. I don't know how 'tekoäly' was chosen but it might have been because it's shorter and/or the grammar is nicer.

It's very good for taking roughly structured data and turning it into strongly structured data. E.g. someone with low data literacy or a program not designed to allow data export gave you something that needs to be made useful... In my day to day work (aside from programming) I find it helpful in writing stories and documentation.

I was working on some code I didn't originally write that looped over a directory recursively and took audio fingerprints of files, then saved them in a SQLite database. I pasted the code in, then gave orders like "Rewrite this to use pathlib." to which it happily did.

Okay, next I notice it uses a .glob of "*.*" and so that makes me a bit suspicious.

"Will this code work for audio files without extensions?" Nope. It fixes it.

Okay, now I notice the original code builds up all the fingerprints in memory, then adds to the database. So I order it:

> Add files to the database as we go, and print progress every thousand files

Boom, just as I would have done it.

Then I note that we don't seem to have good indexes, and it's like you're right! and puts an index on the audio fingerprint field.

Then I ask it to save more info than just the file path, and it adds the size of the file. Then of course, I tell it:

> This is remarkably slow, how can we significantly speed it up?

To which it replies:

> Here's an example implementation using multiprocessing:

Awesome, it works!

Oh, here's an error I got when I ran it when the audio fingerprinting library errored, handle that.

Paste in the error, it fixes it.

The thing is, I would have no problem with doing any of this stuff at all, it just made it so incredibly much faster, and if it does something you don't like, you can so easily correct it. Hope this helps you understand how I use it!

I pay $0 for Google with no delays. I don’t understand why I’d want to pay for information that’s less reliable. (I’m being slightly dense, but not really)

It's not useful for every search, but I find it useful when I can't come up with a search query that will give me the results I want.

For example, ask it to recommend a good paper to read on some topic, then use Google to find the paper. If it made up the paper, you'll find out soon enough.

Also, when you remember something very vaguely, it can be a way to get a name you can search on.

I use it to shortcut the process of knowing what to search by generating code examples. Eg, I know what I want to do, I'm just not sure what methods $framework provides for doing so. I develop a sense of it's limitations as I go (eg, some frameworks it gives me methods that have been moved or renamed or deprecated, and so I check for those things with those frameworks).

When I was learning to program, and I'd get stuck to the point I needed to ask for help, it was a whole production. I'd need to really dot my 'i's and cross my 't's and come up with a terse question that demonstrated I had done my due diligence in trying to find the answer myself, pose the question in a concise and complete way, and do all of that in about 250 words because if it was too long it would get lost in the froth of the chatroom. And I'd still get my head bitten off for "wasting" the time of people who voluntarily sat in chatrooms answering questions all day. And I can understand why they felt that way (when I knew enough to answer questions I was just as much of a curmudgeon), but it was a pain in the ass, and I've met people who really struggled to learn to code partly because they couldn't interface with that system because they weren't willing to get their heads bitten off. So when they got stuck they spun their wheels until they gave up.

ChatGPT just answers the damn question. You don't have to wait until you're really and truly stuck and have exhausted all your available resources. It doesn't even have the capacity to feel you've wasted it's time.

I'm concerned about LLMs for a million different reasons, and I'm not sure how people who don't already know how to code can use them effectively when they make so much stuff up. But when I realized I could just ask it questions without spending half an hour just preparing to ask a question - that's when it clicked for me that this was something I might use regularly.

Google is a worse product than gpt for some queries. And it's funded by ads which distorts incentives.

Take a look at these four examples: https://simonwillison.net/2023/Apr/7/chatgpt-lies/#warn-off-...

Each of those are things that would take WAY longer for me to research or answer using Google.

lmao imagine being so cheap you care about 20 bucks.

I feel like alot of engineers are here in just in denial. Google is trash nowadays. ChatGPT can write code at the level of a good junior dev. The comparison is just so laughable.

It's a considerable sum for many people in many countries. Free on the other hand is equally free everywhere.

> I don’t understand why I’d want to pay for information that’s less reliable.

It's faster. Much faster in a lot of cases. Think of ChatGPT not as a precision source of answers, but rather a universal clue generator.

Yes, ChatGPT frequently (perhaps even usually) produces subtly defective answers. Most of the code it generates won't interpret, compile, etc. without errors, frequently exhibiting simple minded mistakes. That doesn't make the output useless. Indeed, it is a great time saver. That may seem counterintuitive, but I assure you it holds.

I had to write some Perl today. I end up dealing with that about once a year. I have a deep understanding of the language (going back to serious use in the 90's,) but it's been 20+ years since I dealt with it frequently enough to achieve any sort of "flow." ChatGPT and its imperfect answers were extremely helpful in getting through this efficiently. If I had to guess I was 3-4x faster than same exercise using convention search engines.

Google, now more then ever, feels like a ghetto. Useful results from Google are always buried among low quality bait that takes time to wade through, sorting wheat from chaff. ChatGPT obviates that nonsense, producing distraction free, ad free results. Also, Google isn't some fount of truth; lots of wrong stuff makes it to the top of Google.

Sign up for the API and use the playground. You don't get the plugins, but you pay per usage. GPT-3.5 is super cheap, and even GPT-4 isn't that expensive. My first month, when I had only access to GPT-3.5, I didn't even break $1.00; and now that I've gotten access to GPT-4, I'm at about $3. I've only once had it tell me that it was too busy for the request; I tried again 30 seconds later and it worked.