"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
@hushable I feel you. My workplace has been the same. The latest requirement is that instead of 2 week sprints (which more or less amount to having most of a week of QA and potential last minute fixes) to “sprints of a couple of days because we’re an AI-first company now”
Cool, but 1) before the mandate to use AI people were already under pressure to use it and there have been no productivity gains, so this seems optimistic, 2) are we outsourcing QA to the hallucination machine?
Time to start tokenmaxing and make this as costly a decision as possible
Oops, i had Fable selected all month.
Time for some quality control and optimization.
Claude, take our codebase and generate a set of unit tests to evaluate the code functionality based on the current snapshot.
Now port this code to Rust and add comments describing how each function is different. Run the same set of unit tests to evaluate the new Rust code against the existing code. If any unit test fails by more than 0.01% remove the Rust code and regenerate it from scratch.
Evaluate both the current code and Rust code for readability and documentation. Generate full source documentation for each line of code.
etc etc. You'll have half a datacenter cranking for a week. They won't appreciate the bill...
The attempt to imagine this with stock thinkmeat has left me insensate, emotionally bereft, and mildly insane.
You mad genius. You just named my next Thinkpad.
You were already insensate, emotionally bereft, and mildly insane. The prompt to generate unit tests made you aware that you are mildly insane as you had to create a test for insanity and fail yourself is you scored zero.
It's a fair cop, guv
I have a coworker who went full AI, and I think is actually losing it, like sees them as people. He was telling me he's been asking chatgpt for relationship advice and how I should use it to get advice on my relationships (my wife and I are fine because I prefer talking to her than the robots).
On Friday he was spamming me with "issues" that codex found with my PR. I think he sent over 20 bulletpoints. 1 of them was a small bug that would only apply in a way far out edge case but good to fix. The rest was absolute nonsense, or evidently not an issue if a huge read it and thought about the context.
I ignored him all day, but still had to take the time to read and verify everything. At 4pm he sends me a message asking me to fix the bugs or he can have codex fix them. I told him they were nonsense and not applicable, explained each one in simple terms, then 20 minutes later I got a message "codex said there's still these issue: <the same fucking list that he evidently didn't read>".
This has been going in for a while. I spend half my week fixing his AI slop that he genuinely thinks is gold. I swear 4 months ago he was a very solid developer, today I wouldn't trust him to write an HTML page.
I think the easiest way to make an immediate change is to never call/talk to, or about it, as you would a person.
I dont use LLMs at all, but my wife does an i always correct her to say "It said..." when she tells me "He said..." every time she uses it instead of google to get an answer. (And then tell her to find me a trusted website that backs up the LLM as well)
Dehuminise it so theres always a part of your brain treating it as a machine/device, not a person you can trust.
Its like saying "my car broke down, she isnt feeling well".
The car broke down, it needs a part, do i trust my car? Usually but i wont be suprised if it doesnt work some times.
And i dont expect it to be able to swim if i try make it instead of using the vehicle i should have been using.
I spend half my week fixing his AI slop that he genuinely thinks is gold.
God I hate this. The sheer number of times I've heard "and I didn't even write a line of code" and I have to bite back the "yeah, I can fuckin' tell."
It can happen to anyone.
I think the first documented cases were inside Google, from scientists who know exactly how they work:
https://www.cbc.ca/news/science/ai-consciousness-how-to-recognize-1.6498068
I seem to remember a debacle earlier than 2022, but can’t find it… but point is, I suspect its quite a human psychological vulnerability. So if they recover, I wouldn’t be too hard on your coworker.
That's really sad... That is worse than what I thought was annoying at my work today. Claude indicated something was an issue that has been "an issue" for years and never actually been an issue. It suggested over engineering the fuck out of something we just plainly don't need to do. I was annoyed we spent 10+ minutes talking about that. What you described is so much worse 😬
Same seems to happen with some healthcare software in Germany. For example, Compu Group Medical is in the news for massively losing customers for bad software quality, and they have gone all-in on AI.
I get notifications of my prescriptions being sent off to the pharmacy via a service called RXInform.
They recently incorporated an LLM feature. All this does for the end user is take a few seconds to type of what the prescription is, where it’s being sent, how much it’ll cost without insurance, if there are any coupons, and the insurance price.
All of this is incredibly simple and was successfully displayed using a template. The way the information is presented never changes. The only thing the LLM does is increase the time it takes to actually see the full text while making some server somewhere works harder than needed.
Edit: fixed typo
Also while feeding personal health information into a black box somewhere with no real way of knowing what BlackBox© company is collecting or selling in the mean time.
Whileas the pharmacy close to my house will get what you ordered if you just mail them a photo of your last package.
Myepic is used pervasively in US healthcare system, they likely use AI in some form. because the MDs use AI to write thier notes for sure.
I'm retired, so I don't have to deal with this AI garbage... but I have a friend who's a manager of a decent-sized office at an accounting firm.
He told me recently that, not only are all employees required to use AI in their daily work, but my friend has to write weekly reports to his bosses on how AI is improving workflow in the office. He's not allowed to say, "It's not." Only positive feedback. They need to justify its continued use, so he needs to find benefits to report to the higher-ups.
Sadly, he's the only one who doesn't like/use AI. Everyone else in his office has basically replaced their jobs with it. Every report that comes across his desk is AI nonsense, which he has to spend time fixing because his subordinates don't know how to write reports without AI assistance.
I do not envy my friend.
