Reposted by Josh
Overnight I trained a Jev-like model to output probability distributions over a set of 40 emoji, which runs in RAM at ~150ms per query.
This was super easy and cost ~$10. It works pretty well on a weak training set, and with a day or two of work I could make this sing.
![Marc Brooker is a Distinguished Engineer at AWS, works on AWS S3, and is known for having high standards in software engineering. He has written this about the future of code reviews (emphasis mine):
“There’s been a ton of talk about the role of humans in code review, and when and how humans should be signing off on changes.
I believe that, long-term, humans have no role in routinely reviewing code.
Short-term, many teams do it for good reasons. Quality. Understanding. Compliance. Those reasons exist, but we should be (and are) working on ways to make them go away.
The “mixed mode” where humans look at some code reviews, assisted by review tools, is also valuable today. But probably even more transient.
The idea that humans will reliably look through code to find the increasingly rare issues that automated tools miss seems like a fantasy.
What does the future [without code reviews] look like?
Review tools, powered by LLMaaJ-style techniques, static analysis, and automated reasoning methods. Principled approaches to testing and validation (PBT, DST, model-guided fuzzing, etc). Correct-by-construction techniques.
The combination of these things will lead to (and in many cases already lead to) better outcomes than traditional code review, both for large and small issues. Soon, much better outcomes than code review.
The days of Code review as a tool for increasing software quality appeared numbered, and the end may already be here (but unevenly distributed).](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ctphcgyhnllfluywokibtrgb/bafkreiexqaz44a6lhqkf4tdor7sze6lcr3usfktajowjlczfmfpob6wsg4)
