Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A ChatGPT mistake cost us $10k

asim.bearblog.dev

Post reply on HN
Print view View original

321–330 of 526 posts

Re: A ChatGPT mistake cost us $10k

#321

Post by cuppypastexprt »

beala wrote: post #295
AndyKelley wrote: post #193

Two more problems identified solely from the screenshot: * you have two competing subscription id columns. * a uuid is not a string, it is a 128 bit integer. If your database limits to 64 bit integers then use a 64 bit integer for the id instead of a string, or use an array of 128 bytes.

The StripeCustomer table has the same issue. There's both an `id` column and a unique `customerId` column. Presumably the `id` column is useless and could be removed. Also, is there a way to set up foreign key constraints on `userId` with this ORM? That seems like another oversight.

In that blog they say they have now added very robust unit and integration tests, so I don't think this is an issue.

Re: A ChatGPT mistake cost us $10k

#322

Post by n_ary »

My questions are, why did you decide to move on to Python/FastAPI, if you did not understand it well? My second question is, why did you copy-paste as-is from CGPT without doing a review of whatever?

I understand time constraints, but there should be a law or something forbidding using whatever any GPT vomits. In fact, many does it blatantly, so my employer banned using GPT codes and only gave us access to it for /entertainment/ usage.

Re: A ChatGPT mistake cost us $10k

#323

Post by johnny22 »

MrJohz wrote: post #290
sdwr wrote: post #281

Earlier quoted context omitted.

> When you have an argument like foo=obj.whatever(), the obj.whatever() is evaluated at the time the definition of the function is being processed, not at the time when the function is being called. This can't be correct, surely? What if .whatever() relies on internal state that changes after obj is initialized (or after the function surrounding foo is declared, not sure what you're saying)?

It is correct, it's one of the most surprising things about Python and it causes a number of mistakes, even for experts. The easiest way to see this is by running something like this and seeing what gets printed out and when: print("1. start") def function(arg=print("2. func definition")): print("4. func call") print("3. after definition") function() function() function() You should see that the print statement in th…

I find this fascinating because while I don't write much python, this is the behaviour I would assume to be correct based on everything else I've seen. I wouldn't expect it to be evaluated on every call. Definitely shows that your previous familiarity can trip you up (or not)

Re: A ChatGPT mistake cost us $10k

#324

Post by lolinder »

asveikau wrote: post #302
wavemode wrote: post #65

No, a lack of monitoring cost you $10K. Your app was throwing a database exception and nobody was alerted that this was not only happening, but happening continuously and in large volumes. Such an alert would have made this a 5-minute investigation rather than 5 days. If you haven't fixed that alerting deficiency, then you haven't really fixed anything.

I think deploying and then going to sleep is the red flag here. They should have deployed the change at 9am or something and had the workday to monitor issues.

You can deploy and go to sleep if you have monitoring and alerting and someone getting paged. It shouldn't be a human monitoring for issues anyway, so the only reason to choose 9am over bedtime should be that you don't want to risk a late night page, not that someone will actually be checking up actively on the deployment.

Re: A ChatGPT mistake cost us $10k

#325

Post by johnny22 »

gravescale wrote: post #224
bogota wrote: post #167

Earlier quoted context omitted.

This is getting more common. I have already had people try to tell me how something works from a chat gpt summary. This would have led to us taking a completely different direction… 5 minutes of reading the actual docs and I found out they were wrong. Now at a new company i have caught several people copy pasting gpt code that is just horrendous. It seems like this is where the industry is headed. The only thing i ha…

Like people who post "here's what ChatGPTx said" instead of their own answer. Quite literally, what is the point? However, I don't think it's really bad for the technical industries long term. It probably does mean that some companies with loose internal quality control and enough shiftless employees pasting enough GPT spew without oversight will go to the wall because their software became unmaintainable and not use…

> Like people who post "here's what ChatGPTx said" instead of their own answer. Quite literally, what is the point?

Yeah I've seen this and i hate it. If i wanted to know what chatgpt said I'd just ask it myself.

