Google crawls billions of web pages, indexes that content, and then ranks it in order of the most relevant answers. It then opens a list of links to click. ChatGPT offers something even more appealing to intrepid internet users: a single answer based on its own search and synthesis of that information. ChatGPT has been trained on millions of websites to collect not only human-like chat skills, but also the data itself, as long as it goes online before the end of 2021.(1)
Over the past month, I’ve gone through my Google search history and cataloged the responses, posting my 18 Google queries to ChatGPT. Then I went back and Googled it again to refresh my memory. The bottom line is, in my opinion, ChapGPT’s answer was more helpful than Google’s in 13 out of 18 examples.
“Useful” is subjective, of course. What do I mean by the term? In this case, clear and comprehensive answers. During Thanksgiving, a question about whether condensed milk or evaporated milk is better for pumpkin pie prompted a detailed (and slightly verbose) answer from ChatGPT explaining how condensed milk makes for a sweeter pie. (Of course, it was superior.) Google basically provided a list of links to recipes for me to click on, without giving me a definitive answer.
This highlights the main threat of ChatGPT to Google. It provides a single, immediate response that does not require additional scanning of other websites. In Silicon Valley speak, it’s a “frictionless” experience, the holy grail when online consumers prefer fast and easy-to-use services.
Google has its own version of aggregated answers to some queries, but these are compilations of the top-ranking web page and are usually short. It also has a proprietary language model called LaMDA, so good that one of the company’s engineers thought the system was vulnerable.
So why doesn’t Google create custom responses to queries like ChatGPT? Because anything that prevents people from scanning its search results will hurt Google’s operating business model of getting people to click on ads. About 81% of Alphabet Inc.’s $257.6 billion in 2021 revenue came from advertising, most of which came from Google’s pay-per-click ads, according to data compiled by Bloomberg.
“It’s all about ‘Let’s let you click a link,'” says Sridhar Ramaswamy, who oversaw Google’s advertising and commerce business between 2013 and 2018 and says generative search from systems like ChatGPT will disrupt Google’s traditional service. search for business “in bulk”.
“It’s a better experience,” he said. “The purpose of Google search is to get you to click on links, ideally ads, and all other text on a page.” Ramaswamy co-founded a subscription-based search engine called Neeva in 2019, and he plans to introduce his own generative search feature in the coming months, which can aggregate web pages with notes.
ChatGPT does not disclose data sources. In fact, there’s a good chance its creators can’t say how they came up with the answers they came up with. This points to one of his biggest weaknesses: Sometimes his answers are plain wrong.
Stack Overflow, a question-and-answer site for coders, temporarily banned its users from sharing tips from ChatGPT on Monday, saying that thousands of answers posted by programmers from the system were often wrong.
My own experience confirms this. When I entered my 12-year-old daughter’s English essay question into the system, it offered a long and eloquent analysis that sounded authoritative. But the answer was also riddled with errors, such as a literary character’s parents dying when they didn’t.
What’s troubling about this flaw is that inaccuracies are hard to spot, especially when ChatGPT sounds so confident. According to Stack Overflow, the system’s responses “can usually be good.” By OpenAI’s own admission, they often sound plausible. OpenAI initially trained its system to be more cautious, but the result was that it refused questions it knew the answer to. Put another way, the result is something like a college student bluffing their way through an essay after not studying. Free thirst.
It’s not clear how widespread ChatGPT’s bugs are. One estimate making the rounds on Twitter is 2% to 5%. It could be more. This will force internet users to avoid using ChatGPT for important information. Another strength of Google: it makes money primarily from operational search queries for products and navigational searches on other sites, such as people typing in “Facebook” or “YouTube.” Such queries accounted for many of the top 100 Google searches of 2022. As long as ChatGPT doesn’t offer links to other sites, it doesn’t encroach too deeply on Google’s turf. But both problems can develop over time. As OpenAI expands its model training to more relevant parts of the web, ChatGPT may become more accurate. To that end, OpenAI is working on a system called WebGPT, which it hopes will provide more accurate answers to search queries, including source citations. A combination of ChatGPT and WebGPT can be a powerful alternative to Google. And ChatGPT already provides more accurate answers than OpenAI’s previous systems.
ChatGPT gained 1 million users in about five days. This is an extraordinary milestone: it took Instagram 2.5 months to reach this figure, and Facebook ten months. OpenAI isn’t publicly speculating about its future applications, but if its new chatbot starts sharing links to other websites, especially those that sell things, it could pose a real threat to Google.
More from Bloomberg Opinion:
Creative AI poses some messy problems: Parmy Olson
ChatGPT could make democracy more confusing: Tyler Cowen
The AI deleted my scenario. Can It Break Hollywood?: Trung Phan
(1) ChatGPT was refined from a model in OpenAI’s GPT-3.5 series of large language models trained on a mixture of text and code before Q4 2021.
This column does not necessarily reflect the opinion of the editorial staff or Bloomberg LP and its owners.
Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, he is the author of We Are Anonymous.
More stories like this one are available at bloomberg.com/opinion