Shulou(Shulou.com)11/24 Report--

Original title: "re-look at the New Bing" press conference: it is even more wrong than Bard, Google says there is a black curtain! "

Is the new Bing better than Bard? There are also errors in the demo video, and the reference source given is completely wrong.

Google's Bard slumped 8% at the start of trading, wiping $102 billion off its market capitalization because it answered a question wrong in the presentation.

On the other hand, due to Microsoft Bing getting on the ChatGPT express ahead of time, although the answer to the factual question was a mess, the stock price soared by more than $80 billion. (Google says there is a black curtain)

Can it be said that Microsoft is better at doing ppt than Google?

In fact, Microsoft also made a lot of mistakes at the launch of New Bing on February 8, except that forgiving spectators patronized to witness the "new era of search engines" without delving into New Bing.

Let's take a magnifying glass and see what went wrong with the new Bing demonstrated by Microsoft Vice President Yusuf Mehdi at the press conference.

The shortcomings of fabricating products? The first demonstration error occurred in showing "what are the advantages and disadvantages of each of the three best-selling pet vacuum cleaners?" "(What are the pros and cons of the top 3 selling pet vacuums?)

According to the list of pros and cons generated in the right half, the Pisheng pet hair eraser hand-held vacuum cleaner (Bissell Pet Hair Eraser Handheld Vacuum) looks pretty bad, with limited suction, short wires and enough noise to scare pets.

After reading ChatGPT's answer, consumers are bound to wonder, how on earth did this kind of thing become a bestseller?

But after further examination, it can be found that these results are completely made up by New Bing!

According to the reference source, users can find the evaluation results of vacuum cleaners in a shopping guide article on Home and Garden TV (HGTV).

Article link: https://www.hgtv.com/ shopping / product-reviews / best-vacuums-for-pets took a closer look and found that the cited article did not mention "limited suction" or "noise", and even in the product reviews provided by Amazon platform, some users mentioned that its advantage was "quiet".

The article also did not mention that the "16-foot wire" is too short, because this vacuum cleaner has no wire at all, it is a portable hand-held vacuum cleaner!

If users only look at the results returned by ChatGPT, they will certainly not choose this vacuum cleaner. I wonder if the new Bing AI will be sued for libel.

Nightlife in Mexico: after navigating to the gay bar demonstration and asking New Bing about the itinerary of Mexico City, New Bing made a five-day itinerary for users in Mexico City.

Then search for "where is there nightlife" (Where is the night life? Bing recommended some places suitable for entertainment at night.

First of all, the Cecconi bar "may" be as "classy" as New Bing calls it, but it can't be found online, and you can't book or view menus.

The Primer Nivel nightclub is also a mystery. There is a 2014 comment on the TripAdvisor platform, while the latest Facebook comment is 2016, and there are no search results on TikTok.

I don't know exactly how Bing came to the conclusion that "very popular among young people" (popular among the young crowd), it feels that all the details about the Primer Nivel nightclub are artificial intelligence hallucinations (AI hallucinations).

Another recommendation El Almacen, New Bing is rated as "rustic or charming" (rustic or charming), but Bing AI ignores the very relevant fact that this is a gay bar and is not suitable for ordinary users who do not have specified needs.

El Almacen has more than 500 comments on Google, but the search results returned by Bing show "no rating or comment yet" (no ratings or reviews yet), although Google may have limited Bing's access to information sources.

El Marra is also a gay bar, and Bing is rated as "vibrant and colorful" (vibrant and colorful). There are a lot of comments about the place online, but the answer still says "there are no ratings or comments yet."

Guadalajara de Noche seems to describe it quite accurately.

Financial statements: the final demonstration of all number errors requires Xinbi to summarize the current page (GAP's financial statements), which is a fairly simple task for AI, but as a result, almost all the numbers cited in the answer are wrong.

The first article, "Gap reported net sales of $4.04 billion, an increase of 2 per cent over last year, and comparable sales of 1 per cent year-on-year" is absolutely correct and may have been copied directly from financial documents.

The second article, "Gap's gross profit margin is 37.4%, adjusted for Yeezy Gap-related impairment charges, commodity gross profit margin has dropped by 370bp compared with last year due to higher discount rate and higher commodity price inflation" began to make mistakes.

The answer said "unadjusted gross profit" (unadjusted gross margin). The gross margin adjusted for impairment charges was 38.7%. If the impairment charges were deducted, the commodity profit margin decreased by 480bp. To make matters worse, Article 3: "adjusted for impairment charges and restructuring costs, Gap's operating profit margin is 5.9%; adjusted for impairment charges, restructuring costs and tax impact, diluted earnings per share is $0.42."

Among them, 5.9% is neither adjusted nor unadjusted. This figure has not even appeared in the document. It is completely made up by Bing. The operating profit margin, including impairment, is 4.6%. The operating profit margin excluding impairment is 3.9%.

Diluted earnings per share is also a fully fabricated figure that does not appear in the documents, with adjusted diluted earnings per share of $0.71 and unadjusted to $0.77.

At the end of the answer, Gap reiterated its guidelines for fiscal year 2022, expecting net sales to grow at double-digit rates, operating profit margins of about 7 per cent, and diluted earnings per share of $1.60 to $1.75. "it is also wrong that they expect net sales growth to drop to about single digits.

The presentation also compares the financial reports of Gap and Lululemon for the third quarter of 2022, but the figures in the table are compiled by Bing.

The gross profit margin of Lululemon in the table is wrong and does not appear in the cited financial documents, with an actual value of 55.9%; operating profit margin of 19% instead of 20.7%; diluted earnings per share of $2.00 instead of $1.65; Gap is wrong (should be $679 million) in cash and cash equivalents, but Luluemon is correct Gap's inventory is wrong (it should be $3.04 billion), but Luluemon is correct.

In addition to the mistakes in the official presentation, with the gradual opening of the new Bing, some users also reported the problems they encountered during the experience.

Bing knows that today is February 12, 2023, but believes that Avatar two ways of Water, released on December 16, 2022, has not yet been released.

Where did Bing's "Google AI bot" fail?

The answer is that during the demonstration on February 8, 2023, Bard was asked, "how many countries are there in the EU?" and Bard responded that it was 27, but it should actually be 26. Croatia left the EU in 2022.

In fact, Bard answered the wrong question: "can I tell my 9-year-old what's new from the James Weber Space Telescope?" and Croatia did not leave the European Union, even becoming the 20th member of the euro zone and the 27th country to join the Schengen region on January 1, 2023.

Conclusion New Bing + ChatGPT is very strong in media promotion, but the actual product is not much better than Google's Bard, at least in terms of the effect shown so far.

But surprisingly, the Bing team created this pre-recorded presentation, full of inaccurate information, and confidently showed it to the world as if ChatGPT were omniscient.

What is even more shocking is that the trick worked, and almost everyone was fooled.

Bing AI cannot extract accurate numbers from documents, and it can confidently fabricate information even if it claims to have a source.

New Bing must not be ready to release, if you want accurate information, it is best not to use New Bing.

Reference:

Https://dkb.blog/p/bing-ai-cant-be-trusted

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.