Further on Current Al Chatbots: What They're Good At, and What They're Not
See my previous piece: “Claude 3 vs ChatGPT4: A Comparative Review” (March 17, 2024). And compare Ethan Mollick's latest post, excellent: “I, Cyborg: Using Co-Intelligence” (March 14, 2024).
The items in this blogpost are all based on my personal experience, using ChatGPT4 and other AI chatbot tools fairly intensively, in a wide range of use cases, both work and personal.
Good at
Rewriting a draft, to correct typos, bad grammar, and improve bad style. For writers with above average style, won’t improve style, as Mollick points out. As it regresses towards a bland, average mean.
Translating. Esp. ChatGPT4. For standard dialect, about the same as Google Translate. But for non-standard writing, better than Google Translate, since it can be better understood based on context.
Suggesting titles, based on a draft.
Coding. I’ve used it for Python (for web scraping and word counts, see here); SQL (I’ve used it for Google’s Big Query); Excel formulas (writing them, based on natural language, and parsing and explaining complex ones); regex (in Google Sheets, see here); Javascript (used it for Google Apps Script, see here)
Converting a table in an image to a text-based table, using OCR. Claude is especially good at this. Windows Snipping Tool is very good at OCR in general, see here.
Tagging images with attributes, and describing its content in general. ChatGPT4 is especially good at this.
Generating creative text, in all languages. See here. Claude is especially good at this, as I mentioned in the previous post.
Bad at
I haven’t had success with getting ChatGPT to write in direct style, with few adjectives.
Reading Hebrew handwriting.
Critical analysis, such as analysis of historical texts, debates, and any kind of fact checking.
ChatGPT’s internet browsing is a very weak feature. It essentially just summarizes a page or two from top links in search results. It used to be better, but was neutered.