Isnt the answer to then send on the un-fixed reports and say "Oh i had AI check it and it said that was all good, are you saying AI isnt working?"
. He’s not allowed to say, “It’s not.” Only positive feedback. They need to justify its continued use, so he needs to find benefits to report to the higher-ups.
Those higher-ups should not be in charge.
This feels like the emperor's new clothes, except no one listens to the children pointing out the folly.
This feels like the emperor’s new clothes,
I guess it does not only feel like that. It's the same thing.
my friend has to write weekly reports to his bosses on how AI is improving workflow in the office. He's not allowed to say, "It's not." Only positive feedback. They need to justify its continued use, so he needs to find benefits to report to the higher-ups.
All right Claude, you know the drill. Write this week's report.
AI is improving workflow in the office by writing these reports for me so I can do actual work.
CLAUDE.md:
- Write a weekly report on how AI is improving workflow in the office
- Only positive feedback
- MAKE NO MISTAKES!
my friend has to write weekly reports to his bosses on how AI is improving workflow in the office. He's not allowed to say, "It's not." Only positive feedback. They need to justify its continued use, so he needs to find benefits to report to the higher-ups.
"Please make us blind to all consequences!"
At an accounting firm no less... Who gets the blame if some number turns out hallucinated? The lower level employees or your friend? (assuming C-levels never carry blame)
That is going to be a glorious dumpster fire. You have my sincere sympathy.
I was fired for pointing out dangers of AI and data centers. I would do it again
Thank you.
I've been looking for a job for the past 6 months. Every interview I had (more than 50) hinted that some Claude or Copilot was being used more and more. Some companies even want to hire "junior AI engineer." How the fuck can you hire a junior with a non-existing degree and that knows nothing about a technology so recent?
I found one and only one company who officially rejects AI, and I hope they'll hire me.
Let me guess, 10 years experience needed?
Not yet, but they all have various job titles like AI developer, AI architect, AI devops, AI anything... It's like they stick the word AI to traditional roles and expect people to understand what it's all about (or pretend it's a job that exists at all, it's a scam after all).
WTF?!? Where are you getting all of these interviews? I’ve been looking for a job for 19 months and had two interviews.
Unions solve this issue btw. Would love to see software development work more like carpentry or public works with a trade union and work hall than the current cult of personality or culture of staying at a single employer your entire life.
Fork the repos that matter now, and squirrel them away somewhere, just in case someone gets froggy and gives Claude write access to your repos and fucks the history of something interesting like that. Always good to have a backup contingency.
I'm of the opinion that companies that make stupid AI decisions so they can fire people shouldn't get bailed out by a good decision made by one of the people they're planning to fire.
Let them burn.
Because the goddamn suits think AI is going to make them more money, despite that it practically uses mined/stolen data and consumes more electricity than a small city.
For the most part, I am an AI sceptic. My firm has had access to and beta tested every Google AI model since before the AI had a name (yes, well before Bard). We also have access to ChatGPT, Claude and locally run models on dedicated hardware. I'll save you the suspense, at the moment it's all mostly trash. Barely a proof of concept. Definitely not production ready, but there are a few exceptions I, personally, have found.
Gemini Pro is excellent at identifying and diagnosing sick garden plants if you take and upload a photos of the plants. Unfortunately, it still cannot do simple arithmetic. A simple SUM that any spreadsheet does automatically by just highlighting the cells, and somehow it was still off by one.
For software dev, the only one product we have found to be useful is Claude, BUT it really depends on the model. Haiku is a joke. Sonnet is not bad, if not bad is ~50% accuracy. Opus is fairly descent at building scaffolding that you can then correct and complete, but don't expect efficient code.
The only model we have ever tested that can deliver ALMOST fully functional code, reliably diagnose errors, and review code is Mythos/Fable. Genuinely not absolute shite, BUT that shit is extremely slow and outrageously expensive. I seriously doubt that your company is forking out the money for Claude Fable as their default model, and if they are, they must not care about ever making a profit.
There was another similar post not that long ago.
Manager says "we don't touch code any more, we only tell claude what to do". Part of the thinking is that claude will improve as you teach it.
There's a lot of hype and hyperbole from both sides. Managers thinking claude can replace entire teams and commentary saying it can't explain how to tie your shoes.
I haven't used claude itself so I just don't know, but things like Mistral and other models available on huggingface just aren't in this kind of league.
I feel like the atrophy of cognitive function and high level skills is the other side of the AI-in-the-workplace coin. It just seems obvious to me, and I'm sure to any professional, that they wouldn't have developed their skill if they didn't... develop their skill. You can't really develop coding skills by looking at what claude did.
Part of the thinking is that claude will improve as you teach it.
This is just a complete misunderstanding (by the manager) of how LLMs work. Sure, they probably log conversations as training data for a new model, but session to session, it does not improve from talking to it. The model is fixed. There are strategies which boil down to having the LLM write a diary of things it should "remember" which it subsequently has to reread when relaunching, but those suffer from fundamental limits, only so much can be stored before it runs out of context.
It's so frustrating how much imaginative thinking and anthropomorphization is applied to these tools...
Sounds like someone doesn't know about how copyright works for AI code or any possible liability involved in distributing it. The walls will come tumbling down one of these days, then they'll turn to you to fix it.
Where I work the code quality is so bad LLMs were a net improvement
Better polish your CV and stick the popcorn in the microwave. Start sending applications when the first customer complaints start popping up. You should have a new job by the time the company’s reputation is in flames.