A.I. Chatbots Are Being Used Incorrectly. What to Tell Them Is Here. - Liberty

Monday, July 31, 2023

A.I. Chatbots Are Being Used Incorrectly. What to Tell Them Is Here.

 We may direct chatbots toward high-quality data to reduce the generation and dissemination of false information.


A.I. Chatbots Are Being Used Incorrectly. What to Tell Them Is Here.
Ariel Davis

Liberty CNN, One ultimately encounters what is known as hallucinations, the propensity for artificial intelligence to invent information, while using A.I.-powered chatbots like ChatGPT and Bard — wow, they can compose essays and recipes! — who finally fall victim to their seduction.

The chatbots can't help but make mistakes since they rely on information gleaned from all across the internet to estimate what to say. It may be a great buzzkill when they fail, as when they post a cake recipe with outrageously incorrect flour quantities.

But as more and more commonplace digital solutions include A.I., it's critical to understand how to put technology to work for us. After evaluating hundreds of artificial intelligence (A.I.) products over the last two months, I concluded that the majority of us are inefficiently utilizing the technology, mostly due to the tech firms' bad guidance.

When we ask the chatbot questions and then trust that the responses they generate on their own are accurate, which is how they were intended to be used, the chatbots are the least helpful. However, A.I. can do useful jobs with a high degree of accuracy when instructed to utilize data from reliable sources, such as reliable websites and research articles.

Sam Heutmaker, the creator of the artificial intelligence start-up Context, claimed that if you give them the correct knowledge, they can do amazing things with it. However, 70% of the results you acquire on their own won't be reliable.

The chatbots produced understandable responses and helpful advice with the simple change of telling them to deal with certain data. That changed me over the last several months from a pessimistic skeptic of artificial intelligence to a passionate power user. Because the suggestions came from my preferred travel websites, the trip I took utilizing an itinerary created by ChatGPT was successful.

Reducing the creation and dissemination of false information may also be accomplished by directing the chatbots to specified, high-quality sources, such as websites from reputable media organizations and academic journals. Let me describe some of the methods I used to get assistance with my research, cooking, and trip arrangements.


Meal preparation

Chatbots like ChatGPT and Bard can create recipes that seem excellent in principle but don't function. An early A.I. model developed recipes for a Thanksgiving feast in November that included recipes for a thick cake and an exceptionally dry turkey as part of an experiment by The New York Times Food desk.

Additionally, I saw disappointing outcomes while using seafood recipes created by AI. However, when I experimented with ChatGPT plug-ins—, outside applications that integrate with the chatbot—that situation changed. (Users of plug-ins, which may be enabled in the settings menu, must be subscribers to ChatGPT4, the most recent version of the chatbot, who pay $20 per month for access.)

I chose Tasty Recipes from the list of available plug-ins for ChatGPT since it gets its information from the well-known media outlet BuzzFeed's Tasty website. I then requested that the chatbot use website recipes to create a dinner plan that included seafood dishes, ground pork, and veggie sides. A menu of everything-in-the-fridge spaghetti, grilled tofu tacos, and lemongrass pork banh mi was suggested by the bot, along with links to the recipes on Tasty.

I utilized a plug-in called Link Reader to create meal plans utilizing recipes from various magazines, like Serious Eats, by pasting in a web link. The chatbot instructed me to visit the websites to read the recipes and gathered information from them to generate meal plans. It was more effort, but it was better than the meal plan the A.I. generated.



Research

I used ChatGPT and Bard to refresh my recollection of earlier games by describing their narratives while I was doing research for an essay on a well-known video game series. Important aspects regarding the narrative and characters in the games were incorrect.

After experimenting with several different A.I. technologies, I concluded that it was critical to focus on reliable sources and immediately confirm the correctness of the data while doing research. I ultimately came upon a tool that does that: A well-liked free online application among academic academics and attorneys called Humata. AI.

The software allows you to upload documents, like PDFs, and then a chatbot provides answers to your inquiries regarding the content while highlighting important passages in the uploaded text.

I used PubMed, a government-run search engine for scientific publications, in one test to post a research article. I would have needed hours to summarize the large paper, but the program did it in minutes. I quickly skimmed the highlights to make sure the summaries were correct.

The program was created by Cyrus Khajvandi, one of the founders of Austin, Texas-based Humata when he required assistance reading complicated scientific publications as a researcher at Stanford, he said. According to him, the issue with chatbots like ChatGPT is that since they depend on antiquated conceptions of the web, the data may be devoid of essential context.



Travel Preparation

A Times travel journalist recently requested ChatGPT to create a trip plan for Milan. Among other mishaps, the bot directed the writer to visit a key area of the city that was vacant since it was an Italian vacation.

When I asked for a trip schedule for my wife, myself, and our dogs in Mendocino County, California, I had greater success. I requested ChatGPT to include ideas from some of my favorite travel websites, including Thrillist, which is owned by Vox, and The Times's travel section, much as I did while preparing dinner.

The chatbot created an itinerary with dog-friendly eateries and activities in a matter of minutes, including a farm with wine and cheese pairings and a train to a well-known hiking route. This saved me a lot of time preparing, and the dogs had a great time as well.



Bottom Line

Although Google and OpenAI, which closely collaborates with Microsoft, claim to be striving to eliminate hallucinations in their chatbots, we may already profit from A.I. by controlling the data that the bots use to generate their responses.

To put it another way, according to Nathan Benaich, a venture investor who invests in A.I. businesses, the fundamental advantage of training computers with massive data sets is that they can now utilize language to replicate human thinking. He said that combining such capacity with high-quality information is a crucial first step for us.

Comments


EmoticonEmoticon