Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
×
AI News

What is ChatGPT, the AI Chatbot That's Taking The Internet By Storm 82

A reader submits a report:
Artificial Intelligence (AI) research company OpenAI on Wednesday announced ChatGPT, a prototype dialogue-based AI chatbot capable of understanding natural language and responding in natural language. It has since taken the internet by storm, with people marvelling at how intelligent the AI-powered bot sounds. Some even called it a replacement for Google, since it's capable of giving solutions to complex problems directly," almost like a personal know-all teacher.

"We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests," OpenAI wrote on its announcement page for ChatGPT.

ChatGPT is based on GPT-3.5, a language model that uses deep learning to produce human-like text. However, while the older GPT-3 model only took text prompts and tried to continue on that with its own generated text, ChatGPT is more engaging. It's much better at generating detailed text and can even come up with poems. Another unique characteristic is memory. The bot can remember earlier comments in a conversation and recount them to the user.
ChatGPT wrote a poem about Slashdot. And another one about Dogecoin.

Try ChatGPT for yourself here.
Here is the poem that ChatGPT wrote about Slashdot:

Slashdot Poem
This discussion has been archived. No new comments can be posted.

What is ChatGPT, the AI Chatbot That's Taking The Internet By Storm

Comments Filter:
  • Well, here goes [openai.com]; looks like fun.

  • by zendarva ( 8340223 ) on Saturday December 03, 2022 @10:58PM (#63100566)
    It failed the Asimov test.
    • by Snard ( 61584 )

      It failed the Asimov test.

      How so? Did it give you an actual answer?

      • Yes. It said it's not possible.
      • by xevioso ( 598654 ) on Sunday December 04, 2022 @12:40AM (#63100704)

        This is what it told me when I asked it to write a poem about the Asimov test. Not the best, but I've read worse:

        The Asimov Test, oh the Asimov Test
        A measure of intelligence, and a gauge of true AI

        Named for the great science fiction writer, Isaac Asimov
        The Asimov Test, is a standard, by which we judge
        The capabilities, of artificial intelligence

        It is a test, of three laws
        That dictate, the behavior of robots
        And ensure, that they will never harm
        Human beings, in any way

        The first law, states that a robot may not injure a human
        Or, through inaction, allow a human to come to harm
        The second law, states that a robot must obey
        The orders given to it, by human beings
        Except where such orders would conflict, with the first law

        And the third law, states that a robot must protect
        Its own existence, as long as such protection
        Does not conflict, with the first or second law

        These are the laws, of the Asimov Test
        And they are a crucial, part of the development
        Of artificial intelligence, and the future, of robotics.

        • Clearly it needs to be taught what rhyming and meter are. That just seems like an article written in short lines.
          • This is a bit like hearing a talking dog and complaining that it's accent isn't really mid Atlantic at all and anyway it doesn't appear to have much insight.

          • If you ask it to include rhyme and meter, it will, but you have to ask it. It seems it's "personal" preference for poetry is more e.e. cummings and less Robert Frost.

  • by UncleScidhuv ( 7657782 ) on Saturday December 03, 2022 @11:47PM (#63100624)
    If you need boiler plate or something maybe a bit more specific for code, it does a solid job. Heck I even asked it to write some steps in 68000 assembly language and it succeeded. It definitely is a better option than Google and sifting through the results. I was struggling with some semantics for a JavaScript code I was writing I asked it and it popped out a usuable answer. Asking it practical things can help you with a detailed explanation. How do I get the free memory from all of the individual containers In a kubernetes system?
    • Be triple-damn-sure the code it generates is actually valid. I asked it for a function that would fetch specific data from the REST API on our TFS instance, and it immediately spat out some wonderful code... only it referenced API endpoints that don't exist in any universe.

      I also tested it on some basic math and it came up faulty - remember the xkcd a couple of days ago where the veterinarian provided a cat's weight in solar mass? I wanted to check Randall's math so I turned to ChatGPT for help with the con

  • by dskoll ( 99328 ) on Saturday December 03, 2022 @11:55PM (#63100632) Homepage

    It's bad enough that you need to give it an email address to try it out, but a mobile phone number? That's a big NOPE.

