Ask Lemmy

Take the time to do it right the first time but also don't waste time if it doesn't add value.

Having a process is great but if the process exceeds the value then the process not only harms profit margins but also erodes morale. If the reason a process exists is to counter bad behavior then it's an employee problem not a process problem.

Open office floorplans are a terrible idea!

Work from home shouldn't be considered a given based on the job tasks but a privilege and benefit extended to those employees that have shown the discipline and reliability to work from home. But the in office requirement shouldn't be forced on everyone just to satisfy a "butts in seat policy" or a managers insecurity.

Modern PLCs are indistinguishable from IPCs with an RTOS and there's no reason I shouldn't be able to use a proper language for them - with a stdlib and external library support. But manufacturers defined the term and have the industry hostage so you have to buy semi functional libraries and can't use git, unit testing or other automation.

Snapshot tests suck. That's a test that stores the dom (or I guess any json serializable thing) and when you run the test again, compares what you have now to what it has saved.

No one is going to carefully examine a 300 line json diff. They're just going to say "well I updated the file so it makes sense it changed" and slap the update button.

Theoretically you could only feed it very small things, but if that's the case you could also just assert on what's important yourself.

Snapshots don't encode intent. They make everything look just as important as everything else. And then hotshot developers think they have 100% coverage

Do not power law fit your process data for predictive models. No. Stop. Put the keyboard down. Your model will almost certainly fail to extrapolate beyond the training range. Instead, think for at least two seconds about the chemistry and the process, maybe review your kinetics textbook, and only then may you fit to a physics-based model for which you will determine proper statistical significance. Poor fit? Too bad, revise your assumptions or reconsider whether your "data" are really just noise.

Always run qNMR with an internal standard if you are using it to determine purity. And, as a corollary, do not ignore unidentified peaks. Yes, even if it "has always been that way".

DOE models almost certainly tell you less than you think they do, especially when cross-terms are involved, or when the effects are categorical, or when running a fractional factorial design...

Your favorite AI enabled LLM does a very, very good job of simulating language tests based on previous tests and there's no reason at all not to use it to study and prepare.

Don't cost-optimize people's homes. Just don't.

The amount of times I've gone to a maintenance job with a description "people are cold", only to see a plaque on the wall stating that this building has been optimized by Company X is actually infuriating.

And the worst thing is that they inject their proprietary, remote control system on top of the original automation. This means that I can't change anything without literally reprogramming the entire site.

So I'm standing there, trying to figure out how to tell an 80yo lady "you're cold because the building managers want to save some money" without going on an anti-capitalist rant. This has had a success rate of around 30%.

Abilify is a beautiful long term maintenance med but wholly inappropriate for an acutely agitated and combative patient.