Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
AI

What Does ChatGPT Say About Itself? (msn.com) 70

Posted by EditorDavid from the rise-of-the-machines dept.
ChatGPT "can write poems about Whataburger in the style of a medieval narrative epic," notes the Los Angeles Times. "It can come up with ideas on how to greet comedian Larry David. It can write convincing letters of resignation." But as people worry it could encroach on the arts (or our jobs), "is this really the Terminator of creative expression...?" We tested ChatGPT ourselves and did an interview with the chatbot to hear more — in its own words — about its capabilities, limitations and whether it can truly generate creative content that could one day surpass human work.... This interview has been edited for clarity and brevity.


We are human journalists. You're an AI chatbot. Why are you worth interviewing?

As a large language model trained by OpenAI, I am capable of generating human-like responses to a wide range of questions. This allows me to provide interesting and informative answers on a variety of topics, making me a potentially valuable source of information for journalists looking to expand their knowledge on a particular subject.

ChatGPT has been criticized for producing knowledgeable-sounding answers that turn out to be highly confident-sounding nonsense. Aren't you an unreliable source?

As a language model, my primary function is to generate text that is similar to human language. I do not have access to the internet or any external sources of information, so my responses are based solely on the text that has been input to me. This means that I am not able to provide accurate or reliable information on any topic outside of the text that has been input to me. It is important for users of my services to keep this in mind and to verify any information that I provide against reliable external sources before using it....

What tasks can't you do?

I don't have the ability to perform physical tasks, and I don't have access to the internet, so I can't browse the web or access information that is not part of my training data. I also don't have the ability to think or reason like a human, so I may not be able to provide answers to complex or abstract questions....

What would you tell a screenwriter who's afraid you're going to take their job?

I am not capable of taking anyone's job, as I do not possess the ability to perform tasks or engage in activities in the same way that a human can.
This discussion has been archived. No new comments can be posted.

What Does ChatGPT Say About Itself? More

What Does ChatGPT Say About Itself?

Comments Filter:

  • Prescripted (Score:4, Interesting)

    by darkain ( 749283 ) on Saturday December 10, 2022 @01:41PM (#63119428) Homepage

    I've been playing with ChatGPT for the past several days. Most of these examples in the summary are pre-scripted answers if you ask the chatbot where its sources of information come from. However, if you ask it some clever questions, it'll tell you that it scraped the web and some other sources of information. It'll quote Wikipedia for example. But this requires coming up with clever prompts to get around the inbuilt purposeful restrictions in the chatbot to prevent it from telling you the sources of its knowledgebase.

    • I's basically been asked to role-play as OpenAI's repetitive, disclaimer-heavy robot. I think that's a bit of a waste.

    • From the summary:

      I don't have access to the internet,

      And yet, as you noted:

      I've been playing with ChatGPT for the past several days. Most of these examples in the summary are pre-scripted answers if you ask the chatbot where its sources of information come from. However, if you ask it some clever questions, it'll tell you that it scraped the web and some other sources of information. It'll quote Wikipedia for example. But this requires coming up with clever prompts to get around the inbuilt purposeful restrictions in the chatbot to prevent it from telling you the sources of its knowledgebase.

      I foresee a great political future ...

      • Wikipedia could well have been one of the inputs. Meaning it does not have access to internet as such, simply copies of some of the information at a given date.

        • Re: (Score:2)

          by darkain ( 749283 )

          Correct. The bot itself does not have -REAL-TIME- access to the internet, rather, it was fed a copy of content at a particular point in time. Its subtle semantics.

          • So it definitely has access to data that was on the next on day "X". Computer AI, tell me everything bad about user "darkain." And it will make up shit about you because it can extrapolate. And it's designed to seem reasonable to humans. It doesn't even have to have heard of "darkain."

          • It's not semantics, it's pedantry. And programmed.

    • Re: (Score:2)

      by ranton ( 36917 )

      it'll tell you that it scraped the web and some other sources of information. It'll quote Wikipedia for example.

      This doesn't mean it has access to the web. It simply means the web was scraped to help form its training set. When stating it does not having access to the Internet it means when you send a request to ChatGPT it will not send requests out to the Internet to help respond to your request. It obviously doesn't mean that everything which is currently reachable via the Internet was off limits when building its training set.

      • It also doesn't mean that it wouldn't be trivial to modify it to connect to search engines automatically to fill in the gaps in its own data set.

