Isn't that "scheduled sampling"? In that case they also shift the input distribution toward that of the model, which possibly is even more crucial than shifting the output distribution?
You probably mean "e.g." as "for example", not "i.e."?
This might be on purpose and part of the training data because "for example" just sounds much better than "e.g.". Presumably for most purposes, linguistic naturalness is more important than fidelity.
Except for (vector) graphics design, desktop publishing, page layout designing and reporting software...
The reason is of course that most printed items are unica and non-variable, but
proper parametric and constraint-based would definitely be useful in report generation and signage where print formats vary or where designs need to be automatically adjusted based on texts and other content varying in size/length.
The "lesson" does not seem so clear according to the article you have linked:
> Schneider claimed that the decision had been political, “not made on the basis of facts”
> “We do not see any compelling technical reasons for a change to Windows and Microsoft Office… We solve compatibility and interoperability problems by providing MS Office, mostly virtualised, at workplaces that need to work together with external offices on office documents.”
One point of critique was the difficulty involved with 3rd party programs on the Linux laptops. But with MS Teams, Zoom etc. running in the browser and/or having actual Linux clients these days, this may be much more feasible today.
The opposition claimed the decision to abandon Linux was a purely political one. The mayor who pushed the return to Microsoft, Dieter Reiter (SPD), was a known Microsoft fanboy, but he (of course) denied personal investment in the reversal.
As others have noted, it is more about the maximum complexity increasing than mean or median. Simple structures keep existing as long as they have their niche, and a human's niche is not (yet) that of viruses.
This also reminds of Gall's law that complex systems evolve from simpler ones.
You can also see it in neural nets, where larger ones have a higher spatiotemporal resolution and can do more complex things.
More model capacity allows to model the environment and self more accurately which allows to outperform other structures in negentropy consumption often at the cost of the other structures (zero sum).
This exerts selection pressure toward increasing complexity.
That also largely explains group and country disparities.
I am not sure that non-evolving things really fit into the same pattern. A burning fire does not necessarily displace inert matter, nor did it arise from competition.
Physics and chemistry are more fractal-like possibly the result of enumeration of all computational structures (see Tegmark's mathematical universe hypothesis or Wolfram's ideas on the computational universe). Not fractal-like in terms of self-similarity (although there is some at different scales), but fractal-like in terms of chaotic complexity like a pseudorandom number generator but with more rule-like structures in between. Wolfram also classified such computational patterns.
To this list I'd add pulling the jaw backward/inward.
Moving the jaw forward and then to the right has the biggest effect for me, causing the ringing on the left ear to increase. It's asymmetric in that moving the jaw to the front left has only a very small effect on the right ear.
Since when are military spooks and political opportunists better at deciding on our technological future than startups and corporations? The degree of global policing and surveillance necessary to fully prevent secret labs from working on AI would be mind-boggling. How would you ensure all government actors are sticking to the same safety standards rather than seizing power by implementing AI hastily? This problem has long been known as quis custodiet ipsos custodes - "who guards the guards themselves?".
Even though p-values can be hacked, they are very useful when they aren't. At p = 0.1 I'd ignore the finding because there would be a 10% chance it was explained by random chance. p = 0.01 would pique my interest. p < 0.001 I'd accept it as true, but I'd still watch out for systematic biases such as comparing new to old factories.
Of course each emergency is one too much, but I wonder what are the sample sizes? Are the counts statistically significantly different from each other? Another thing to consider: Emergencies are rare events and so a small difference in circumstances can make the outcomes vary substantially. Is a well oiled manufacturing pipeline like a Audi factory comparable to a new factory that has not rounded all sharp corners yet?
In this study recycled glass bottles were not better when it comes to microplastics. It's everywhere. In lids, detergents etc. In fact, single-use plastic bottles did best!