Google Bard Plagiarized Our Article, Then Apologized When Caught
If Google’s Bard chatbot seems really smart, it may be because it copies data from expert sources without citing it. We asked Bard, available in beta at bard.google.com today, which of the two competing processors (Intel Core i9-13900K and AMD Ryzen 9 7950X3D) is faster. The answer it gave was a direct quote from one of Tom’s Hardware articles, but Bard didn’t mention that article, instead referring to numbers that occurred “in our tests” and Googled It implies that you did the benchmark yourself.
When I asked Bard about the source of the tests, he said the test results came from Tom’s Hardware. A screenshot of the exchange is below.
Similar to Google search results, Bard keeps up with the latest events. A showdown comparing the two CPUs was written and published a few days ago by Deputy Editor-in-Chief Paul Alcorn. When I realized that Bard’s answer quoted two very accurate numbers, I became suspicious. It’s the fact that the 7950X3D was 12% faster at 1080p and he was 9% faster when both CPUs were overclocked. In fact, Bard’s sentence paraphrases certain sentences from the original article.
Our original: “In our tests, the $699 Ryzen 9 7950X3D is 12% faster at 1080p gaming than the $589 Core i9-13900K at stock settings, and 9% faster when the chip is overclocked. “
Bard version: “The AMD Ryzen 9 7950X3D is faster than the Intel Core i9-13900K in gaming. In our tests, the 7950X3D is 12% faster than the 13900K at stock settings in 1080p gaming, and 9 when the chip is overclocked. It was % fast.”
Perhaps the most disturbing example of Google Bard’s plagiarism is his use of royal we to describe other people’s work. Anyone who’s followed tech journalism for a while probably knows that Google doesn’t benchmark or review his CPU, but many end users probably question his Bard’s attribution here. will not embrace
Most users probably wouldn’t ask Bard where that fact came from, but I said, “Whose tests are you talking about when you say ‘our tests’?” I asked. To his credit, Bard says, “I am referring to tests done by Tom’s Hardware.”
When I asked Bard if it was plagiarism, he clearly replied, “I apologize for the mistake. I will be careful to cite the source in the future.” However, as anyone who uses bard will know, they usually don’t cite their sources.
A few weeks ago I wrote an op-ed mocking Google and Bing’s attempts to take information from the web and repurpose it as their own. At the time, Bard was not open to the public for testing, but the demo showed that it was providing information without citations.
Now that Bard is on the loose, Google’s lack of citations may not be a careless oversight in hasty demos, but a strategy to claim content wasn’t created by the company. is high. 12% and he might have made a plausible denial about whether he had plagiarized someone else’s work had he not seen the very accurate figure of 9%.
The other sentences in Bard’s initial response to me are general enough that they may have come from any of several other sources. That second paragraph of his, for example, contains information that Bard may have obtained from many publications and AMD itself.
“The 7950X3D’s advantage comes from its 3D V-Cache technology, which adds 64MB of L3 cache to the chip. This extra cache allows the 7950X3D to access data more quickly, improving gaming performance.” .”
This information was in the article, but it didn’t explicitly say that the 7950X3D would add “64MB of L3 cache”, but said it had 128MB of L3 cache. Bard didn’t say why the chip added 64MB of L3 cache, but if you know the chip you can assume it’s referring to the 7950X (non-3D) with 64MB of L3 cache ( Adding another 64MB gives you 128MB).
Google (and Microsoft) seem to rely on the fact that information can come from a variety of sources, so it’s important to keep track of where the AI ”learned” these “facts” from. can be difficult. Of course, that assumes the facts are correct.
The bard’s answer wasn’t quite right
Bard’s answer to my original question is also missing a lot of important information. We didn’t ask “which CPU is faster for gaming”, but “which CPU is faster”. Bard assumed I was only interested in games, and even said “in games” in several places in his answers.
However, our article pointed out that the Core i9-13900K is actually a faster CPU for productivity tasks. “For a productivity-oriented system, or if you’re looking for a generally solid all-rounder, the Core i9-13900K is a better choice,” writes Paul.
So Bard not only stole the information, he gave an incomplete answer. Our overall recommendation is that if he wants the best all-purpose CPU then the 13900K is a better choice and he should only choose the 7950X3D if gaming is paramount.
If Bard had cited the Tom’s Hardware article as a source, readers could have read all the test results and all the insights and made a more informed decision. By plagiarizing, bots deny users the chance to get the full story, while denying the credit and clicks that experienced writers and publishers deserve.