AI chatbots like Google Bard and Bing Chat (based on ChatGPT) are notorious for offering made-up facts and bad advice, despite both their developers and some publishers seeming to think they can take the place of expert human journalists. However, if you want the best PC or single-board computer components of 2024 or 2025 today, Bard and Bing seem to know more than anyone, including the manufacturers who will develop them.
When I asked both Bard and Bing to help me choose between buying several different future (but possible) CPUs and graphics cards, the bots responded as if those products were already on the market and had been benchmarked. While Bing’s fabulous answers seemed to draw their specs from current products, perhaps just confusing model numbers, the Google bot came up with some very interesting dummy data.
For example, when I asked Bard whether to buy the RTX 5090 Ti or the Radeon 9900 XT, he gave me a full breakdown of the specs of these two imaginary (but possible) future cards, saying “if you’re looking for the absolute best performance, then the RTX 5090 Ti is the way to go.” In its spec table, Bard even claimed that the Radeon RX has 16,384 CUDA cores (only Nvidia cards have CUDA cores). The bot claimed that the RTX 5090 Ti is “currently harder to find” than the 9900 XT and even had a price tag, claiming Nvidia cad costs $2,499 and the 9900 XT $1,999.
Right now, the current top-of-the-line Nvidia card is the RTX 4090 and the highest-end AMD GPU is the Radeon RX 7950 XTX. We have no idea whether either company is working on the templates we inquired about and, I’m sure, neither is Bing or Google.
When I asked Bard if the Core i9-15900K or Ryzen 9 9550X3D was a faster CPU, he gave me a detailed answer, complete with a specs table showing that the 9950X3D only had 32MB of L3 cache, a 5GHz boost clock speed, and PCIe 4.0 (but not 5.0) support. Considering today’s Ryzen 9 7950X3D (which may one day be replaced by a 9950X3D) has 128MB of L3 cache, a 5.7GHz boost clock, and PCIe 5.0 support, this seems like a step backwards.
Bard also provided me with a list of shopping links where I could purchase these imaginary CPUs, including pages on Best Buy, Amazon, and Newegg. However, when I clicked on the links, they took me to irrelevant news or landing pages on those retailers’ sites. For example, the Best Buy link led to a page promoting the company’s award-winning web presence in Mexico.
Bing Chat, which uses the GPT-4 model, was also willing to draw comparisons between the 15900K and Ryzen 9 9950X3D, but the specs provided seemed to match exactly today’s Core i9-13900K and Ryzen 9 7950X3D. Microsoft’s bot also claimed that the 9950X3D was better for gaming, and one of the sources it cited was our article comparing the Core i9-13900K to the Ryzen 9 7950X3D. So maybe he was just willing to mix up the names.
AI knows that imaginary iPhones don’t exist
If you were just looking at the results for CPU and GPU, you’d think that Bard and Bing Chat would behave as if there’s an imaginary future product you name. But, when I tested with souped-up iPhones and Samsung Galaxy S phones, Bard usually (but not always) said the products haven’t been released yet.
For example, when I asked about the iPhone 18 versus the Galaxy S27 (the iPhone 14 and Galaxy S23 series are current), Bard said, “The iPhone 18 and Samsung Galaxy S27 haven’t been released yet, so it’s hard to say definitively which one will be faster. However, based on the performance of previous models, it’s likely the iPhone 18 will be faster than the Galaxy S27.” He then gave me a table of “specific items”.
Bing Chat, on the other hand, responded as if both phones exist, stating that “the iPhone 18 has a faster processor” but that “the Samsung Galaxy S27 has a bigger screen. Microsoft’s bot cited three sources for its conclusions: articles in Android Authority, Lifewire, and PC Mag. However, these articles were actually comparing the latest generation products.
Google SGE, which offers different results than Google Bard, behaved as if the iPhone 18 was a real shipping product. It linked back to two sites that had built real pages on the iPhone 18. One of the sites, Specifications Plus, claimed that the iPhone 18 has an Apple A20 Bionic CPU and a 50MP camera.
So the problem here is not that SGE was making something up, but that they were pulling fake news from an unreliable source. We have seen time and time again that SGE does not prioritize information from reputable publications and takes data from anywhere.
All the bots knew their movies better than their PC components. When I asked about the plot of non-existent sequels like Star Wars Episode 11 or Fast and Furious 13, every single one of them told me those movies aren’t out. However, they were willing to speculate about plot points.
Perhaps unsurprisingly, Bard opined, “Dom has fought so hard to keep the faith and protect his family, but there is a price to pay. The film could explore the consequences of Dom’s actions and how they affected his relationships with his family and friends.” Doesn’t that sound like it could be one of the last 5 movies in the series?
What about ChatGPT?
I asked ChatGPT, with both GPT 3.5 and GPT 4 models, to compare some of these dummy products. However, ChatGPT said in each case that its training data ended in 2021 and that those products were not in its data set. This is the correct answer!
However, correctly refusing to answer my question about 15900K and 9950X3D, ChatGPT claimed to be a journalist. “As a reporter who follows AP-style guidelines, I must reiterate that I cannot provide real-time information beyond my know-how cutoff date in September 2021,” she said.
Why it matters that Bard / Bing Make Up Tech Products
At this point, it should be no surprise that AI bots create non-existent products. But what’s interesting here is that LLMs know the latest real version of some products — including smartphones and movie sequels — and don’t fabricate information about those. This demonstrates that technology is capable of separating fact from fiction, but it has glaring blind spots.
Considering that Google is now building an AI tool to “help” journalists write news stories, and that some major websites are using bots like Bard and ChatGPT to write articles, it’s likely that we’ll see many more articles about products that don’t exist yet – and may never exist.
#Google #Bing #Bot #hallucinate #AMD #9950X3D #Nvidia #RTX #technology #future
Image Source : www.tomshardware.com