The Future of Everything is Lies, I Guess

A critical, long-form essay by a software engineer examining the current state of LLMs and ML systems. It covers what LLMs actually are (statistical text completers, not reasoning machines), their persistent tendency to confabulate and lie, the 'jagged frontier' of their capabilities (excelling at some complex tasks while failing trivially at others), and the uncertainty around whether scaling will continue to yield improvements. The author draws on personal anecdotes of LLM failures across image editing, factual queries, and agentic tasks, while acknowledging genuine advances in areas like protein folding and code generation. The piece is explicitly one-sided, aiming to fill in the critical gaps left by AI boosterism.