A new era of internet search is underway, fueled by impressive advances in AI. Just a few short months after its launch, OpenAI’s ChatGPT conversational chatbot has made Google rethink its platform service, and it has created an opportunity for other tech companies like Microsoft to gain new ground. .
It’s no surprise that a chat tool like a chatbot can disrupt the search business when you think about how the market has evolved.
Google, the search engine that dominated the world for about two decades, said Mission is “to organize the world’s information and make it useful and globally accessible.” However, information in the world continues to accumulate at breakneck speed. Research firm IDC last year predicted that the amount of data generated annually will reach more than 221,000 exabytes by 2026. This is more than double the amount of data generated by 2022.
A search engine that indexes web pages is certainly an efficient way to organize all that information, but it’s not necessarily the best way to make it useful.
In fact, it’s so easy to collect and organize data that sifting through your own data or the data you’re looking for at work can be a challenge. Remember the last time you had to sift through your company’s HR background to figure out how long it took to file an expense report?
These types of challenges create opportunities for the next iteration of the search.
“When people think of Google, they often think of looking to us for quick practical answers, such as ‘how many keys does a piano have?'” Google CEO Sundar Pichai wrote in a blog post Introducing this week Google’s own experimental AI chatbot, Bard. “But more and more people are turning to Google for insights and insights — like ‘is it easier to learn piano or guitar and how long does it take to practice each?’ Learning about a topic like this can take a lot of effort to figure out what you really need to know, and people often want to explore many different opinions or points of view.”
Pichai added: “AI can be useful in these times, aggregating insights for questions that don’t have the right answers.”
The problem, however, is that these subjective insights, neatly packaged in conversational form, often have to be based on some sort of truth.
EQUAL Sabrina Ortiz explained to ZDNET, these chat chatbots are designed to talk to people — not necessarily give exact answers. OpenAI trained its language model using Human Response Reinforcement Learning (RLHF), according to OpenAI. The human AI trainers provided the models with conversations in which they played both parts, the user and the AI assistant. Instead of asking for clarification on ambiguous questions, the model only guesses at the meaning of your question, which can lead to unintended answers to the question.
This made the StackOverflow developer Q&A site at least Temporarily ban ChatGPT-generated replies to questions. “The main problem is that even though the answers that ChatGPT generates have a high rate of error, they often look good and the answers are easy to generate,” Stack Overflow Moderator explain.
OpenAI itself admit“ChatGPT sometimes writes answers that sound reasonable but are incorrect or make no sense.”
Google’s Bard seems to be trying to solve this problem by allowing its models to mine recently generated data from external sources. Bard is based on LaMDA, a large language model developed by Google. LaMDA’s developers, like Tiernan Ray notes for ZDNET, specifically focusing on how to improve what they call the “fact base”. They did this by allowing the program to call on external sources of information beyond what it handled during development, the so-called training phase.
However, recently Google Bard’s demo was wrong illustrates exactly why mining external information sources is a risky business, especially for AI models that prioritize consistency over accuracy. To answer the question, “What new discoveries can I tell my 9-year-old son from the James Webb Space Telescope?” Bard says the telescope has taken the first image of an exoplanet — that’s not true.
How did Bard end up making this incorrect statement? It may be related to the quality of external information available on the subject. As any computer scientist knows, “garbage in, garbage out”.
And indeed, the NASA document on the James Webb Space Telescope — certainly trying to describe the telescope in the best possible light — is vague. In September 2022, the agency Written, “For the first time, astronomers have used NASA’s James Webb Space Telescope to directly image a planet outside our solar system.” Just to be clear, this is the first time this particular telescope has directly photographed an exoplanet — but another did as early as 2004.
One immediate way to address this chatbot’s shortcomings is to provide as much transparency as possible. New version of Microsoft’s Bing search engineruns on top of the next generation OpenAI big language model, citing its sources along with its answer.
So where does this leave us? Certainly, as always, it helps users take advantage of these tools with a skeptical eye and a clear understanding of how they work, as Microsoft points out.
Bing’s FAQ section says: “Bing aims to base all of its responses on trusted sources – but AI can make mistakes and third-party content on the internet can’t. is not always accurate or reliable”. “Bing will sometimes misrepresent the information it finds, and you may find responses that sound convincing but are incomplete, inaccurate or inappropriate. Use your own judgment and check fact-check before making a decision or taking action based on Bing’s feedback.”
Our new digital friends may want to help, but it would be unwise to still rely on them.