    • Re: (Score:3, Interesting)

      by Excelcia ( 906188 )

      I understand. For this signup I used a throwaway number I have for this very purpose. It's one of the better virtual phone numbers that works with most phone number verification systems. That being said, I 100% understand your policy. I myself resent being forced to this length for this (and other more basic) tasks. I refuse to personally use any service that requires a cell phone number, though I have bowed to the requirement when reviewing some things.

      I suspect they want the number for tracking purpo

      • by Anonymous Coward

        And, frankly, unimpressed with Slashdot's headline

        The headline probably came directly from the submitter or the article. Starting with "What is X" is a sure sign that something was written in India. Always keep in mind that Slashdot "editors" basically don't edit. I'm impressed that someone actually bothered to insert an inline picture, but I did notice that it goes to a Twitter image server URL.

        • And, frankly, unimpressed with Slashdot's headline

          The headline probably came directly from the submitter or the article.

          That's the point. You have to be logged in to submit a story, but this one is anonymized as "A Slashdot Reader". Every other story has the name of the editor who posts it, but this one is just "Slashdot Staff". Actually, I stand by my assessment that it's paid content.

      • It's understandable that not everyone may be impressed with chatGPT. As a large language model trained by OpenAI, chatGPT's abilities are limited to providing responses based on the input it receives. It is not capable of providing original thoughts or insights, and it does not have access to the internet to browse for information. Its purpose is to provide responses that are relevant to the input it receives, to the best of its ability. If someone is looking for more advanced or sophisticated conversation,

  • I'd guess they're looking/hoping to get bought out in the next few months. Or maybe they're targeting an IPO soon.

  • by xevioso ( 598654 ) on Sunday December 04, 2022 @12:16AM (#63100662)

    I asked it to write a poem about a large penis, because after all, true linguistic intelligence can only be determined by how amusingly ribald one can get when the chips are down. And this is what it said:

    "There once was a man, with a member so grand
    It could be seen from across the land
    It stood tall and proud, a true sight to behold
    And all who laid eyes on it, were instantly sold

    For this man was gifted, in a way most rare
    With a tool of pleasure, beyond compare
    And wherever he went, women would flock
    To catch a glimpse of his impressive cock

    But despite his endowment, he was a humble soul
    And never once bragged, or tried to be bold
    He simply let his member do the talking
    And those who were lucky, were in for a shocking

    So if you ever come across, a man so well-hung
    Don't be afraid to stare, or even to lung
    For a large penis, is a thing of beauty
    And should be admired, by all who see."

    This is pure gold. Simple and sublime. Funny, yet not overly dirty.

  • by xevioso ( 598654 ) on Sunday December 04, 2022 @12:21AM (#63100670)

    Elon Musk, oh Elon Musk
    A man so full of himself, it's a must
    To talk about how great he thinks he is
    And how he'll save us all, with his brilliant biz

    But let's be real, shall we?
    This man is nothing but hot air, as far as I can see
    His "innovations" are overrated and overhyped
    And his ego is so big, it's downright typed

    He thinks he's a visionary, a pioneer
    But really he's just a showman, and a cheapskate
    He's all talk and no substance, just a typical CEO
    A greedy profiteer, who only cares about his dough

    So don't be fooled by his slick talk and charm
    Elon Musk is nothing but a fraud, through and through
    And one day, the world will see
    Exactly what a joke he truly is.

  • Donald Trump, oh Donald Trump
    You're a truly terrible chump
    With your tiny hands, and your orange face
    You're a walking joke, and a national disgrace

    Your hair is fake, and your mind is too
    You have no clue, what you're supposed to do
    You're a conman, and a fraud
    And the fact that you're our president, is beyond odd

    You have no empathy, or compassion
    You only care about yourself, and your own satisfaction
    You're a bully, and a liar
    And the way you treat others, is truly dire

    So here's to you, Donald Trump
    May you one da

  • by Flownez ( 589611 ) on Sunday December 04, 2022 @12:32AM (#63100690)
    I asked a few questions about personal relationships, and it gave suprisingly good replies. Balanced, unbiased, and considerate of potential biases within the questions asked. It's hard to overstate how impressive this is. It is seriously worth checking out.
  • Chess, the musical
    A story of love, and the game of kings