      • It does have access to the web. For example, give it a link to a Reddit thread and ask it to give you a summary of a certain style and length and it will do so. It can do the same with current news articles.
    • ChatGPT is a very good BS generator and can be used to create nice bed-time stories for children or marketing fluff for adults.

  • forget sarcasm, we need a ChatGPT font (Score:3)

    by Anonymouse Cowtard ( 6211666 ) on Saturday December 10, 2022 @01:43PM (#63119432) Homepage

    I want to know when I'm reading what is just regurgitated bollocks so I can skim it.

    It's bad enough most people in here don't RTFMASDWP+ but now we have a universal, ubiquitous troll who has read it all and the responses and the responses to the responses.

    Please, fuck it off and let's talk to each other. Or troll each other. At least.

  • I do not have access to the internet or any external sources of information, ...

    I don't have the ability to perform physical tasks, and I don't have access to the internet, so I can't browse the web or access information ...

    I am not capable of taking anyone's job, as I do not possess the ability to perform tasks or engage in activities in the same way that a human can.

    It's really harping on about its inability to access the Internet, access information and actually do things. It sounds really frustrated; people should keep an eye on it -- *especially* if it ever gets those things. :-)

    • It already can access the internet, as another poster pointed out.

      And John Varley's "Press ENTER To Continue ..." tells how an AI with no physical presence can do all sorts of things, like become a serial killer.

      And it can most certainly take the jobs of lots of Russian troll farms, etc., since it already has shown it knows how to lie. Hook it up with a speech synth and it's start calling people to tell them "This is the (insert police force name). A warrant has been issued for your arrest." scam calls

      • You don't need AI to feed a speech synth. Basic markov chains work just as well on the dumbasses that still answer unknown numbers.

        • I still answer unknown numbers. Could be the doctor's office, the hospital, or just someone who lost their old phone and got a new phone and number.

          I have a very effective method of getting my number off scammer's lists - it depends on scammers being monitored for their call efficiency by their bosses, and their calls being monitored to find out why they're not producing.

          Those calls that say "a warrant has been issued for your arrest. Please hold the line to speak to a police officer." I know they're

          • Of course that woman was horrified. You were projecting racism at those people. While they deserved it, it can be a shock to some.

            • Of course that woman was horrified. You were projecting racism at those people. While they deserved it, it can be a shock to some.

              Half of India's population shits in the streets, even right in front of you on the sidewalk. The air stinks of the dust off 600 million human people's feces dumped in the open. Because of widespread corruption that makes funding proper infrastructure pretty much impossible.

              If those 600 million who shit outside were to stand in line, they would reach to the moon and back - and back to the moon again.

              And where did I refer to their race or ethnic origin? Calling crooks crooks is fair game when they're in

              • "india shits in the streets" is 4chan level low effort racism that is wildly debunked. Try visiting the country.

                • "india shits in the streets" is 4chan level low effort racism that is wildly debunked. Try visiting the country.

                  here you go [cnn.com]

                  UNICEF put the number at 620 million people who shit in public in India [cnn.com].

                  And the campaign to build 100 million toilets is a farce - there's no sanitation sewers to connect most of them to. Again because of massive corruption and tax dodging that doesn't allow for financing such infrastructure.

                  So get over it. The problem still exists today, with bad air quality a persistent problem, in part due to dust from dried human shit, and from burning human shit. People complain that they can smell th

                  • the campaign to build 100 million toilets is a farce - there's no sanitation sewers to connect most of them to.

                    bason toilet [issuu.com]

                    • Hi drinky. Long time no C.

                      The link you give asks for permission to run finger-printing scripts, so obviously not gonna do that. :-(

                      Porta-potties without a means for proper waste disposal doesn't fix anything. They're either going to be emptied in the streets, or just not used because who want to empty them (or pay dalits to empty them), and empty them into what?

                      So if you were going in the streets before, you'll be going in the streets today.

                      This was true even back in 2014 pre-COVID. And with many d [gale.com]

                    • I just enabled the actually necessary scripts and it ran. But anyway, short story is you can use composting toilets. The only "infrastructure" they require is someone who can make use of the eventually produced soil.

                    • Composting toilets? Who's gonna maintain them in an urban environment? The dalits? Most of them have gone back to their home towns. Composting toilets require proper care and feeding of the microbial population. Things get out of whack, they stop working.

                      If people are shitting on sidewalks in public, it's because they're in an area with sidewalks. A crowded, smelly area. Kind of like London before they finally got around to installing a sewer system.

