How Susceptible Are LLMs to Typos? (part 1)

This site does not allow itself to be embedded. You can still read it on the original site — the toolbar below keeps your place in the directory.

A recent theme in conversations at work is that “prompts are fragile.” Word choice and word order can have large impacts on LLM responses, and every user input is a potential attack vector for the LLM equivalent of a SQL injection attack. But before I go off on that tangent, having this repeated discussion got me thinking – “Can I quantify how sensitive LLMs are to…

Read on aimlbling-about.ninerealmlabs.com ↗