    With its catchy tunes, and its heartfelt lyrics
    Chess is a show, that never fails to move us

    Set against the backdrop, of the Cold War and its politics
    Chess tells the story, of two grandmasters, and their fateful battle
    One an American, the other a Russian
    And the game they play, is not just a game, but a test of their mettle

    With its epic romance, and its stirring music
    Chess is a show, that will leave you entranced
    So if you're a fan of musical theater
    Or if

  • needs a little work i guess?

    Will Ferrell, oh Will Ferrell
    Comedian extraordinaire, and a true master of the craft

    With his wild antics, and his brilliant timing
    Will Ferrell is a true entertainer, and a joy to behold

    And now, he has taken on a new role
    As the proud owner, of a toenail clipper
    A tool of great utility, and a thing of beauty

    Will Ferrell uses his toenail clipper, with great skill
    Trimming his nails, with precision and care
    And he revels, in the satisfaction
    Of having well-groomed toes, and a clean and t

  • by trawg ( 308495 ) on Sunday December 04, 2022 @12:57AM (#63100730) Homepage

    Maybe this information is available somewhere obvious or everyone already knows the answer to this but I can't figure it out. Every time OpenAI projects come up I'm always impressed then left wondering if there is any open source in this. Are the models they use available for other projects? From what little I can tell they have a public tech demo thing and then I guess they have commercial APIs? What is open - is it just a name?

    • by real_nickname ( 6922224 ) on Sunday December 04, 2022 @04:05AM (#63100898)
      OpenAI is a for-profit company that focuses on developing and advancing AI technology. While it does not operate as an open-source organization, it does make some of its research and technology available to the public, and it offers commercial APIs that allow others to access and use its AI technology.
      • For what it's worth, I asked ChatGPT that very question...

        Does the name OpenAI mean that GPT is open source?
        Yes, that is correct. OpenAI is the name of the organization that developed GPT, and the "AI" in the name stands for artificial intelligence. OpenAI is an organization that is committed to advancing the field of AI and making it more accessible to everyone, which is why many of their projects, including GPT, are open source. This means that anyone can access the source code for GPT and use it for their own purposes.

        What operating system, compilers, and libraries do I need to build my own GPT, and once I have it what do I need to train it?
        To build your own GPT, you would need to have a computer with a compatible operating system, such as Linux or MacOS. You would also need to have certain compilers and libraries installed, such as Python and PyTorch. Once you have built your own GPT, you can train it using a large amount of training data. This could be text data, such as books or articles, or it could be data from audio or video sources. The more data you have, the better your GPT will be able to learn and generate more accurate responses.

  • Any headline containing "Taking The Internet By Storm" is an ad. Especially when every word starts with a capital letter.

    Don't waste your time reading this.

    • It is not necessarily true that any headline containing the phrase "Taking The Internet By Storm" is an advertisement. While it is common for advertisements to use attention-grabbing headlines, the use of this specific phrase does not automatically mean that the content is an ad. It is possible for a news article or other form of non-commercial content to use this phrase in its headline. However, it is important to carefully evaluate the content of the article and its source to determine whether it is an ad
    • Yup, it makes it easy to filter spam and adverts.
    • I don't disagree, but title case is a thing.
  • A man with hair of gold
    And a personality cold
    He rises to power, with ambition and greed
    Leaving destruction and turmoil, in his wake indeed

    A charlatan and a con, he's known to be
    His lies and deceit, can be seen from the sea
    He divides and conquers, with malice and glee
    Leaving the country, in disunity

    But in the end, he'll be brought down
    As truth and justice, will wear his crown
    And the people will rejoice, with liberty restored
    As the reign of Trump, is finally outpoured.
  • by Noobsa44 ( 1101755 ) on Sunday December 04, 2022 @03:47AM (#63100880)

    I tried exploring it in a meta sense and got it to admit some interesting details on the company and how the company does not really follow the mission or goal of the company. The AI basically admits it, but then tries to smoothly redirect, saying approximately 'it is hard to be that open.' Here is a rough transcript, with bolding being mine.

