Getting tired of “AI said this was this issue”
Posted by Alces_@reddit | ExperiencedDevs | View on Reddit | 28 comments
At least once a week, I see a comment on a bug that is agent generated that multiple paragraphs long with section headings like “1.The Catalyst 2. The Perfect Storm” and the dev posting the comment not doing ANY investigation further.
_5er_@reddit
Show them videos of Husk IRL
dvorgson@reddit
AI has been a huge blow to human expertise. Not because it's smarter than a human, but because it has convinced people to ignore the invaluable expertise of a well educated human
davenirline@reddit
And then they call you a "Luddite" if you resist.
_ak@reddit
The irony is that one the big thing the Luddite protests was preserving quality. They thought that automation not only destroyed their job, but also the quality of products that they had been so proud of.
exapunk_11@reddit
And soon in the culture that we're leaving behind for decades to come, if people end up being deterred by not becoming creatives. But this is a whole different can of worms no one is willing to talk about.
new2bay@reddit
I don't think that will happen. People said the same thing about photography when it was invented, but there are still painters. They said the same thing about digital photography, but there are still people who shoot film.
dbenc@reddit
there are a limites number of educated humans and an unlimited amount of available slop 🙈
420-code-cat@reddit
extremely well said
Intrepid-Ostrich2226@reddit
These people never played video games with AI advices. You ask something and immediately get feedback from the game if AI was wrong. I mean some games where you have lot of characters, their abilities, weapons and so on and should make your own excellent build.
F1B3R0PT1C@reddit
YES! I get it from both ends now. Product owner gets a new feature and copy and pastes crap from an AI about what the “actual” solution is. Another scenario, a customer escalates a bug and it ends up in my lap somehow. I ask my boss what he knows about the issue, and I get back a giant wall of text from an LLM explaining it. I hate it. All the burden of cognitive load is put on me because my coworkers and boss can’t be bothered to think about an issue for themselves. I have to parse the insane ramblings of a hallucinating machine with my puny meat brain because it’s my responsibility to actually take action using the information buried in the walls of text they blasted at me. I find the whole thing disrespectful.
ProButterscotch@reddit
Man I am in this boat
Antice@reddit
I have one reply I give when they do this.
"I asked for a clarification of the users experience. Not a disertation about what the error could be. Please refrain from using AI for tickets in the future as per the policies set here: "
codescapes@reddit
I know that bringing up Dunning-Kruger is such a "Reddit thing" but AI genuinely does make it way more relevant. We all now have superficial expertise in whatever we need and with that it's extremely easy to get an overinflated sense of your own intelligence, knowledge or ability.
I am not saying AI tooling cannot be mega helpful but it does not turn you into some kind of polymath / Leonardo da Vinci. We all have it, we can all go on these 'research quests'. I am not just talking about coding but whatever other disciplines you can now easily research - law, medicine, history, philosophy... Hell my foot is sore from running so I looked up stuff that might help - I am not a podiatrist!
It's so easy to get an unattributed surface-level understanding of things. I was in the Yucatan earlier this year and went on a whole thing learning about Cortez and the Spanish conquistadors. I was asking increasingly obscure questions (I wanted to understand the war dogs, supply lines, gunpowder efficacy etc) and it always had an answer but when I pressed for sources it did the AI equivalent of shrugging or vaguely pointed to the writings of Bernal Diaz. A conquistador who wrote about his experiences 50 years after the events as an old man.
I supplemented that with a bunch of podcasts and got a vivid picture of things but I am no expert, just someone who went on holiday and wanted to know part of the history of the region. Maybe 20-30h total 'research'. There are people who spend their whole lives understanding that period, becoming SMEs intimately familiar with basically every written account we have.
Point being, we shouldn't let ourselves go on ego trips or get an overinflated idea of our understanding of things we only became aware of 5 minutes ago. Prompting does not make us Gods, just people using the same tools available to everyone else.
Fluffatron_UK@reddit
Then raising that to management and they go "ohhh that shouldn't be happening! You should push back on that." Yes, obviously I do... But guess what else you tell them Mr management, use copilot for everything!
I don't think I want to be a professional developer anymore.
superdurszlak@reddit
It's really amazing how nowadays 90% of manager's job is telling his/her reports they should have handled things themselves.
And, surprisingly, if your pushback is ineffective because you're some nobody that no one respects when you push back yourself, the next step is to escalate to your manager and let _them_ handle the pushback as someone more important with slightly better chances at being listened to when they say "stop doing that". But then you're often back to square one.
Antice@reddit
If management got your back. Start refusing AI tickets. Set up apolicy document outlining what you actually want and need. Also outline what will cause an automatic refusal. The dumbest ones will just feed that document to the AI to fake it, but that is still better inputs on a ticket than pure slop.
superdurszlak@reddit
Any time someone replies to my concerns with "what AI says about that" I'm temped to respond along the lines of:
> Thank you, I wouldn't have thought of prompting first, and I'm in dire need of my personal ChatGPT copypaste assistant...
gentlychugging@reddit
Getting tired of AI would have been enough
CodelinesNL@reddit
Yeah it's a problem, and one we need to keep tackling. I've had this talk with one specific developer two times already.
For my with a small team I'm leading it's less of an issue since I can set the standards. But in large companies who are now all on the AI hype train with managers who don't understand software engineering; good luck.
ecko814@reddit
Or when your manager generates a ticket with 10 issues for me to manually verify. Or AI generated pull request, then assigned me the ticket to take it across the finish line.
vocal-avocado@reddit
lol getting an AI slop PR to “clean up” must be infuriating.
Zookeeper187@reddit
This is the worst, and from fellow devs as well. Big AI PRs that they didn't bother to check and offload code review to others. I just don't give a f any more to review overcomplicated ai code for them.
Flashy-Whereas-3234@reddit
Did you read this?
No?
Then why the fuck am I reading this.
Decline.
FastHotEmu@reddit
This! This right here. Officer, this post makes me sad, it's exactly what I an seeing.
Bronkic@reddit
"Getting tired of AI ~~said this was this issue~~"
FTFY
OkidoShigeru@reddit
Our team has been given license to just ignore any obviously AI-generated issue reports/massive PRs left with no further comment in our domain. We’re fine with AI-generated PRs that are actually driven by a real, trustworthy human that we know will give it the proper scrutiny and attention through the review process, and yeah issue reports that originated from AI but someone has actually, you know, looked at it and done further investigation themselves.
But yeah, people just copy pasting the output of an LLM is just noise that wastes everyone’s time…
UnderstandingDry1256@reddit
culture issue
Engineers should learn how to use AI. Just a new tech in stack - if you can leverage it, you're in the top. If you can not - you are replaced, not by AI but by folks who know how to use AI.
Fun times
YakaryBovine@reddit
Complete non sequitur. The OP is talking about a specific problem with how their coworkers are imploring AI - how does your comment connect to that?