Jannah Theme License is not validated, Go to the theme options page to validate the license, You need a single license for each domain name.

How Google taught AI to doubt itself


It is Platforma newsletter on the intersection of Silicon Valley and democracy from Casey Newton and Zoë Schiffer. register here.

Today let’s talk about a breakthrough from Bard, Google’s answer to ChatGPT, and how it solves one of the most pressing problems with chatbots today: their tendency to make things up.

Since the day chatbots arrived last year, their creators have warned us not to trust them. The text generated by tools like ChatGPT does not rely on a database of established facts. Instead, chatbots are predictive: they make probabilistic guesses about which words seem correct based on the enormous corpus of text on which their underlying large linguistic models were trained.

As a result, chatbots are often “wrong,” to use the industry term. And it can fool even highly educated people, as we saw this year with the case of the lawyer who submitted quotes generated by ChatGPT – without realizing that each case was fabricated.

This state of affairs explains why I find chatbots mostly useless as research assistants. They will tell you anything you want, often in seconds, but in most cases without citing their work. As a result, you end up spending a lot of time researching their answers to see if they are true, which often defeats the purpose of using them.

Google highlights the new feature allowing you to check Bard’s answers.
Screenshot: The Edge

When it launched earlier this year, Google’s Bard featured a “Google It” button that submitted your query to the company’s search engine. This made it slightly quicker to get a second opinion on the chatbot’s output, but still put the burden on you to determine what is true and false.

However, starting this week, Bard will do a little more work on your behalf. Once the chatbot responds to one of your queries, pressing the Google button will “verify” your response. Here’s how the company explained it in a blog post:

When you click the “G” icon, Bard will read the response and evaluate whether there is content on the web to support it. When a statement can be evaluated, you can click on the highlighted phrases and learn more about the complementary or conflicting information found by the search.

When double-checking a query, many sentences in the answer will be green or brown. Answers highlighted in green are linked to the cited web pages; hover over one and Bard will show you the source of the information. Brown’s highlighted responses indicate that Bard does not know where the information came from, highlighting a likely error.

When I double-checked Bard’s answer to my question about the history of the band Radiohead, for example, it gave me a lot of sentences highlighted in green that matched my own knowledge. But it also made this sentence brown: “They have won numerous awards, including six Grammy Awards and nine Brit Awards.” » Hovering over the words showed that the Google search had shown conflicting information; indeed, Radiohead have (criminally) never won a single Brit Award, let alone nine of them.

“I’m going to tell you about a tragedy that happened in my life,” Jack Krawczyk, Google’s senior product manager, told me in an interview last week.

Krawczyk had cooked swordfish at home and the resulting smell seemed to permeate the entire house. He used Bard to research ways to get rid of them, then double-checked the results to separate fact from fiction. It turns out that cleaning the kitchen thoroughly wouldn’t solve the problem, as the chatbot initially stated. But placing bowls of baking soda around the house might help.

If you’re wondering why Google doesn’t check answers like this Before I showed them to you, too. Krawczyk told me that, given the wide variety of ways people use Bard, double-checking is often unnecessary. (You generally wouldn’t ask him to double-check a poem you wrote, or an email he wrote, etc.)

An answer from Bard showing which lines could be saved with a Google search (green) and which could not (brown).
Screenshot: The Edge

And while double-checking is a clear step forward, it often requires you to pull up all those citations and make sure Bard interprets those search results correctly. At least when it comes to research, humans still hold AI’s hand as much as ours.

This is nevertheless a welcome development.

“We may have created the first language model that admits to making a mistake,” Krawczyk told me. And given the challenges of improving these models, ensuring that AI models accurately confess their errors should be a high priority for the industry.

Bard received another big update on Tuesday: It can now connect to your Gmail, Docs, Drive, and a handful of other Google products, including YouTube and Maps. Extensions, as they are called, allow you to search, summarize, and ask questions about documents you have stored in your Google account in real time.

For now, it’s limited to personal accounts, which severely limits its usefulness, at least for me. It’s sometimes interesting as an alternative way to browse the web: it did a good job, for example, when I asked it to show me some good videos for getting started in interior design. (The fact that you can play these videos inline in Bard’s response window is a nice touch.)

But the extensions also have many errors, and there are no buttons to press here to improve the results. When I asked Bard to find my oldest email with a friend I’ve been messaging in Gmail for 20 years now, Bard showed me a message from 2021. When I asked him which messages from my inbox might need a quick response, Bard suggested spam with the subject line “Hassle-free printing is possible with HP Instant Ink.”

This works best in scenarios where Google can make money. Ask it to plan an itinerary for a trip to Japan, including information on flights and hotels, and it will give you a good selection of choices from which Google can take a cut of the purchase.

Eventually, I imagine third-party extensions will come to Bard, just as they previously did to ChatGPT. (They’re called plugins there.) The promise of being able to get things done on the web through a conversational interface is huge, even if the actual experience is mediocre.

The long-term question is to what extent AI will ultimately be able to verify its own work. Today, the task of directing chatbots to the right answer still weighs heavily on the person typing the prompt. Right now, tools that push AIs to cite their work are essential. Eventually, however, we hope that more of this work will fall to the tools themselves – and without us always having to ask.


Not all news on the site expresses the point of view of the site, but we transmit this news automatically and translate it through programmatic technology on the site and not from a human editor.
Back to top button