I wish they’d offer an llm free version with no cap on searches. Their products are too expensive and it feels like it is mostly to pay for the llms. I can’t justify paying that much for a product I am never going to use.

One of my big beefs with ML/AL is that these tools can be used to wrap bad ideas in what I will call “Machine legitimacy”. Which is another way of saying that there are many cases where these models are built up around a bunch of unrealistic assumptions, or trained on data that is not actually generalizable to the applied situation but will still spit out a value. That value becomes the truth because it came from some automated process. People cant critically interrogate it because the bad assumptions are hidden behind automation.

Yes they lost the civil war. We know this because the CSA does not exist any more. I think that losing your entire country and failing to achieve any strategic objectives is a pretty good definition of “lost a war.” The fact that traditions of racism still existed in the southern states during reconstruction does not mean that the CSA won. That’s silly.

Thanks for reaffirming my bias that new cars suck.

I am really concerned the next car I need to buy, which is probably 20 years off, is going to be this trend cranked to 11. With software and hardware I can find alternatives and hack my way around the “you paid for it, but we own it and can do whatever we want to it” mentality that tech companies push, but cars seem like a whole different world when it comes to the “you paid for it, but we own it” mentality.

This really isn’t a study, so much as a lit review. Sort of. Anyway, in the fully remote section they cite three studies that argue show a fall in productivity. The first (Emmanuel and Harrington (2023)) found an 8% drop in call volume as a call center shifted to fully remote work at the onset of the pandemic. But their comparison group was a group of call center employees who were always remote. So even if you buy the argument that the change call volume is solely attributable to a drop in productivity, you cannot conclude that the productivity shift was caused by working from home, the group that shifted from on-location to remote work did 8% worse than the group than the always remote work!

The second study (Gibbs, Mengel and Siemroth (2022)) is, again, an analysis of call-center employees (this time in India) who shifted to remote work at the onset of the pandemic. They find no change in productivity, but that employees are working longer hours at home, which they argue means a real 8-19% drop in productivity.

The final study (Atkin, Schoar, and Shinde (2023)) is another firm from India which involved a randomized controlled study which finds an 18% drop in productivity for data entry work.

So, just taking their lit review at face value, one of their studies directly contradicts their argument, yet they somehow present it as if it is evidence of a causal relationship between working from home and productivity. Another study shows no effect, so they break out some razamataz math to try to turn no effect into a negative effect. Only one of the three studies shows a plausible effect.

Since these are the only three papers they cite to support their argument that fully remote work causes a drop in firm productivity, let’s look at them in more depth.

If you go to their references section, you find that there is not a Emmanuel and Harrington (2023) cited. Hey, that a bad sign. There is an Emmanuel and Harrington 2021, but its an unpublished paper. Maybe it got published and they just forgot to update the cite? I plugged the title into google scholar, and find one result, with no copy of the working paper, and no evidence of any sort of publication record from any journal. Plugging the title into regular google returns a “Staff Report” of the federal reserve bank of NY. So not a peer reviewed article. They employ whats known as a difference-in-difference design to compare employees who shifted from fully in person to fully remote. They report a 4% reduction in productivity for these workers, not the 8% reported in the original article. I just skimmed the article, so maybe they get their 8% figure someplace else. What is interesting to me though is that their DID models seem to show there is not any difference between the different groups for most of the periods of observation. IDK. I’d have to read more in-depth to make up my mind.

It seems like these conclusions, whatever you make of them should really only be applied to call-center work during the pandemic.

Meh, its not a perfect correlation (and the time series for the poverty map and the diabetes map are different), but most chronic diseases tend correlate with poverty pretty well. You should look at a map of obesity. It follows the same form.

That is really just a map of poverty.

That seems like good news for Mastodon