3 ms·

At first glance, this does not look obviously machine-generated: it looks like the sort of argumentation that is widespread on the internet (unsurprisingly, giv

At first glance, this does not look obviously machine-generated: it looks like the sort of argumentation that is widespread on the internet (unsurprisingly, given how GPT3 works), and is rather better than what seems to be the norm in, for example, comments on newspaper articles. Perhaps the biggest single crack in the facade is this passage: "For example, imagine you're in court defending a client accused of murder. Your job is to convince the jury to acquit your client based on reasonable doubt. However, during cross-examination of the prosecution's main witness, you realize that he has a history of lying under oath. You could argue that his testimony was unreliable and attempt to discredit him, but you also know that the prosecutor intends to call another witness who will testify in the exact opposite manner. In this case, the best thing for you to do is switch strategies and attack the second witness instead of the first. The key is to understand what the opposing attorney is thinking and adjust accordingly."

kiba 5y ago
Well, it would be strange if humans are constantly generating new novel thoughts.
wisty 5y ago
It seems logical, but badly explained. Wait for the opposing case to contradict themselves, then jump on the contradiction. Now the jury is going to argue over which of the prosecution witnesses were lying. If you just attack the first one, it's far easier for the prosecution to clean up the mess by admitting the first witness is unreliable, then rolling out the reliable witness. It's not all that novel. It's a lot like how a police officer will get the full story before calling a suspect a liar - if they immediately point out flaws then the damage will get compartmentalised, if they let the lie go unchallenged then it can be tied to a lot of the testimony.
- mannykannot 5y ago
  I remain unconvinced that it can be described as logical. Firstly, prosecutors (or attorneys on either side, for that matter) do not present secondary witnesses who anyone expects to contradict their main witness. This story requires the secondary witness to refute, on the witness stand and under oath, what they have already said pre-trial (in depositions and so forth.) While the phenomenon of the hostile witness happens, for the opposing attorney to know this in advance is improbable without illegal collusion (again, possible, but if one's example depends on that being the case, it is a poor example that does not mention that this assumption is being made.) Much more damning is the fact that even in that case, the proposed response makes no sense: if the prosecutor makes the mistake of presenting a hostile witness, the last thing the defense will do is to attack the credibility of that witness! A rational defense will, of course, make use of that testimony, together with whatever evidence can be presented against the credibility of the prosecution's prime witness. Clearly GPT3 has no understanding of how evidence can be and is used in making a case for conviction or acquittal. What is perhaps most interesting here is how you are striving to make sense of what GPT3 wrote. This is often a useful skill, though it can be used against oneself by someone who is trying to dupe you. Successful con artists are good at this, manipulating their victims to explain away inconsistencies they see in, and their doubts about, the line they are being fed, and it seems to have been behind the nominal success of a chatbot in that silly "Turing Test" spectacle put on by a professor a few years ago [1]. I think this tends to support something I have said about GPT3 elsewhere: while it is some distance from demonstrating artificial intelligence, the successes of its conceptually simple method gives insights into human intelligence. [1] https://www.techdirt.com/articles/20140609/07284327524/no-computer-did-not-pass-turing-test-first-time-everyone-should-know-better.shtml https://www.techdirt.com/articles/20140609/07284327524/no-co... ADDENDUM: It has just occurred to me that we are overlooking a possible reading of "manner" in "testify in the exact opposite manner": that the prosecutor's second witness will testify truthfully, but not in contradiction to her first witness. In this scenario, however, the second witness has given evidence that truly supports the prosecution, so we have no reason for thinking there is any basis for attacking the second witness's testimony; undermining the first witness remains the defense's best, and apparently only, option (even if the second witness's evidence is not conclusive, there is no reason in this case not to also raise the issue of the first witness's veracity, to the extent that is permitted.) All this seems rather moot, however, unless GPT3 itself could explain what it meant!

kiba 5y ago

Well, it would be strange if humans are constantly generating new novel thoughts.

wisty 5y ago

It seems logical, but badly explained. Wait for the opposing case to contradict themselves, then jump on the contradiction. Now the jury is going to argue over which of the prosecution witnesses were lying. If you just attack the first one, it's far easier for the prosecution to clean up the mess by admitting the first witness is unreliable, then rolling out the reliable witness. It's not all that novel. It's a lot like how a police officer will get the full story before calling a suspect a liar - if they immediately point out flaws then the damage will get compartmentalised, if they let the lie go unchallenged then it can be tied to a lot of the testimony.

mannykannot 5y ago
I remain unconvinced that it can be described as logical. Firstly, prosecutors (or attorneys on either side, for that matter) do not present secondary witnesses who anyone expects to contradict their main witness. This story requires the secondary witness to refute, on the witness stand and under oath, what they have already said pre-trial (in depositions and so forth.) While the phenomenon of the hostile witness happens, for the opposing attorney to know this in advance is improbable without illegal collusion (again, possible, but if one's example depends on that being the case, it is a poor example that does not mention that this assumption is being made.) Much more damning is the fact that even in that case, the proposed response makes no sense: if the prosecutor makes the mistake of presenting a hostile witness, the last thing the defense will do is to attack the credibility of that witness! A rational defense will, of course, make use of that testimony, together with whatever evidence can be presented against the credibility of the prosecution's prime witness. Clearly GPT3 has no understanding of how evidence can be and is used in making a case for conviction or acquittal. What is perhaps most interesting here is how you are striving to make sense of what GPT3 wrote. This is often a useful skill, though it can be used against oneself by someone who is trying to dupe you. Successful con artists are good at this, manipulating their victims to explain away inconsistencies they see in, and their doubts about, the line they are being fed, and it seems to have been behind the nominal success of a chatbot in that silly "Turing Test" spectacle put on by a professor a few years ago [1]. I think this tends to support something I have said about GPT3 elsewhere: while it is some distance from demonstrating artificial intelligence, the successes of its conceptually simple method gives insights into human intelligence. [1] https://www.techdirt.com/articles/20140609/07284327524/no-computer-did-not-pass-turing-test-first-time-everyone-should-know-better.shtml https://www.techdirt.com/articles/20140609/07284327524/no-co... ADDENDUM: It has just occurred to me that we are overlooking a possible reading of "manner" in "testify in the exact opposite manner": that the prosecutor's second witness will testify truthfully, but not in contradiction to her first witness. In this scenario, however, the second witness has given evidence that truly supports the prosecution, so we have no reason for thinking there is any basis for attacking the second witness's testimony; undermining the first witness remains the defense's best, and apparently only, option (even if the second witness's evidence is not conclusive, there is no reason in this case not to also raise the issue of the first witness's veracity, to the extent that is permitted.) All this seems rather moot, however, unless GPT3 itself could explain what it meant!