Could you elaborate a bit more on how you do this, or point to resources on it? Do you mean putting constraints on possible parameter values, or just on the loss function values itself? Thanks.
Of interest, Winston Chuchill took up bricklaying while performing many other duties, became a member of his local bricklaying union, and build a bomb shelter in his back garden.
I think many experiences do compound. Most people who take the traditional path of working straight after school trade security for reference experiences which help in understanding what they want, and helping deal with challenges further down the line. For some this is fine, security is what they want and the 'what ifs' and opportunity cost on growth experiences will not bother them. For others this may cause more existential problems down the line.
As a generalising anecdote, some of the greatest people I have worked with had very non-traditional paths early on, and I find them to have a greater 'big picture' sense, more resiliency and more humour in the face of adversity. But are not as rich in their 40s as others.
I'm surpised this is getting downvoted. The OPs assertion that listening is easier than speaking makes sense in the context he describes, a closed testing environment.
In a more practical setting, as someone who is living in another country and learning the language, it is far easier for me to convey my meaning with limited vocabulary, than to listen and interpret another speaker, and especially a group conversation.
Certainly listening is easier than me speaking at the same proficiency as the other speaker, but in terms of what matters for using the language on a day to day basis, it is the other way around.
I have some sympathy to this argument, even though I agree with you in overall that based on what we know from the millions of vaccines being administered, they appear to be very safe. But points in favor of the FDA argument, which is mostly around emergency authorization requiring a lower standard of evidence:
- The pharmaceutical industry is notorious for manipulation of trial data to make their treatments looks as beneficial as possible, as they are incentivized to do so. In the current case Pfizer and others have still not made the trial data public, and probably never will [0][1]
- The main study on the Pfizer vaccine contained safety data for an average of about two months - on this basis the vaccine was declared to be extremely safe. Obviously longer term effects cannot be captured in this timeframe [2]
- Due to the requirements to generate a clear public health message, almost anything that may suggest a risk with vaccines is downplayed. See for example how VAERS data is routinely dismissed (with some reason)[3], whereas data on COVID hospitalization and death rates is largely unquestioned, despite containing some room for interpretation [4].
Again, the the totality of evidence suggests that the vaccine are safe and effective. But it is not crazy or evil, as some would suggest here, to question some of what we are being told and how the data is being presented to us.