Copilot does a much better job, but it also outputs irrelevant stuff.
I also miss FastGPT-style source referencing. In case of FastGPT the relatively low number of references (/sentence) may be a good indicator of answer quality.
Experimenting with #FastGPT#LLM to help me find the right API's to use in my #Ghidra script. While the answer puts me in the right direction, it includes misinformation and code that won't compile (solution in alt text).
"China’s largest #antivirus firm, Qihoo360, is an investor of offensive capabilities firms and may be selling PII of individual antivirus customers to an offensive company it funds that does intelligence work for government clients."