Open AI’s AI-powered chat tool ChatGPT has been in the news since its release, to the point where we’re talking about direct competition with Google and predicting the Mountain View company’s near death. Yet all of this is within the normal order of the evolution of research tools. And ChatGPT is still far from competing with the flagship engine…
Over the past few weeks, Open AI’s ChatGPT tool has made headlines with its quite impressive (also in French) AI-powered question and answer system. The system certainly has its limits, but it remains efficient for a certain number of requests, especially informational ones.
Here are two examples of questions for which the ChatGPT answer is of pretty good quality:
Questions about SEO And Google, replies from ChatGPT. Source: Plenty
But it should be noted that often the answers are full of inaccuracies, even errors. Here are some examples among many others, with explanations in the legend:
Question about Spirou, created by Rob-vel and not Franquin. A bellboy is not really a waiter. Spip is a squirrel and not a hamster. The other designers didn’t wait for Franquin to die to play the character. Etc. Source: Plenty
Question about Iznogoud : here Goscinny is the writer and not the artist (although he was a writer before). But for Iznogoud he is clearly the screenwriter and not the artist. Jean Tabary is French (although born in Stockholm) and not Franco-Belgian. Source: Plenty
Question about Abraracourcix : The Gallic village chief’s wife’s name is Bonemine and not Bonnaire. Source: Plenty
Wine question Clevener : No mention at the beginning of Heiligenstein, cradle of the Kleveners from 1742 e single city producing this wine. Heiligenstein is also on the Lower Rhine and not on the Upper Rhine. Etc. Source: Plenty
We see from these few examples (which don’t want to be a scientific study on the reliability of ChatGPT but which could be repeated almost indefinitely) that this tool can hardly be used without verifying the result on other discarded sources. Reliability is clearly not there (yet)… Ask ChatGPT a question about a topic you know well and you’ll know right away. Today it is inconceivable to use ChatGPT as a lambda search engine for your daily information search needs.
However, the tool is notable for the form of the answers it returns (structured and long, descriptive sentences, without spelling errors) and in this it promises what it will give when the many current errors are corrected. Today it can be estimated that 90% of the content of the answers given is quite reliable. But when, in this answer, 10% are wrong, it does not allow us to have unalterable trust in the instrument. In this, it seems impossible to us that today it can compete with Google. On the other hand, it could also show what search engines will be like in a few years, in an obvious way.
Red alert to Google
In any case, the release of this product has caused a “red code” in Google and the leaders of the Mountain View company seem to have asked its engineers to speed up development (obviously underway for many years). around LaMDA) of an equivalent technology. But it is clear that the dimensions of effective operation between ChatGPT and Google are not the same:
- the degree of errors returned by ChatGPT today would not be accepted if Google returned them to its official engine (but that might be the case on an experimental tool that is sure to be launched in 2023).
- the amount of information produced and indexed by Google (hundreds of billions of web pages) does not have a common size with that of ChatGPT.
- Same for the volume of requests processed every second.
Sundar Pichai, CEO of Google, and Jeff Dean, Head of AI at Alphabet, stall: This is an area where we need to be brave and responsible. So we have to strike a balance (…) For research, questions of veracity are really important; and for other applications, polarization, toxicity, and safety issues are also critical After all, Sam Altman, CEO of OpenAI, says no more and recently admitted the limitations of his tool: “ ChatGPT is incredibly limited, but good enough in some areas to give the deceptive impression of being great. It would be a mistake to rely on it for anything important right now (…) there is still a lot of work to be done in terms of robustness and veracity. »
One must therefore definitely go beyond the “Wow” aspect. (which is demonstrably real) of ChatGPT and think it’s mostly a vision of what search engines will be in the medium term (knowing that on the Internet the notion of medium term is relative: here we are only talking about a few years). It’s clear that the “10 blue links” will soon disappear, in their current form, however. It is also likely that 2023 is a crucial year at this level and that Google will show us things soon, perhaps as soon as the Google I/O event next May.
That remains to be seen then how SEO will adapt to this new standard, these new standards of evolution. Why adaptation can be important and not so easy…
#ChatGPT #worries #Google #believable #News #SEO #engines
Comments
Post a Comment