The Chicago Journal

Microsoft AI actually had errors in demo last week

Microsoft Last week, two key tech firms engaged in a contest to showcase their progress in artificial intelligence.

Early versions of the AI-powered search engines from Google and Microsoft were on exhibit at two different gatherings.

The fact that Microsoft’s event took place a day before Google’s, whose disastrous failure resulted in a drop in Alphabet’s shares, gave them an edge.

More than a million people attempted to join and utilize Microsoft’s new tool in the first 48 hours as a result of the presentation’s widespread notice.

According to Microsoft CEO Satya Nadella, the industrial revolution may have “brought to knowledge work” technology.

Its AI validated concerns regarding accuracy, therefore the achievement was not without problems.

The demo

In the experiment, the ChatGPT-inspired AI system of the Bing search engine examined financial data, including that from Gap and Lululemon.

The chatbot’s results showed a fault in that it overlooked certain data when compared to the actual reports.

Additionally, the viewers noticed that some of the figures looked to be false.

Dmitri Brereton, an independent search researcher, wrote the following on Monday in a post on Substack:

“Bing AI got some answers completely wrong during their demo. But no one noticed. Instead, everyone jumped on the Bing hype train.”

Brereton also drew attention to apparent factual flaws with the demo’s irregularities in the specifications of the vacuum cleaner and the travel plans to Mexico.

The researcher said he wasn’t purposefully trying to find errors.

Brereton didn’t become aware of the errors until he tried to compare the Microsoft and Google AI reveals in his writing.

Meanwhile, AI experts referred to the errors as “hallucinations.”

The tendency of tools to create data based on in-depth language models is referred to as hallucinations in artificial intelligence.

When Google put on a similar event, its AI system also produced factual errors that were easily detected.

AI and search engines

Google and Microsoft are working to integrate new types of generative AI into their search engines as a way to showcase their progress.

The rivalry intensified after OpenAI debuted ChatGPT in November.

Microsoft granted OpenAI billions of dollars.

Due to billion-dollar valuations in private financing rounds, several companies, such Stability AI and Hugging Face, had great growth during this time.

Read also: BuzzFeed wants to take advantage of AI

Google, on the other hand, was wary of incorporating AI-generated answers into its search engines because it needed to maintain its image for providing the best results.

The business had safety worries as well.

However, during its introduction, Microsoft emphasized the potential short-term exposure of portions of the public to its technology.

“I think it’s important not to be in a lab,” Nadella added. “You have to get these things out safely.”

Demo problems

When Bing debuted their AI solution, there were problems with the corporate profitability outcomes.

Yusuf Mehdi, a Microsoft marketing executive, went to the Gap investor relations website and gave Bing AI instructions to emphasize the company’s November third-quarter numbers.

The AI-generated results showed that the summary included the following errors:

  • The stated gross margin for Gap was 37.4%, but once Yeezy was dropped, it rose to 38.7%.
  • The company’s operating margin was 4.6% as opposed to 5.9% (Gap’s report omitted this information).
  • Diluted earnings per share (adjusted) were $0.71 as opposed to the $0.42 that was reported. The report from Gap showed an adjusted income tax benefit of nearly $0.33.
  • According to Gap, net sales would decrease sequentially in the fourth quarter by the mid-single digits, which will result in lower revenue for the whole year. The operating margin is not given a forecast, however.


Microsoft is aware of the errors and expects the Bing AI to keep making them.

“We’re aware of this report and have analyzed its findings in our efforts to improve this experience,” said a Microsoft spokesperson.

“We recognize that there is still work to be done and are expecting that the system may make mistakes during this preview period, which is why the feedback is critical, so we can learn and help the models get better.”

Opinions expressed by The Chicago Journal contributors are their own.