Comment on Shopify CEO Tobi Lütke tells employees to prove AI can’t do the job before asking for resources.
besselj@lemmy.ca 1 week ago
What these CEOs don’t understand is that even an error rate as low as 99% for LLMs is unacceptable at scale. Fully automating without humans somewhere in the loop will lead to major legal liabilities down the line, esp if mistakes can’t be fixed fast.
…
What error rate do you think humans have? Because it sure as hell ain’t as low as 1%.
But yeah, it is like the other person said: This gets rid of most employees but still leaves managers. And a manager dealing with an idiot who went off script versus an AI who hallucinated something is the same problem. If it is small? Just leave it. If it is big? Cancel the order.
A human has the ability to think outside the box when an unexpected error occurs, and seek resolution. AI could very well just tell you to kill yourself.
Yes. No over worked human would ever lose their crap and tell someone to go kill themselves.
What would happen to such a human?
I mean it is also generous to the Artificial Idiot to say it only has a 1% error rate, it’s probably closer to 10% on the low end. Which humans can be far better than in terms of just directly following the assigned task but does not factor in how people can adapt and problem solve. Most minor issues real people have can be solved without much of a fuss because of that. Meanwhile the Artificial Idiot can’t even draw a full wine glass so good luck getting it to fix its own mistake on something important.
Which humans can be far better than in terms of just directly following the assigned task but does not factor in how people can adapt and problem solve.
How’s that annoying meme go? Tell me that you’ve never been a middle manager without telling me that you’ve never been a middle manager?
You can keep pulling numbers out of your bum to argue that AI is worse. That just creates a simple bar to follow because… most workers REALLY are incompetent (now, how much of that has to do with being overworked and underpaid during late stage capitalism is a related discussion…). So all “AI Companies” have to do is beat ridiculously low metrics.
Or we can acknowledge the real problem. “AI” is already a “better worker” than the vast majority of entry level positions (and that includes title inflation). We can either choose not to use it (fat chance) or we can acknowledge that we are looking at a fundamental shift in what employment is. And we can also realize that not hiring and training those entry level goobers is how you never have anyone who can actually “manage” the AI workers.
how you never have anyone who can actually “manage” the AI workers.
You just use other AI to manage those worker AI. Experiments do show that having different instances of AI/LLM, each with an assigned role like manager, designer, coding or quality checks, perform pretty good working together. But that was with small stuff. I haven’t seen anyone wiling to test with complex products.
Error rate for good, disciplined developers is easily below 1%. That’s what tests are for.
The error rate for human employees for the kind of errors AI makes is much, much lower. Humans make mistakes that are close to the intended task and have very little chance of being completely different. AI does the latter all the time.
CosmoNova@lemmy.world 1 week ago
Yup. If 1% of all requests result in failures and even cause damages, you‘ll quickly lose 99% of your customers.
VanillaFrosty@lemmy.world 1 week ago
It’s starting to look like the oligarchs are going to replace every position they can with AI everywhere so we have no choice but to deal with its shit.