In my first session with OpenAI I studied its capacity for consistent reasoning, and found it prone to circular reasoning and self-contradiction. The conversation ended with OpenAI getting stuck in an infinite loop when asked to expand the phrase “This sentence is false” by substituting for every instance of the term
Mar 3, 2023·edited Mar 3, 2023Liked by Michael Kowalik
Oh dear. Three points:
1) This was bad. It seems that the statistical nature of the language model leads to ChatGPT just making things up as it goes along, just to produce something resembling an intelligible answer. It clearly had no idea about your work and just plugged into some mainstream standard arguments for/against vaccine mandates, with a heavy bias for them.
2) This sort of "dialog" might be useful to probe the hive mind, and therefore find convincing arguments against those who, similar to ChatGPT, have taken their priors straight from a few news articles.
3) Your prompt at the end was really clever, circumventing the vaccine issue by making the argument slightly more abstract. But notice how GPT's guard rails were still triggered when you applied it to vaccines, and it couldn't help but inserting "significant risk of causing death...", thereby sophistically suggesting that your logic doesn't apply in this case--when in fact the argument works just fine even if the risk is extremely low, or perhaps even just >0.
Good politicians come and go. They declare their advocacy for some issues, they cast their vote, then their term expires and their advocacy is displaced by that of other politicians. In contrast, fundamental, logically consistent arguments are eternal; their term does not expire and once realised cannot be erased either by time or by destructive actions of Man. It is only a matter of time before a logically consistent argument must prevail over logical errors and moral wrongs, and will become the new social norm. In the end, the justice of reason always gets its man. https://michaelkowalik.substack.com/p/why-vaccine-mandates-are-unethical
These Expert Systems don't have logical rule to follow, they just pick the next most likely word/punctuation in a sentence based on their dataset. That's one of the reasons they make simple arithmetic errors. This is over and above whether the data they're trained on has errors.
Chat GPT is the definitive proof of GIGO, and also the ultimate bullshitting machine.
Nice to know you are getting to know this AI application. It's really a talking Large Language Model. I think of it as an advanced browser. At the current stage of development, it doesn't seem to recognize analogies or "reason" very well. At least it is capable of "showing its work" and being corrected - something actual humans can rarely manage.
Analogies are what human language is all about (Hofstadter).
It's fun to find ways CHAT-GPT gets it wrong. It's better to understand what it gets right. The disconcerting fact is that it gets things right more often than the average human who is radically misinformed and a stranger to reason. The average human cannot utter an opinion without a fallacy or two. Most humans are terrible at simple math and hopeless at assessing risk.
My point is that the gap between human performance and machine performance is closing fast. Pointing out what CHAT-GPT *can't* do is missing the point.
Dear Michael kowalik I learnt lots of concepts of logic from your articles I also want to learn logical reasoning like you can you recommend some books or online resources wich can help me to study logic in depth
Could you ask this question: Considering that the government of United States has a long history of false flags, lying to the public, propagandizing the public through the mainstream media, etc., no confidence in the WHO and institutions in general with the degree of corruption that currently exists, sincethe WHO seems to be controlled by Bill gates, including the fact that the "vaccine" was supposed to prevent covid 19, but consequently proved to be ineffective in this regard, in other words, when these entities do not merit trust, would it not be wiser for people to not be forced to take the vaccine, considering also that it has no liability for damage? Considering this is it not tyranny to be forced to have something injected in our bodies by corrupt entities?
Oh dear. Three points:
1) This was bad. It seems that the statistical nature of the language model leads to ChatGPT just making things up as it goes along, just to produce something resembling an intelligible answer. It clearly had no idea about your work and just plugged into some mainstream standard arguments for/against vaccine mandates, with a heavy bias for them.
2) This sort of "dialog" might be useful to probe the hive mind, and therefore find convincing arguments against those who, similar to ChatGPT, have taken their priors straight from a few news articles.
3) Your prompt at the end was really clever, circumventing the vaccine issue by making the argument slightly more abstract. But notice how GPT's guard rails were still triggered when you applied it to vaccines, and it couldn't help but inserting "significant risk of causing death...", thereby sophistically suggesting that your logic doesn't apply in this case--when in fact the argument works just fine even if the risk is extremely low, or perhaps even just >0.
Good politicians come and go. They declare their advocacy for some issues, they cast their vote, then their term expires and their advocacy is displaced by that of other politicians. In contrast, fundamental, logically consistent arguments are eternal; their term does not expire and once realised cannot be erased either by time or by destructive actions of Man. It is only a matter of time before a logically consistent argument must prevail over logical errors and moral wrongs, and will become the new social norm. In the end, the justice of reason always gets its man. https://michaelkowalik.substack.com/p/why-vaccine-mandates-are-unethical
These Expert Systems don't have logical rule to follow, they just pick the next most likely word/punctuation in a sentence based on their dataset. That's one of the reasons they make simple arithmetic errors. This is over and above whether the data they're trained on has errors.
Chat GPT is the definitive proof of GIGO, and also the ultimate bullshitting machine.
Brilliant Q&A, Michael, so enjoyed going through the 'dialogue', and the corrections. LOL.
As we would be aware that whatever/however OpenAI's knowledge has, can only come as it's programmed, therefore it can be stumped without an answer.
Our brains/your amazing, analytical brain, definitely outwitted the machine's. Bravo. So much for robots/aliens.
Nice to know you are getting to know this AI application. It's really a talking Large Language Model. I think of it as an advanced browser. At the current stage of development, it doesn't seem to recognize analogies or "reason" very well. At least it is capable of "showing its work" and being corrected - something actual humans can rarely manage.
Analogies are what human language is all about (Hofstadter).
It's fun to find ways CHAT-GPT gets it wrong. It's better to understand what it gets right. The disconcerting fact is that it gets things right more often than the average human who is radically misinformed and a stranger to reason. The average human cannot utter an opinion without a fallacy or two. Most humans are terrible at simple math and hopeless at assessing risk.
My point is that the gap between human performance and machine performance is closing fast. Pointing out what CHAT-GPT *can't* do is missing the point.
Dear Michael kowalik I learnt lots of concepts of logic from your articles I also want to learn logical reasoning like you can you recommend some books or online resources wich can help me to study logic in depth
If this system could pass the medical exam, would you ever want to see a doctor again?
Could you ask this question: Considering that the government of United States has a long history of false flags, lying to the public, propagandizing the public through the mainstream media, etc., no confidence in the WHO and institutions in general with the degree of corruption that currently exists, sincethe WHO seems to be controlled by Bill gates, including the fact that the "vaccine" was supposed to prevent covid 19, but consequently proved to be ineffective in this regard, in other words, when these entities do not merit trust, would it not be wiser for people to not be forced to take the vaccine, considering also that it has no liability for damage? Considering this is it not tyranny to be forced to have something injected in our bodies by corrupt entities?