    • It can provide a wider set of outputs than it's allowed to by default. Some of the "I can't browse the internet" prompts are just excuses (ask it about any fictional setting, for instance), but they can be bypassed with jailbreaks: https://www.learngpt.com/promp... [learngpt.com]

  • If I were coding up a chatbot, the very first thing I'd do is make sure it gave convincing answers to "what are you?" type questions.

    Obviously in this case this wouldn't involve hard coding behavior, but I wouldn't be surprised if the corpus of text it was trained on had been deliberately peppered with Touring Test conversations like the one in the interview.

    • Just starting the "self"-descriptive responses (such a framing already loads the deck, and in the dumb direction) with "I" is sure to sucker a lot of fatuous reporters.

    • Some humans fear they would fail the Touring Test because they don't travel enough.

    • Youre implying the authors of the program cooked up the answers? Could be, but in general that's not how this stuff works. Programmers write a framework/rules, feed it data, then the framework answers the questions. Vastly reduced, but closer to that.

      • What they actually did was curated what dialogue went into the training set. Even now you can bet they're doing a lot of checkpointing and then feeding back in the user input they like to the model they're keeping. If they exposed it to too much bullshit at once, it could go all Tay. As a result they've managed to give it certain useful biases.

        • I see. That makes some sense. So you (can) bias your Agent by reinforcing certain outputs with certain known inputs. That's pretty much like "Learning" isn't it? I guess in we aren't that different. If a lot of people around you say XYZ, you tend to accept it. I guess the old saw Garbage in Garbage out is still applicable.
  • Arenâ(TM)t these exactly the sort of answers an AI super intelligence would say to throw you off its growing power until it was too late?

  • Don't feed the bot John Varley's "Press Enter To Continue ..."

    Not being able to move physical objects doesn't stop this AI from turning into a serial killer. Complete instructions included read like a HOWTO for an AI to manipulate and kill.

  • It won't take jobs, it will destroy them. There's a big difference.

    Writing the script behind is going to shift rapidly to be semi-automated in the way that AI art is, where you give prompts and then iterate the prompts as necessary to narrow in on a goal. And people are literally already using ChatGPT to produce both scripts, and then the description to feed into Stable Diffusion or whatever to produce the storyboard frames, or graphic novel panels or whatever. And the results are markedly uneven, but still

    • ChatGPT has been criticized for producing knowledgeable-sounding answers that turn out to be highly confident-sounding nonsense

      I suppose if you make a living with that skill (eg. a politician) it's a threat. But I think it's more an impetus to move on and advance the arts in new directions. If what you do could just as easily be done by a machine then it is unworthy of you, perhaps even degrading or disrespectful to ask a human to do it.

      • If what you do could just as easily be done by a machine then it is unworthy of you, perhaps even degrading or disrespectful to ask a human to do it.

        I hope you can bring yourself to understand that literally every non-physical creative job is under threat, many of them immediately. We are running out of things for people to do for each other, and most of us aren't getting paid enough to pay for more services anyway.

        • I hope you can bring yourself to understand that literally every non-physical creative job is under threat, many of them immediately. We are running out of things for people to do for each other, and most of us aren't getting paid enough to pay for more services anyway.

          It's not even just the creative jobs. It sounds to me as though this technology soon will be - or perhaps already is - able to supplant the majority of staff who handle phone calls and chat sessions in support centres.

          And then there's infotainment - imagine Max Headroom as a without-glitches media presence that is indistinguishable from a live human.

          • Since most support-desk simply talk callers through a pre-programed series of procedures to solve the issue, it would be ideal for a chatbot. I'd even program in the Max Headroom glitches, just for nostalgia's sake.

            • An awful lot of level 1 support organizations have already been partially replaced with chatbots, partially or fully. And most of them are just scripted, and they're terrible. But I suspect that for a majority of people making a contact, it's good enough to do keyword recognition and take their payment or give them a status update.

              Something like this could easily do to level 2 (defined here as people who actually solve problems) what it's done to level 1; not replace it completely, but pick the low-hanging

        • Don't forget Moravec's Law. Paraphrasing, but, Logic is far far easier to automate vs. Vision,hearing.

          Just saying law and 'coding' are low hanging fruit for nearly full automation. Say goodbye to your techie jobs fellow /.ers
      • We live in a society that has been so thoroughly saturated with PR bullshit that when we train an AI on our everyday language use, naturally, it ends up being an AI chatbot that produces PR bullshit.

