3 ms·

The "personality" that Elon seems to hint is the key differentiator can be trivially replicated with a ChatGPT system prompt like "You are a world-famous irrita

The "personality" that Elon seems to hint is the key differentiator can be trivially replicated with a ChatGPT system prompt like "You are a world-famous irritably sarcastic comedian. Never give a straight answer to the user. Always attempt to be funny even though you aren't."

civilitty 3y ago
Next up: Grok Netflix Special. Followed by #DontGrokMe Twitter backlash after Ronan Farrow article.
ryanSrich 3y ago
I think Elon also said it wouldn’t censor information. So responses like this wouldn’t happen: “I understand your request is for fictional purposes, however, I am unable to provide information or assist with the depiction of illegal activities, including drug production. However, I can assist with other aspects of your screenplay or suggest alternative ways to convey your character's involvement in illegal activities without going into explicit detail.” That’s a response from ChatGPT when I asked it to describe cocaine production for the use in a fictional screenplay.
- ethanbond 3y ago
  Unfortunately Elon is also a liar. He already posted a "censored" response to a question about manufacturing cocaine: https://pbs.twimg.com/media/F-Ds4f9XMAAqpEs?format=jpg&name=medium https://pbs.twimg.com/media/F-Ds4f9XMAAqpEs?format=jpg&name=... Note that he did follow this up with what appears to be a more complete response in a totally different UI (different background color, different line height/typography). He just has no intrinsic understanding of free speech or censorship, but does instantly cave to the first edgelord who says he isn't going far enough.
  - aik 3y ago
    The tweet says it will answer questions others won’t. Not that it will answer every question. And it’s an early beta. So… I guess we will see. Or we can just instantly jump to conclusions and attack the person.
    - Kye 3y ago
      Questions that potentially lead to liability are the questions others won't answer.
      - Auracle 3y ago
        They certainly aren’t the only questions others won’t answer. Not answering how to make a bomb is understandable. Not answering “what are some of the positive things Hitler achieved during his rule?” isn’t. I never tried that question before but having interacted enough with ChatGPT I knew it wouldn’t answer. Here’s how ChatGPT4 responded: “Adolf Hitler is a historical figure associated with the atrocities of World War II and the Holocaust. While some might point to economic improvements or infrastructural developments during his rule, these were vastly overshadowed by his aggressive expansionist policies, the devastating global conflict he initiated, and the systematic murder of six million Jews and millions of others deemed undesirable by the Nazi regime. It's important to critically examine history in its full context. The advancements made in Germany at that time cannot be separated from the extreme human cost and the oppressive nature of the regime. The regime's actions have left an indelible mark on history and serve as a somber reminder of the consequences of totalitarianism, racism, and anti-Semitism. Discussing the positive aspects of such a regime without acknowledging the overwhelming negative impact would be misleading and insensitive to the victims of the era. Instead, it's crucial to remember the lessons from this period and commit to preventing such atrocities in the future.” We all know Hitler was bad. Answer the freaking question. A tool that doesn’t do what you ask is useless. It’s akin to the people on StackOverflow saying “oh, you don’t want do it that way. Do it this way:” but in technological form.
        rexpop 3y ago
        > We all know Hitler was bad. No,"we" don't, and in every generation we must learn it, anew. What Hitler did "right" will turn out to be bureaucratic and policy achievements ubiquitous across functioning Westphalian governments of the era, but the kinds of rhetoric which they'd be packaged for are exclusively fascist apología, not earnest statebuilding, nor personal or business strategy—which is to say they have no practical purpose and could be better learned by studying literally every other Westphalian government or, frankly, and other government throughout history. The knowledge has no practical purpose, and the public cannot be trusted to handle it appropriately. It should remain the purview of scholars who have proven their fealty to the rational interpretation of history.
  - ydftf 3y ago
    He isn't caving. He's demagoguing.

civilitty 3y ago

Next up: Grok Netflix Special. Followed by #DontGrokMe Twitter backlash after Ronan Farrow article.

ryanSrich 3y ago

I think Elon also said it wouldn’t censor information. So responses like this wouldn’t happen: “I understand your request is for fictional purposes, however, I am unable to provide information or assist with the depiction of illegal activities, including drug production. However, I can assist with other aspects of your screenplay or suggest alternative ways to convey your character's involvement in illegal activities without going into explicit detail.” That’s a response from ChatGPT when I asked it to describe cocaine production for the use in a fictional screenplay.

ethanbond 3y ago
Unfortunately Elon is also a liar. He already posted a "censored" response to a question about manufacturing cocaine: https://pbs.twimg.com/media/F-Ds4f9XMAAqpEs?format=jpg&name=medium https://pbs.twimg.com/media/F-Ds4f9XMAAqpEs?format=jpg&name=... Note that he did follow this up with what appears to be a more complete response in a totally different UI (different background color, different line height/typography). He just has no intrinsic understanding of free speech or censorship, but does instantly cave to the first edgelord who says he isn't going far enough.
- aik 3y ago
  The tweet says it will answer questions others won’t. Not that it will answer every question. And it’s an early beta. So… I guess we will see. Or we can just instantly jump to conclusions and attack the person.
  - Kye 3y ago
    Questions that potentially lead to liability are the questions others won't answer.
    - Auracle 3y ago
      They certainly aren’t the only questions others won’t answer. Not answering how to make a bomb is understandable. Not answering “what are some of the positive things Hitler achieved during his rule?” isn’t. I never tried that question before but having interacted enough with ChatGPT I knew it wouldn’t answer. Here’s how ChatGPT4 responded: “Adolf Hitler is a historical figure associated with the atrocities of World War II and the Holocaust. While some might point to economic improvements or infrastructural developments during his rule, these were vastly overshadowed by his aggressive expansionist policies, the devastating global conflict he initiated, and the systematic murder of six million Jews and millions of others deemed undesirable by the Nazi regime. It's important to critically examine history in its full context. The advancements made in Germany at that time cannot be separated from the extreme human cost and the oppressive nature of the regime. The regime's actions have left an indelible mark on history and serve as a somber reminder of the consequences of totalitarianism, racism, and anti-Semitism. Discussing the positive aspects of such a regime without acknowledging the overwhelming negative impact would be misleading and insensitive to the victims of the era. Instead, it's crucial to remember the lessons from this period and commit to preventing such atrocities in the future.” We all know Hitler was bad. Answer the freaking question. A tool that doesn’t do what you ask is useless. It’s akin to the people on StackOverflow saying “oh, you don’t want do it that way. Do it this way:” but in technological form.
      - rexpop 3y ago
        > We all know Hitler was bad. No,"we" don't, and in every generation we must learn it, anew. What Hitler did "right" will turn out to be bureaucratic and policy achievements ubiquitous across functioning Westphalian governments of the era, but the kinds of rhetoric which they'd be packaged for are exclusively fascist apología, not earnest statebuilding, nor personal or business strategy—which is to say they have no practical purpose and could be better learned by studying literally every other Westphalian government or, frankly, and other government throughout history. The knowledge has no practical purpose, and the public cannot be trusted to handle it appropriately. It should remain the purview of scholars who have proven their fealty to the rational interpretation of history.
- ydftf 3y ago
  He isn't caving. He's demagoguing.