Re: A ChatGPT mistake cost us $10k

#326

Post by Aurornis »

hitekker wrote: post #186
Aurornis wrote: post #5

On one hand, thanks for being honest about a story of how this bug came to be. On the other hand, I don’t think advertising the fact that the company introduced a major bug from copy and pasting ChatGPT code around and that they spent a week being unable to even debug why it was failing. I don’t know much about this startup, but this blog post had the opposite effect of all of the other high quality post-mortem posts…

That's an alright takeaway: the team made a rookie mistake and then they made a PR mistake by oversharing. Otherwise, I think this comment thread is a classic example why company engineering blogs choose to be boring. Better ten articles that have some useful information, than a single article that allows the commentariat to pile on and ruin your reputation.

> Better ten articles that have some useful information, than a single article that allows the commentariat to pile on and ruin your reputation.

Pile-on aside, the problem with this blog article is that it doesn't really have much of a useful takeaway.

They didn't even really talk the offending line in detail. They didn't really talk about what did to fix their engineering pipelines. It was just a story about how they let ChatGPT write some code, the code was buggy, and the bug was hard to spot because they relied on customers e-mailing them about it in a way that only happened when they were sleeping.

It's not really a postmortem, it's a story about fast and loose startup times. Which could be interesting in itself, except it's being presented more as an engineering postmortem blog minus the actionable lessons.

That's why everyone is confused about why this company posted this as a lesson: The lesson is obvious and, frankly, better left as a quiet story for the founders to chuckle about to their friends.

Re: A ChatGPT mistake cost us $10k

#327

Post by sneak »

TheDong wrote: post #303
sneak wrote: post #278

Earlier quoted context omitted.

TBH, if the backend were written in Go, this probably wouldn’t have happened to the extent it did. Somewhere in a log a descriptive error would have shown up. One of the reasons I use Go whenever possible is that it removes a lot of the classic Python footguns. If you are going to rewrite your backend from Javascript, why would you rewrite it in another untyped, error-prone language?

In python it's harder to ignore errors than in Go. In go, I've definitely seen: tx, err := db.Tx() defer tx.Commit() // silently ignores the error on committing, which is the important one That would have masked this error so it didn't get logged by the application. In python, if you ignore an exception entirely, like I did that error above, you instead get an exception logged by default. Python's exceptions also inc…

Yeah, each have their benefits. I run all my go projects through golangci-lint which is required for merge to master/main, so not checking the error value would not have made it to prod.

I suppose there are probably similar checkers for Python that would have caught the passing of a scalar value instead of a function.

Perhaps this is an argument for mandatory linting in CI.

Re: A ChatGPT mistake cost us $10k

#328

Post by pfisherman »

la64710 wrote: post #25

I dont get it , line 56 every time it was invoked would call the uuid4 function to generate a unique id isn’t it? Or was the issue due to the uuid4 function not getting invoked for any reason?

You should generally be setting attributes in the init method of your class, unless you want it to act like a class attribute - shared by all instances. The exception is some special types of classes such as dataclasses or pydantic objects, which do it automatically but require a post init method if you want to do anything more than set values passed to the constructor.

Tbh, if they were writing something for prod they should have factored out the uuid generation and passed values into a pydantic dataclass w/ validators for all the fields. Just sayin.

Re: A ChatGPT mistake cost us $10k

#329

Post by arjvik »

primitivesuave wrote: post #314

I spotted the error instantly. With all due respect to your team - this has nothing to do with ChatGPT and everything to do with using a programming model that your team does not have sufficient expertise in. Even if this error managed to slip by code review, it would have been caught with virtually any monitoring solution, many of which take less than 5 minutes to set up.

To be fair, if I wasn't looking for this bug I never would have spotted it. That being said, you're entirely right that any monitoring or even the most basic manual testing should have instantly caught this.

Re: A ChatGPT mistake cost us $10k

#330

Post by spencerchubb »

paul7986 wrote: post #287
spencerchubb wrote: post #267

Earlier quoted context omitted.

A large language model is trained on vast amounts of text to predict the next token. That tool will have no idea what the current date is, unless the developers augment it by telling it the current date.

Few weeks or more ago I discovered Burger King has a cotton candy slurpee/icee ... I've been enjoying one once to a few times a week. But not all locations have it yet chatGPT seemed to know which locations have it & each one it told me ..they had it available to buy. Very useful so today went to one closest to me which always has it but today they said we no longer do. Thus I asked chatGPT has been discontinued as w…

That is indeed a weakness of chatgpt - it's not so good with super current information

That is because it is only trained on info up to a certain date

It's good to know what tools are good and bad at

Post reply on HN