Does ChatGPT think?
– or my brief chat with #ChatGPT?
This ‘conversation’ took place a few days after version 4 was released. First, I gave #ChatGPT a simple task to continue 16, 32, 64, 128, … which it did impeccably:
Clearly it had recognized 2 power n. But does this require thinking? This sequence can be easily found on the Internet (Google “knows” it) and computers are very good at memorising. Will ChatGPT be able to solve a task which is less likely to be solvable by memorising? What if we ask it to?
Knowing the earlier sequence 16, 32, 64, …., it only takes to notice that 16 consists of digits 1, 6 and 32 of 3, 2,?to think of?how to continue this sequence. In fact, this does not take very much thinking, one could argue that this is still no more than simple pattern recognitions. But one could also argue that it does count as?thinking,?at least to some degree. It certainly isn’t very likely that one could solve is by having memorising something.
So, how does #ChatGPT do here??
Not very well. It not only failed to spot the commonalities between the multi-digit and single digit numbers that follow them, but also made a mistake in missing 32786, which it did figure out correctly earlier. But this can happen to a thinking human too. Let’s try again.?
Although the hint hasn’t helped #ChatGPT very much, interestingly, it did re-discover 32756. But it hasn’t shown much thinking. Can we try to nudge it more??
领英推荐
Not only ChatGPT makes a factual mistake, it still does not recognize what 32 contains both digits 3 and 2. But at least it is polite.?
Great! There is hope!
2048, 2, 8, 16 is where my patience comes to an end.?
ChatGPT reminds me of a bad student who does not understand anything, remembers something, and tries to bluff his/her/its way through an exam.?
But perhaps this is quite human? Whatever the answer to the question in the title of this blog is, to me this “discussion” with #ChatGPT shows that humans won’t be out of work just yet. Maybe later.?
(A version of this blog was presented at Coffee with EMBL on 28 April 2023)
Bioinformatics, blockchain and data science
1 年If you spell out the answer, it can do it. Perhaps that's not impressive... perhaps that's incredibly impressive...
Visual Communication of Science Training | Founder @ Seyens Ltd.
1 年"...to me this “discussion” with #ChatGPT shows that humans won’t be out of work just yet. Maybe later." Define later... https://www.bloomberg.com/news/articles/2023-05-01/ibm-to-pause-hiring-for-back-office-jobs-that-ai-could-kill#xj4y7vzkg
(MetabolicChemistry -> (Epi)Genetics -> ProteinBiochemistry) Circularized and Summarised in MSNGSMS
1 年Pretty sure that if we separate human beings into divergent thinking and all forms of IT (robotics, AI, informatics) into a 'service' role - then we'll have the perfect division of labour between man and machine. The core problem though is that a human being is powered by a couple of olives to think and remember ... ... the amount of infrastructure, energy and money required to drive a computer to think and remember (albeit potentially with a larger dataset) is where the technological solution falls down.