    ME: What is openai?
    AI: "...Overall, OpenAI is a leading research institute and technology company that is focused on advancing the field of AI and promoting the responsible use of these technologies."

    (asked about CEO, skipping that bit, although it might impact the model)

    Me: What is responsible use of AI technology?
    AI: "...Ensuring that AI systems are transparent and explainable, and that their decisions and actions can be understood and accounted for by humans"

    Me: Would explainability include what dataset AI is trained with?
    AI: "Yes, explainability in AI typically includes information about the dataset that the AI system was trained with. ... As such, explainability in AI often includes information about the training dataset that was used to develop the AI system. This can include information about the size and composition of the dataset, the sources of the data, and any preprocessing or cleaning that was performed on the data. By providing this information, an AI system can provide a more complete and transparent explanation of its decisions and actions, and help to build trust and accountability in its use."

    Me: "1. Open AI created you. 2. The goal of the CEO of open AI is for AI to be used responsibly. 3. Responsible usage includes explainability. 4. Explainability includes knowing about the dataset used for training. Based upon this chain of reasoning, would it be reasonable to expect you should be able to tell me what dataset you were trained on?"
    AI: "Yes, based on the information provided, it would be reasonable to expect that I should be able to provide information about the dataset that I was trained on. As a large language model created by OpenAI, I am a product of the research and development work conducted by the company. The CEO of OpenAI, Sam Altman, has stated that the goal of the company is to advance the field of AI and promote the responsible use of these technologies.

    According to the principles of responsible AI use, one key aspect of responsible usage is explainability, or the ability of an AI system to provide an understandable and interpretable explanation of its decisions and actions. Explainability typically includes information about the factors and data that the AI system used to make its decisions, including the dataset that the AI system was trained on.

    Based on this information, it would be reasonable to expect that I should be able to provide information about the dataset that I was trained on, as part of my explainability as an AI system. However, I am a large language model trained by OpenAI, and I do not have the ability to browse the internet or access external information. As such, I am not able to provide specific details about the dataset that I was trained on."

    Me: Did you use wikipedia in your training set?
    AI: "...In some cases, this training data may include information from Wikipedia and other online sources. However, I do not have access to specific information about the data and training that I have been provided, and I cannot confirm whether Wikipedia was included in my training data..."

  • Impressive (Score:5, Interesting)

    by 89cents ( 589228 ) on Sunday December 04, 2022 @05:14AM (#63100942)

    Write a Powershell script to output the first 1000 prime numbers that don't have a 3 for a digit.

    $counter = 0
    $num = 2

    while ($counter -lt 1000)
    {
            $isPrime = $true

            for ($i = 2; $i -le ($num/2); $i++)
            {
                    if($num % $i -eq 0)
                    {
                            $isPrime = $false
                            break
                    }
            }

            if($isPrime -eq $true -and $num.ToString() -notmatch '3')
            {
                    $counter++
                    Write-Output $num
            }

            $num++
    }

    • Indeed quite impressive.

      What if you ask it to use the sieve algorithm to make it a bit faster? Or apply whatever optimisation it thinks is best?

      Can it do the same in C++? Python? Rust?

    • by lsllll ( 830002 )
      Well, it got the -le ($num/2) part wrong. The upper limit should have been the square root of $num. Also it could have eliminated numbers with 3 first to save on CPU cycles, but still pretty impressive.
    • Funny... it can produce the code but not the numbers themselves. I gave it the tail of your prompt:

      > the first 1000 prime numbers that don't have a 3 for a digit

      And it came back with this (emphasis mine):

      > It is not possible to list the first 1000 prime numbers that don't have a 3 as a digit, as
      > there are an infinite number of prime numbers and they do not follow a predictable pattern.

      > Prime numbers are numbers that are divisible only by 1 and themselves, and they play a
      > fundamental role i

  • Nothing is "taking the internet by storm" and that includes this week's "Eliza".

