Twitxy Saved

Ryan Fleury

Verified account This account is verified. Learn more

@rfleury

RJF // @dgtlgrove // Making RAD Debugger at @radgametools in @epicgames (opinions my own)
USA dgtlgrove.com Joined November 2016
151 Following 26K Followers 13K Posts

Latest posts 5

Ryan Fleury Verified account @rfleury
It is "midwittery" to restate ~research consensus (or at least one of several popular views) about obvious limitations of present-day LLM-based AI architecture w.r.t. rigorously solving for constraints (which they do not), which persist to this day, and that they are not arbitrarily generalized intelligence.

Non-thinking and psychotic
6 replies 2 reposts 140 likes 4.7K views View on X
Ryan Fleury Verified account @rfleury
Working on the new evaluation compiler for RADDBG. This will clean up MANY rough edges in the current evaluation system. It is much more rigorous about parsing C type expressions correctly, it is language-frontend agnostic, it is much tighter w.r.t. evaluation of debugger metadata, it will support cross-evaluation-context evaluations in a single expression (e.g. thread_0.locals.foo - thread_1.locals.foo), and it is much smarter about caching, so it should be a lot faster too.