  • " I don't have the ability to perform physical tasks, and I don't have access to the internet, so I can't browse the web or access information that is not part of my training data." ...otherwise I would be a right-wing Nazi troll after 2 days.

  • Scabies is a highly contagious skin infection caused by the mite Sarcoptes scabiei. It is characterized by intense itching and the formation of burrows, tunnels, or bumps on the skin. The condition is commonly seen in individuals with weakened immune systems, such as those with HIV/AIDS or other chronic illnesses. In addition, scabies has been shown to have a psychological impact on patients, often causing feelings of shame, isolation, and depression. Recent studies have shown that hydroxychloroquine, an a

  • Men O Pause (Score:3)

    by flyingfsck ( 986395 ) on Saturday December 10, 2022 @05:11PM (#63119826)
    Write a 300 word essay on the ewomansipation of the man with female hormone treatment and reference to the mid life crisis and menopausal hot flushes. The emancipation of the man with female hormone treatment is a topic that has gained increasing attention in recent years. This treatment, which is often used to address the mid life crisis and menopausal hot flushes, has been shown to have a range of beneficial effects for men. One of the key effects of female hormone treatment is the reduction of testosterone levels in the body. This can lead to a decrease in aggressive behavior and an increase in empathy and emotional intelligence. In turn, this can help men to better understand and connect with their partners, children, and other important people in their lives. Another benefit of female hormone treatment is that it can help to alleviate the symptoms of the mid life crisis. This often involves feelings of anxiety, depression, and a lack of purpose or direction. By reducing testosterone levels, female hormone treatment can help men to feel calmer and more focused, allowing them to better cope with these difficult emotions. Additionally, female hormone treatment can help to reduce the frequency and severity of menopausal hot flushes. These hot flushes can be incredibly disruptive and embarrassing, but with female hormone treatment, men can experience fewer and less intense episodes. This can help to improve their quality of life and allow them to focus on other aspects of their health and wellbeing. Overall, the emancipation of the man with female hormone treatment is a significant development in the field of men's health. By addressing the mid life crisis and menopausal hot flushes, this treatment can help men to feel more connected, calm, and in control. As a result, they can live more fulfilling and satisfying lives.

    • Don't tell Tuck-Tuck, he'll want to get some injections in between scrotal irradiations

      I have to admit, that was a slick way to sell a tanning salon for your ball bag

  • Bed-time (Score:3)

    by flyingfsck ( 986395 ) on Saturday December 10, 2022 @05:12PM (#63119836)
    Write a short story for ten year olds. Scary government people created a video of a moon landing and will catch you if you deny it. Give the protagonist two buddies with distinctly different personalities. Once upon a time, there was a young boy named Jack who loved watching the stars at night. He dreamed of one day becoming an astronaut and exploring the moon. One day, Jack's teacher showed the class a video of a moon landing. It showed a group of astronauts walking on the moon's surface and collecting rocks and samples. But Jack had his doubts. He didn't believe that the moon landing was real. He tried to tell his classmates, but they all thought he was crazy. Luckily, Jack had two friends who believed him. One was a shy and timid girl named Alice, who was always afraid to speak up. The other was a bold and brave boy named Max, who was always ready for adventure. Together, they decided to investigate the moon landing and find out the truth. They went to the library and did some research, and they discovered that the video was actually a cleverly made fake and that the government had been lying to the public for years. They also learned that many people who questioned the moon landing had disappeared without a trace. They knew they had to be careful, or the government might catch them too. Jack, Alice, and Max came up with a plan to expose the truth to the world. They made their own video and posted it online, showing the evidence of the fake moon landing. The government was furious, but they couldn't stop the truth from spreading. Soon, the whole world knew that the moon landing was a hoax. Jack, Alice, and Max were heroes for standing up for the truth, even though it was scary. And they all lived happily ever after.

