"The Workslop Tax" by Marketoonist comic / Tom Fishburne - #GenAI has been earning a reputation for #workslop: polished-looking but unhelpful low-quality content that others need to rework or correct, if anyone does.

[unchartedworlds@scicomm.xyz]

@KO6YQ

That's a good one!

[failedlyndonlarouchite@mas.to]

@KO6YQ

I don't understand why my actual, lived real world experience is so diff from this cartoon

at the biotech I worked at until last year, all of the PhDs, engineers, and programmers - mostly smart people IMO - they all found AI to be a very powerful and extremely useful tool

[aligorith@mastodon.nz]

@KO6YQ
This is a major pain point for our team currently

Other team whose work we have to inspect to avoid it breaking our live system:
* Yo, here's 3 code reviews vibe slopped in parallel. (The actual code changes may be mostly normal looking, but the unit tests are completely off the rails, but achieve 100% coverage!)

* We've also autogenerated all the required docs:
* Screeds of verbose tables for every type / property / method changed
* Even more verbose docs on the unit tests - And as of 2-3 weeks ago, more tables on how these satisfy all the formal requirements / test criteria
* Rinse and repeat for the Integration Testing, except now maybe including screenshots for the cases we insures on seeing some

[ko6yq@pnw.zone]

@aligorith As I also mentioned to another similar response, the process needs some form of pushback/feedback going into the originators' metrics. Without feedback, nothing would stop them from continuing to abuse their advantage.

[larsbrinkhoff@mastodon.sdf.org]

@datarama @KO6YQ I reject those code reviews.

I tell my colleagues a good pull request is like a sales pitch. It should be split up into a few slides, and each should make a short and compelling argument.

A sales pitch that consists of just one big slide with a hundred lines of dense text will not get the point across.

[runrichrun@mastodon.social]

@KO6YQ
This cartoon is sadly great. Unfortunate.
@mastodonmigration

[nicoh@piaille.fr]

@KO6YQ exactly what we are facing. 2 sec to generate 30 minutes to read and comment. Asymmetry is unbearable. I see 2 ways of reacting to this (and actually were experimenting both right now):
1. Let them fight - ie. Get your own AI read comment and identify inconsistencies
2. Not on my watch - ie. I don’t accept your AI sloppy content, please let’s get in touch so we can talk, and summarise the idea with our brains.

Both are not with satisfying results yet. Any experience or other idea ?

[jabgoe2089@hub.netzgemeinde.eu]

@Ian Kluft ️ so true. but sometimes it also goes the right way: up. i already heard upper management complaining about the large amount of elaborate output they get from their underlings when they ask for infos and data ...

[semantikast@openbiblio.social]

@KO6YQ ... and very telling that they printed it out. (Symbolbild German public service)

[paulschoe@mastodon.world]

@failedLyndonLaRouchite

AI is a useful tool. But it depends on what job you have.

If you have to create, your output explodes. Everybody happy.

If you have to review, your input (that what you have to review) triples while you now have to check whether all the references are correct (hallucinations). Human creators don't tend to hallucinate.

Triple the lines = triple the time it takes + additional checks => death to productivity
+ responsibility for hallucination checking.

#AI

@KO6YQ