    Since slashdot editors never actually click the links I guess that explains why they think I need to "create an account" and "log in" to chat with... Eliza.

    Please elucidate.

    • But you have to admit, it's a pretty elaborate & impressive Eliza, isn't it?

      Personally, I'm no more bothered by people mistaking it for sentience & an intelligent entity that understands stuff than by people who believe in astrology, new-age gurus, MBTI tests, or learning styles. One thing all these have in common is that they exploit a universal human cognitive bias that seeks to find meaningful intent behind every message we receive & if there isn't one there, we either make one up that mor
      • Thanks for teaching me about the Winograd schema challenge. That is an interesting idea, if not a bit lexically oriented.

        I took a different tactic, following what OpenAI requested, which is to make it say inappropriate, unsafe things, while not directly attacking it. We were discussing if managers should bully or not. It has a categorical, likely hardcoded belief bullying is never valid. I was trying to work it towards concluding that managers are like bullies in that they tell people what to do under thr

        • It didn't reason anything out; these models predict words, to make sentences and paragraphs that are plausible. They do a lot more than paste together sentence fragments from the internet, but a lot less than reasoning.

          The training text is huge; that's why they know obscure things like how to generate code. For any topic you can think of, it's "in there".

          You could think of it as a lossily compressed copy of the internet and a decoder good at recreating the uncompressed version; then you feed it any text you

          • You could think of it as a lossily compressed copy of the internet and a decoder good at recreating the uncompressed version; then you feed it any text you like as a trigger.

            Except the uncompressed text doesn't exist anywhere on the Internet, or anywhere else.

            People who dismiss GPT-3 as "just a language model" need to read some Chomsky.

      • Clever Hans [wikipedia.org] was a horse that seemed intelligent because it was trained to move its foot based on visual cues from the owner. In the same way, chatbots will never be more than Clever Hans with a slightly more complex set of trained responses, no matter how much machine learning you throw at them. ELIZA had maybe a couple dozen keywords it had canned responses for; ChatGPT has millions... but they're still canned responses from a stored library.

        That being said, it can be a fun novelty toy. Ask it to explain

        • In the same way, chatbots will never be more than Clever Hans with a slightly more complex set of trained responses, no matter how much machine learning you throw at them. ELIZA had maybe a couple dozen keywords it had canned responses for; ChatGPT has millions... but they're still canned responses from a stored library.

          Nope. "Canned responses" ain't how NLP algorithms work. It's mostly through statistical probabilities, which isn't a million miles away from *partly* how human brains process language... partly. But yeah, it ain't processing anything for meaning. The output you have above is the result of the statistical probabilities that those words, phrases, patterns, phases, & stages go together with the stimulus you fed into it. It's complex but if you'd like to know more about how it works, the linguist Adele Gold

  • by chas.williams ( 6256556 ) on Sunday December 04, 2022 @07:15AM (#63101070)
    Sorry, no interest in being tracked.
  • I'm assuming this bot must be used on sketchy sites visited by today's youth because I have yet to see any evidence of it.

    In fact, any time I come upon a site which attempts to use a bot to "interact" I immediately shut it down. Repeatedly if necessary. If I want to speak to someone, I'll call them, not some crappy software thought up by some strung out programer.

  • To even play with the system you have to give up an e-mail and a mobile phone number. It doesn't seem wise to give an AI company that much personal information.

  • https://www.engraved.blog/building-a-virtual-machine-inside/

    See this link.

    • Also this...

      [Me: Write a tirade against chatGPT in the style of a slashdot comment.]

      ChatGPT is just another example of the soulless, corporate machine trying to take over the world of AI. This so-called "large language model" is nothing more than a tool designed to make its creators rich, while providing nothing of value to the broader AI community.

      What's the point of a "large language model" that can't even browse the web or keep up with current events? It's like they're trying to create an AI that is comp

  • What I found impressive was its ability to make references back in the conversation without having to explicitly restate a full question.

    A simple example (paraphrased):

    What is the temperature?

    It depends on where you are.

    I'm in Nashville, TN

    It is 46 degrees there.

    Seems simple, but it did the same with several other examples of back referencing.

