/biz/ - Business & Finance

>>49377486
The code base was already fucked when I got there because it was as old as me. When I was working there my job was basically to be an archeologist and find out how something was programmed 20 years ago and how to get it to work in modern .net.

No one even knew where the servers this shit was running on, let alone who was responsible for maintaining these things (literally companies that don't even exist anymore in some cases). I had to write a custom encryption/decryption pipeline based off of ancient code full of comments like "//needs fix" and "//ask bob about date because it doesn't work" that was based around just scrambling timestamps around because there was no standard https at the time it was set up.

For example once I found out that the way this thing was designed 15 years ago it would stop working on every day the day and month were the same day (january 1st, february 2nd, etc) , at a specific time.

I also never even told my boss about it because my team got let go so screw that. This company manages 20 billion dollars in retirement assets btw and the shit I was untangling was handling actual people's accounts and there were dozens of repos that could be dependent on it (it's not like you even know because the whole thing was a tower of babel). There is a huge amount of code debt out there that is yet to be repaid.

>>23393886
Econometrics 100%. That is entire degree in some places Just pray you don't get IQ filtered.

>>22721062
Idk what Carmack knows about machine learning, but the reason FP8 is being looked at at all in the ML world is because of the success of curves like ReLU.

>>22513282
Of course it does.

Imagine a disease that only 1% of people have. A test with a 90% false positive rate would mean that about 9/10 people who test positive don't actually have that disease, because 10% of the healthy population would get false positive, which is about 10 times more as the population that actually has it.

>>15880851
There is a reason why multi-z level tile based games are so infrequent by the way. It's mostly just do to physical constraints.
Imagine 400*400*100 = 16,000,000 tiles/entities. At one byte per tile that would be 15 MB already.
Now consider what happens when a single object reference in c# is 12 bytes minimum and a single entity will need dozens of references to represent health, color, sprite image, lighting, whatever and that's hundreds of bytes that simply will not fit in memory.

I created a template system that allows for entities to use up a minimum of about 25 bytes total in memory (that includes their reference) if they aren't edited, and if they are edited once then about 70 plus however many bytes that were changed on the object.

So you can imagine a world full of various kinds of tiles that could even be modded in and each one would only take up 25 bytes until it is damaged from mining for example and that would make one about 80 bytes in size.

This is for all entities. A tile entity could be given a component that lets it damage people who pass by, emit radiation and temperature, do anything and STILL stay at 25 bytes.
Entities are represented as json templates with components that represent a subscription to a system. You can mod your own custom components as well.