4 ms·
GPT4 shows itself quite capable at generating code and making complex refactorings in the face of changing requirements. When you stray from areas of code, it
by kazinator 4y ago
GPT4 shows itself quite capable at generating code and making complex refactorings in the face of changing requirements.
When you stray from areas of code, it shows serious weaknesses.
Within still the same domain, if you discuss algorithms, it's a little bit out to sea without a rudder. If you say subtly wrong things in your questions, it picks up on those things and pretends it researched them.
Earlier today, it spewed that the Lomuto partitioning scheme used in some quicksort implementations requires one bit of external storage per array element. This is completely false, and comes from a remark I made in an earlier question in the same chat about some possible idea requiring one bit of storage.
Like a parrot, it just took my idea, stripped it of context, and planted it into statements about something else.
Speaking of which, I've noticed a pattern in its behavior of using my ideas and hints, and pretending that it came up with them itself. If confronted about that, it will admit, that yes, sorry it did get that answer from your hint and not from its knowledge and analysis as its answer glibly implies.
GPT4 is intensively being trained on coding in order to impress programmers, who will then evangelize it to non-programmers. (The things it does with code really are impressive.)