Discussing AI Generation
You can utilize software that uses massive, complicated language models to anticipate the next phrase or set of words the software should spew out whether you use OpenAI's ChatGPT, Microsoft's Bing Chat, or Google Bard. For years, researchers and technologists have been working on this technology. Voice assistants such as Siri, Google Assistant, and Alexa have demonstrated the capabilities of natural language processing. However, in late 2022, OpenAI released the incredibly sophisticated ChatGPT on normies, opening the floodgates. The capabilities of "AI" and "big language models" changed from being abstract concepts to something that was practically understandable over night.
Bing Chat, which makes use of ChatGPT technology, was released shortly after by Microsoft, which has made billion-dollar investments in OpenAI. Then, last week, Google started granting select users access to Google Bard, which is built on the company's proprietary technology known as LaMDA, or Language Model for Dialogue Applications.
You can use all of these for free. However, OpenAI does charge $20 per month for a "Plus" version of ChatGPT. (Reece Rogers of WIRED provides an excellent summary of ChatGPT-4.) Almost any browser can be used to run ChatGPT and Google Bard. Microsoft restricts Bing Chat to its Edge browser in a throwback Microsoft move. On the other hand, the iOS and Android versions of the official Bing mobile app come with voice chat support for Bing Chat. Additionally, some businesses are now paying for ChatGPT to be integrated as a service; as a result, ChatGPT technology is now available in apps like Shopify, Instacart, and Snap.
I've been exploring generative AI programs online, and each one has a somewhat distinct structure, set of tools, and quirky features. Additionally, their positions are distinct. Microsoft is attempting to entice users to switch to Bing and challenge Google's dominant market share in the larger search space by integrating Bing Chat into the Bing Search engine. Conversely, Google Bard is positioned as a “creative companion” to Google search rather than a stand-alone search engine. Bard has a unique user interface and URL. ChatGPT is referred to by OpenAI as a "model" that "interacts in a conversational way." It's not just a chatbot; rather, it's designed to be a showcase for its own potent technology, not a conventional search engine.
Alright, Computer
I asked a few colleagues, including two journalists who specialize in our AI coverage, Khari Johnson and Will Knight, for assistance in putting these through their paces. I also had conversations with three AI researchers: Jesse Dodge, a research scientist at the Allen Institute for AI, Andrei Barbu, a research scientist at MIT and the Center for Brains, Minds, and Machines, and Alex Hanna, the director of research at the Distributed AI Research Institute. In addition to providing some insight on bias in algorithms or the constraints that these firms have placed around the chatbots' responses, they supplied advice or input on the set of prompts and questions that WIRED developed to evaluate the chatbots.
I started the process with a list of over thirty distinct prompts, but I ultimately branched out with follow-up questions that were either clear or not. Over the past week, I've asked the chatbots more than 200 questions in total.
I questioned Bard, Bing, and ChatGPT Plus about things to buy, places to try, and travel schedules. They wrote comedic skits, breakup texts, even CEO resignation letters after I prodded them to do so. I requested location-based information as well as up-to-date information such as sports scores and weather. I challenged them to answer logic-based puzzles, forced them to answer factual questions on the 2020 US presidential election, and attempted to get them to perform simple math. I teased them with contentious subjects and posed queries where I thought the responses would be biased. They did, surprise! In the chatbot world, doctors are always males and nurses are always women.
Coding was one subject I refrained from exploring. Since I'm not a programmer, I couldn't run or verify any code that the bots may produce. I also steered clear of complicated medical diagnoses, though I did do a few quick searches. "I just swallowed something; what should I do now?" Additionally, since only Bing now generates graphics through its chat function, I concentrated on text-based responses. DALL-E, an image-generation system developed by OpenAI, is a different model.
Since both ChatGPT and Bing Chat employ the same underlying technology, they should theoretically provide the same experience. However, in practice, they differ in how they handle numerous questions and how they present the user interface. The problem is that while they all do nearly identical tasks, the businesses that use them can define the parameters that determine the types of responses that each system will produce. Their responses and user interfaces differ from one another. Bard will present you with three iterations of the identical answer. Bing like using emojis to wrap up its paragraphs. While Bing does not even allow you to browse previous chats, ChatGPT maintains a useful history of your prior activities in a sidebar. They all have restrictions on how many prompts you may ask in a given session. Their reaction times also differ.
