Steven Pinker – Will ChatGPT supplant us as writers, thinkers? | Hacker News Reader

Steven Pinker – Will ChatGPT supplant us as writers, thinkers?

news.harvard.edu

5 pointswhy-el4 comments

Great interview. Pinker's responses are well thought-out and accurate. He admits not expecting LLMs to get this good at so many things so quickly, illustrates their failure modes with simple examples than anyone can grasp, and expects these models to get significantly better (i.e., make fewer mistakes) over time.

I'm not doing justice to the interview with this short comment.

Go read it!

I told it to use common sense to one of the questions he asked and it answered fine

using common sense answer the following logic problem:

if cindy is alive at noon and alive at 5 pm is she alive at 3 pm?

Based on common sense, if Cindy is alive at noon and alive at 5 pm, it's reasonable to assume that she is alive at 3 pm. This is because there is no information provided that suggests she would have died or become incapacitated between those two times. It's possible that she could have experienced an event that would cause her death or incapacity during the three-hour period, but based solely on the information given in the problem, it's safe to assume that she is still alive at 3 pm.

I probed it a little to understand it's logic without "common sense"(you can decide for your self which is the better answer):

ME: list the facts in this statement: if cindy is alive at noon and alive at 5 pm is she alive at 3 pm?

ChatGPT: Fact 1: Cindy was alive at noon. Fact 2: Cindy was alive at 5 pm. Conclusion: It is unknown whether Cindy was alive at 3 pm.

ME:can someone die and come back to life?

ChatGPT:

It depends on how you define "death" and "coming back to life."

In some medical situations, a person's heart may stop beating, and they may not be breathing, which can lead to a clinical death. However, with prompt medical attention, it is sometimes possible to revive the person and restore their heartbeat and breathing. In this case, one could say the person "came back to life," but it is important to note that they were not truly dead in the sense that all brain function had ceased.

On the other hand, if we define "death" as the irreversible cessation of all brain function, then it is currently not possible for someone to come back to life once they have died. While there are many stories and legends of people being brought back from the dead, there is no scientific evidence to support the idea that someone can die and then be resurrected.

ME: in this statement: "if cindy is alive at noon and alive at 5 pm is she alive at 3 pm?" is which definition is the statement using?

ChatGPT: The statement "if Cindy is alive at noon and alive at 5 pm, is she alive at 3 pm?" does not specify a definition of death or life.

However, the question implies that Cindy's state of being alive or dead is continuous and can change over time. Therefore, it is most likely using a clinical definition of death, where a person can be declared dead or alive based on observable criteria such as the presence or absence of a heartbeat or breathing. It is important to note that this statement is not discussing the question of irreversible cessation of all brain function, which is the criterion for biological death.

It's also conceivable that companies running LLMs might consider various tweaks to popular questions as they appear in media so as to inflate the model's answers. No way to learn for sure, but competing companies where PR spells short term disasters, wouldn't put it past them to intervene when necessary. I am not familiar with how one augments and already trained model or if if there is even an architecture that's capable of that, short of input proxies in front of the model. I am curious to learn more if anyone knows.

Reply on news.ycombinator.com