The field moves faster than any one person can read, and that is normal.
arXiv publishes preprints daily, long before peer review has finished.
Follow a few careful summarisers rather than trying to read everything yourself.
Conference proceedings are a filtered, higher-signal view of the year.
Depth on the foundations ages far better than breadth on the news.
More machine learning papers are published each month than anyone can read in a year.
Accepting that is the first practical step, because completeness is simply not available.
arXiv hosts preprints, which appear the moment the authors post them.
That is why the field moves quickly, and also why the quality varies so widely.
Peer review still happens afterwards at conferences, often months later.
Proceedings from NeurIPS, ICML, ICLR and ACL are a filtered view of the year.
Reading the best paper awards from each is a cheap way to see what the field valued.
Between those, rely on a small number of people who summarise carefully.
Newsletters, paper-reading groups and a few careful writers do the filtering for you.
Choose them for judgement rather than for speed, since being first is worth very little.
Set a fixed and small budget, such as one paper a week read properly.
One paper understood is worth more than thirty abstracts skimmed.
Track a few topics deliberately rather than trying to follow the field as a whole.
Foundations move slowly, and time spent on them keeps paying out for years.
Most of what feels urgent this month will not matter, and the basics still will.
One question on this card. Every option explains itself once you pick, right or wrong.
11 concepts in this module, then a graded test with every answer explained.