Post by @HeavenlyPossum@kolektiva.social

@HeavenlyPossum@kolektiva.social 14 hr. ago

I had to attend mandatory training on the use of “AI” in the workplace today and two of the rules they told us are:

1) we can’t use AI to do our work for us, such as by asking it to reach conclusions that we might pass off as our own work, and

2) we can use AI when we have more data than is reasonable for a person to analyze, by feeding the data into an LLM and asking it to identify patterns for us.

When I tried to explain that these are literally the same thing, and that the latter is just pretending to be the sci-version of the former, the instructor got really uncomfortable and changed the subject.

7 Boosts

0 Likes

Cho

@chojzina@kolektiva.social 13 hr. ago

@HeavenlyPossum I work low-level government work and the AI training literally just consisted of a short (inaccurate) explanation what AI is and the plea to never use this tool without crosschecking its output.

The following dialogue is true, so help me Cthutuhl:

Me: So...you want us to put in all the crucial personal information into a system we have no control over when we all signed paperwork that could cost us our job when we give away crucial information?

Instructor: Uhmm, well...

Me: And then you want us to check the output by doing ALL THE REGULAR STEPS OF OUR WORK as we would do without to make sure it's correct and compliant with German law?

Instr. Yes, exactly.

Me: You just added a step to our regular process. Where exactly is the win in efficiency and productivity?

I swear to the flying spaghetti monster that this broke the man. I do not enjoy being mean but I just couldn't help it.

HeavenlyPossum

@HeavenlyPossum@kolektiva.social 13 hr. ago

It’s hard to convey the scale and scope of how cooked a generation of people are by LLM slop.

“You can use AI to improve efficiency but make sure you double-check all its output to make sure it isn’t just hallucinating bullshit” thanks that’s actually more work and thus less efficient than not using an LLM at all.

“You have to double-check LLM output, but also if the task is too complicated or complex to figure out on your own, you should use an LLM” thanks but if it’s too complicated to figure out on your own, then you can’t double-check the output.

Just absolutely cooked.

Mateusz 🏳️‍🌈

@aemstuz@mastodon.com.pl 12 hr. ago

@HeavenlyPossum Oh come on! We both know well that they're just saying "you have to double-check" not to sound evil but they silently count on you not to check the output and just do your work quicker (and less accurate 🤷‍♂️)

David Gerard

@davidgerard@circumstances.run 8 hr. ago

@aemstuz @HeavenlyPossum the one weird trick

HeavenlyPossum

@HeavenlyPossum@kolektiva.social 13 hr. ago

Back in my day, we had a process for analyzing large quantities of data for patterns: statistics.

It required first collecting a lot of data, and then parsing it, cleaning it, and sorting it. And then it required some knowledge of math or, failing that, at least the software that did the math for you.

LLMs pretend to do the same thing for people who are too lazy to actually collect and parse data and don’t even know that there’s a math option. They just want to feed a bunch of text into an LLM and then pretend that the output is actually related to the input, rather than a hallucination of text that the LLM thinks is statistically correlated to the text you input.

There will come a point at which those few people who have not cooked their brains on LLM slop will be the most important people in the world.

Rich Puchalsky ⩜⃝

@richpuchalsky@mastodon.social 13 hr. ago

@HeavenlyPossum

Ordinary processes of our society are now doing more workplace sabotage than the most dedicated anarchist saboteurs could ever do.

A.A. Delafin

@delafin@climatejustice.social 13 hr. ago

@HeavenlyPossum "AI is only to be used by people who understand the subject matter, so they can check its validity."
vs
"AI gives everyone access to things that were previously only done by experts!"

Similar contradictory statements that are routinely touted by slop-pushers