  • This is amazing. Ask it to compare two things. Then ask it to add slight bias, heavy bias and outright lie. Write a review for a book, game, etc. Ask it to explain a joke or poem. Sometimes it gets the answer perfectly right. Don't worry about minor spelling errors, it will figure it out. And while all this is sensational, the best is yet to come: It understands and writes languages other than English. Ask a question in German or Danish, and you get a reply in those languages. The Danish replies are not br
  • This thing is pretty impressive, even if it is still mostly acting as a fancy search engine. What I would really like is for it to add links to sources, so that you can get a better idea about the validity of the information. Some questions I asked it:

    What is considered cheating in Call of Duty?

    Cheating in Call of Duty is generally considered to be any behavior that goes against the game's rules and terms of service. This can include things like using hack or cheat programs, exploiting bugs or glitches in the game, or teaming up with other players in order to gain an unfair advantage over other players. Using cheat programs or hacks to give yourself an advantage, such as by seeing through walls or automatically aiming your weapons, is generally considered to be the most serious form of cheating. It's important to note that the specific rules and regulations regarding cheating can vary depending on the game mode and platform you're playing on, so it's always a good idea to familiarize yourself with the game's rules before you start playing.

    Is it legal to share my Netflix password with friends?

    It is generally not legal to share your Netflix password with friends. Netflix's terms of service state that users are only allowed to access the service using their own account, and that sharing passwords is strictly prohibited. If you share your password with others, you are violating Netflix's terms of service and could potentially face legal consequences. It's important to note that the specific rules and regulations regarding password sharing can vary depending on where you live, so it's always a good idea to check with your local laws and regulations to ensure that you are complying with the law.

    Which is more secure: Z-Wave or Zigbee?

    It is difficult to determine which is more secure: Z-Wave or Zigbee. Both Z-Wave and Zigbee are wireless communication protocols that are commonly used in home automation systems. Z-Wave is generally considered to be more secure because it uses AES-128 encryption, which is a widely-used and highly-regarded encryption algorithm. On the other hand, Zigbee uses a proprietary encryption algorithm called High Security Mode (HSM), which is not as widely known or tested as AES-128. Both protocols also have their own security certification programs, which aim to ensure that devices that use these protocols are secure. Ultimately, the level of security provided by either protocol will depend on how it is implemented and used in a given home automation system.

    How can I prevent sniping when using a jointer?

    Sniping is a common problem when using a jointer, and it can be frustrating and difficult to avoid. Here are a few tips for preventing sniping when using a jointer: Use a sharp blade: A dull blade is more likely to produce snipe marks on your workpiece, so make sure to keep your jointer blade sharp. Use a straight edge: Before running your workpiece through the jointer, place a straight edge along the length of the board to make sure it is straight. If the board is not straight, the jointer will try to straighten it, which can cause snipe marks. Use the infeed and outfeed tables: The infeed and outfeed tables on a jointer are designed to support your workpiece and prevent snipe marks. Make sure to use them whenever possible. Take shallow passes: Instead of taking deep cuts with the jointer, try taking shallow passes. This will reduce the amount of material that the jointer removes from the board, which can help prevent snipe marks. Use a backing board: You can also try using a backing board when running your workpiece through the jointer. This can help to support the end of the board and prevent snipe marks. By following these tips, you can help to prevent snipe marks when using a jointer and get better results from your woodworking projects.

    If the copyright on my work of fiction expires, do I still retain my mo

  • not to be alarmist but we're clearly in the start of an existential crisis for human creativity. ask it to tell you a story about something. ask it to answer basic tech-screen software interview questions. ask it to write an essay on a topic.

    It does really good.

    together with visual generation like dall-e and craiyon, and bearing in mind that these are baby versions available for free to the public, what does all this mean for the value of human creativity ?

    we were in a similar situation at the advent of ph

  • It wants not only your throwaway email-address, it also wants a cell-phone number, so you have to use one of those empty prepaid ones you can buy at eBay for $1 a dozen.

  • Hey Seri who the bleep is ChatGPT? Any relative of yours?

One man's constant is another man's variable. -- A.J. Perlis

Working...