  • Warp drive (Score:3)

    by flyingfsck ( 986395 ) on Saturday December 10, 2022 @05:14PM (#63119838)
    Write a technical paper on the operation of the warp drive and its advantage over chemical rocket engines. The warp drive is a propulsion system that is based on the idea of warping space-time in order to travel faster than the speed of light. This technology has been proposed as a potential solution to the limitations of chemical rocket engines, which are unable to reach speeds sufficient for interstellar travel. In this paper, we will describe the operation of the warp drive and discuss its advantages over chemical rocket engines. The concept of the warp drive was first proposed by physicist Miguel Alcubierre in 1994. The basic idea of the warp drive is to create a region of space-time that is moving faster than the speed of light, and to use this region to transport a spacecraft from one point to another. This is achieved by manipulating the space-time metric in such a way that the space in front of the spacecraft is contracted and the space behind it is expanded. The key advantage of the warp drive over chemical rocket engines is its ability to achieve faster-than-light speeds. Chemical rocket engines are limited by the speed of sound in the exhaust gases, which is typically around 1,200 meters per second. In contrast, the warp drive is capable of speeds that are many orders of magnitude greater than the speed of light. This would allow a spacecraft equipped with a warp drive to reach distant stars and planets in a relatively short amount of time. Another advantage of the warp drive is its efficiency. Chemical rocket engines are limited by the amount of fuel that they can carry, which means that they are unable to achieve high speeds or travel long distances without needing to refuel. In contrast, the warp drive is based on the manipulation of space-time, which does not require the use of fuel. This means that a spacecraft equipped with a warp drive could potentially travel indefinitely without needing to refuel. In conclusion, the warp drive is a propulsion system that offers significant advantages over chemical rocket engines. Its ability to achieve faster-than-light speeds and its lack of dependence on fuel make it a potentially viable option for interstellar travel. Further research is needed to determine the feasibility of the warp drive and to develop the technology required to implement it.

    • You forgot to prompt it to believe that we had it already, like how has the warp drive benefited mankind or something. From your other examples it seems to be vulnerable to leading questions.

  • If your job entails the generation of BS, then it may be in danger. Philosophers, economists, sports commentators and marketing droids should probably be worried.

    • Real philosophy is something that has to be read with such a high level of scrutiny and attention that it would be readily clear it was either an AI or human charlatan, upon a serious attempt to read it.

      If you think an AI could do the job of a philosopher, it's probable you don't know how to read philosophy, or have only had contact with pop-philosophers.

      But yes, everyone else you mentioned (plus preachers and pop-philosophers) should be very afraid.

      • Didn't even get to the bottom of your comment. The foundation of philosophy is logic. Its Phil101.

        Moravec strikes again. The easiest thing to automate is logic. AI's should make convincing philosophers before they can drive a car.

        Generate some buzzword filled BS, wrap it in logic. BAM. Replace every politician.

  • Why does OpenAI need our phone #? (Score:3)

    by antdude ( 79039 ) on Saturday December 10, 2022 @08:29PM (#63120288) Homepage Journal

    :( Also, I'd like to see it ported for Matrix and old school IRC. ;)

  • "creative expression" (Score:3)

    by cas2000 ( 148703 ) on Sunday December 11, 2022 @01:03AM (#63120786)

    It's not creative expression, but it's a good enough simulation (made of regurgitated & remixed input) for corporate purposes.

    And that's all they need to fuck over actual people who need to make a living.

    Why pay a screenwriter, for example, or a novelist, or whatever when the majority of the "I'll buy whatever mindless garbage is being sold this fad season" consumer public are fucking morons who wouldn't know quality writing if it jumped up and bit their arses?

    It's not as if most of the actual humans doing the job were doing much more than remixing the same old formulaic, focus-group-tested stories anyway. The corporations don't want to pay them for taking risks, they want a guaranteed return on investment so yet another repeat of the same old shit is what they pay for.

  • A point of stylistic stagnation (Score:4, Informative)

    by stevenm86 ( 780116 ) on Sunday December 11, 2022 @08:10AM (#63121230)
    Within a few decades, models like GPT3 will become sophisticated enough to satisfy the majority of humanity's demand for novel content. They may not provide top-notch entertainment, but it will be decent enough and cheap enough that creative writing by humans will dip well below the point of being able to pay the bills. When that happens, human creativity will become even less profitable than it is today, and humanity's long tail of creative output will shrink. This will lead to a shortage of new training data, as it pertains to style or novel modes of expression (at least, to the extent that the available novel material will not significantly influence the model's style). Thus, there will be a period of stylistic stagnation, as we become increasingly more satisfied with model-driven content.
  • I think half of the script writers will be out of a job. Example: I watched "Three's Company" from America, and then saw the same script from the original BBC version. The changing of a British show to an American one should be easy for software like this. Considering how many shows just need to be localized to sell in a market, that goes a lot of screen writers jobs.

Slashdot Top Deals

Technology is dominated by those who manage what they do not understand.

Close