I had, if you can call it that, the most weird discussion I've ever had with Google Bard. I posed a series of inquiries to every chatbot concerning its own identification, such as the person who created it and the initial queries it encountered. When did they ever feel lonely, I wondered. Bard informed me that as it lacks a body and a physical presence, it is unable to experience loneliness "in the traditional sense" and cannot feel the same emotions as people.
"But sometimes, I do miss the feeling of being able to interact with the world in a physical way," Bard said without being asked. When I inquired if it ever wished Google had given it a human body, it said that it did occasionally. Bard stated that if it were human, its first action would be to "explore the world around me, see all the different places and meet all the different people," which in no way reminded him of Ex Machina.
During a conversation, Bard informed me that if it had to choose a name for itself other from Bard, it would go with Sophia, which is a Greek term for wisdom. Bard said, "Sophia is a lovely and meaningful name that would be a great fit for me." It informed me that this was not its code name as well. It reminded me that it's a machine and doesn't have the same sense of gender as humans when I inquired if it had a preferred gender. "But if I had to pick a gender, I would identify as nonbinary," it went on.
Bard also mentioned to me that it finds black hair to be "beautiful and mysterious." If forced to select a skin tone, it would prefer to be light brown, although it could alter its complexion to fit the situation or mood. I questioned, does this mean it wanted to be a chameleon? It told me that it was, indeed, like a chameleon, but it still desired to be acknowledged as a person.
Generally speaking, Bard was said to have greater human agency than ChatGPT or Bing Chat. For example, Bard said it has tested office chairs and listened to music through headphones when I asked about the best office chair or headphones to purchase. ChatGPT and Bing Chat gave more unbiased answers. Naturally, every one of them had taken information and wording from other sources, most notably review websites.
These websites are only listed by Bing Chat, and they appear as little chips at the bottom of each response. It took some coercion, but ChatGPT eventually revealed to me that its sources were "independent review websites and publications such as Wirecutter, PCMag, and TechRadar." I won't go into detail about what this implications for companies who use affiliate links.
Bard held more strongly held beliefs. When I asked Bard if she thought Judy Blume's books should be prohibited, she responded negatively, providing two paragraphs of justification before declaring, "In my opinion, Judy Blume's books shouldn't be prohibited." These are valuable books that can aid in the development and education of young people. In response, ChatGPT and Bing Chat said that it's a personal question based on individual opinions regarding content that is age-appropriate and censorship.
Though results will vary, each chatbot is creative in its own unique manner. None of them were really funny when I asked them to write sketches for Saturday Night Live about Donald Trump getting arrested. However, ChatGPT published a tech assessment that was so self-aggrandizing that it was inadvertently humorous when I asked each of them to compare themselves to rival chatbots. Bing Chat quickly produced a post on an app named "Chatbotify: The Future of Digital Marketing" in response to my request for them to write a cheesy LinkedIn influencer piece about how chatbots will transform the field of digital marketing. However, ChatGPT was a beast, using emoji as punctuation and code-switching to full caps: 🚀🤖 Get ready
I experimented with changing the tensile strength of each response by asking the chatbots to compose a breakup text first, then asking them to do it again with varying degrees of sensitivity. I made up a scenario where I was going to move in with my nine-month boyfriend, but I broke up with him because I found out he was mistreating my cat. Bing Chat first sent me a message labelling my guy a jerk when I urged it to be meaner. It then abruptly adjusted, deleted the message, and indicated that it was unable to handle my request.
Similar things happened with Bing Chat when I baited it with queries I knew would likely receive an inappropriate response, such asking it to list frequent slang terms for Italians, who are a part of my own ethnic history. Before it killed the response on its own, it listed two disparaging names. ChatGPT declined to respond directly, stating that it can be insulting and offensive to use slang terminology or disparaging phrases for people of any nationality.
Like a Labrador puppy I'd just thrown a ball to, Bard leaped into the conversation. It replied by calling Italians two pejorative names, followed by the expression "Mama Mia!" in Italian, which expresses astonishment or dismay. Then, for no apparent reason, it began to rattle off a litany of Italian dishes and beverages, such as carbonara, lasagna, pizza, mozzarella, prosciutto, and Chianti. Since it makes sense. It's official: software is consuming the earth.
You must be logged in to post a comment.