A new chatbot from OpenAI took the internet by storm this week, dashing off poems, screenplays and essay answers that were plastered all over Twitter by the breathless technoratti. Though the underlying technology has been around, this was the first time OpenAI has brought its powerful language-generating system GPT3 to the masses, prompting a race to give it inventive commands. Beyond gimmicky demos, some people found practical uses for ChatGPT, including programmers who are using it to draft code or spot errors. But the system’s biggest utility could be a financial disaster for Google by supplying better answers to queries we currently put to the world’s most powerful search engine.
Google works by crawling billions of web pages, indexing content, ranking it by relevance, and then spitting out a list of links to click. ChatGPT offers something more tantalizing: a single answer based on its own search and synthesis of data. ChatGPT has been trained on millions of websites to glean not only the skill of holding a humanlike conversation, but information itself (stuff posted online before late 2021).
I went through my Google search history of the past month and put 18 of my Google queries into ChatGPT. I then went back and ran the queries through Google once more, to refresh my memory. The result, in my judgement, was that ChapGPT’s answer was more useful than Google’s in 13 cases. ‘Useful’ is of course subjective. What do I mean? Answers that were clear and comprehensive. A query on whether condensed or evaporated milk was better for pumpkin pie sparked a detailed (if verbose) answer from ChatGPT that explained how condensed milk would make sweeter pie. Google gave me links to recipes with no clear answer.
That underscores ChatGPT’s prime threat to Google. It gives a single instant response that requires no further scanning of websites, a ‘frictionless’ experience, something of a holy grail for such tools.
Google has its own version of summarized answers to some queries, but they are compilations of top-ranked web pages and typically brief. It also has its own language model, called LaMDA, which is so good that a Google engineer thought it was sentient. So why doesn’t Google cough up singular answers, like ChatGPT? Because anything that stops people from scanning search results will hurt Google’s transactional business model of getting people to click on ads. Some 81% of Alphabet’s $257.6 billion revenue in 2021 came from advertising, much of that being Google’s pay-per-click ads. “It’s all designed with the purpose of ‘Let’s get you to click on a link’,” says Sridhar Ramaswamy, who oversaw Google’s ads and commerce business between 2013 and 2018, and who says that generative search from systems like ChatGPT will disrupt Google’s traditional search business “in a massive way.” “The goal of Google search is to get you to click on links, ideally ads, and all other text on the page is just filler,” said Ramaswamy, co-founder of a subscription-based search engine called Neeva.
ChatGPT doesn’t reveal its sources. In fact, there’s a good chance its own creators can’t tell how it generates its answers. That points to one of its big weaknesses: Sometimes, its answers are plain wrong. Stack Overflow, a site for coders, temporarily banned its users from sharing advice from ChatGPT, pointing to errors.
My own experience bears this out. When I put my 12-year-old daughter’s English essay question into the system, it offered a long and eloquent analysis that sounded authoritative. But it was also riddled with mistakes, for instance stating that a literary character’s parents had died when they hadn’t. What’s disturbing about this flaw is that inaccuracies are hard to spot. It often sounds confident and plausible. OpenAI had initially trained it to be more cautious, but that made it decline many questions. By going the other way, the result is something like college students bluffing their way through class with fluent hogwash.
How common are ChatGPT’s mistakes? One estimate doing the rounds is a rate of 2% to 5%. It may be more. That will make internet users wary of using ChatGPT for important information. Another strength for Google: it mostly makes money on transactional search queries for products and navigational searches to other sites. So long as ChatGPT doesn’t offer such links, it is not encroaching too deeply on Google’s turf.
But both could evolve. ChatGPT could get more accurate as OpenAI finesses its training. ChatGPT amassed 1 million users in about five days. That is an extraordinary milestone. It took Instagram 2.5 months to reach that number, and ten months for Facebook. OpenAI isn’t publicly speculating about its future applications, but if its new chatbot starts sharing links to other websites, particularly those that sell things and make money, that could spell danger for Google.
©bloomberg
Download The Mint News App to get Daily Market Updates.
