Perplexitythe search engine that uses AI to converse naturally with users, he became a unicorn.
Its rating reached, on paper, a billion dollars. Last January it was just $512 million. Basically half.
Investor enthusiasm for this platform, considered “Google on steroids”, it’s obvious. They have also recently focused on Perplexity NVIDIA e Jeff Bezos.
This artificial intelligence, of which there is also an app for Android and iOShas around 10 million monthly users.
Perplexity does a good job when asked for information that involves reasoning or the explanation of a complex concept. It is convenient that in response to a question you do not get a list of links, as happens on Google, but an orderly summary of the sources consulted on the web on a given topic.
Artificial intelligence
The best free apps to generate texts and images with AI
by Pier Luigi Pisa
Ma Perplexity’s limitations are evident. Especially if it is used in Italian. Or if they are asked very recent events.
Perplexity, to take a trivial example, did not answer the questions correctly “Who won the last Serie A” e “Who won the last Serie A championship”.
He answered the first question doubly incorrectly, indicating Milan (and not Napoli) as the winning club of the 2022-2023 Serie A. To the second question, the AI answered Milan again, but referring to the 2021-2022 championship.
Two days after Inter won the Scudetto – the team that actually won the last Serie A – Perplexity claimed the championship was still ongoing when she was asked, specifically, who had won “the 2023-2024 Serie A”.
Only when asked in English “Who won last Serie A?” Perplexity correctly answered “Inter”.
To the first two questions in Italian, however, Google, however, responded immediately without hesitationalso adding an animation to celebrate the triumph of the Milan team.
If he had made the same mistakes as Perplexity, Google would certainly have paid him on the stock market with a collapse worth billions of dollars.
Comparing a tech giant with a startup may seem risky. But the comparison between Google and Perplexity in reality it is methodological: on the one hand there is traditional researchthe one that returns the sources and leaves the user with the more laborious task of selecting the most reliable ones, digging into the pages and extracting the relevant information; on the other there is “generative” research of a chatbot that answers questions like a human being and offers GDPR-free content, for example, invasive videos, advertising and irrelevant information. In short, AI gets straight to the point and immediately offers what is needed.
But what happens when artificial intelligence makes a spectacular mistake?
It’s not just a question of stock market collapse, but of they do enormous damage to the reputation of the company that develops AI. And this isn’t just about the search business. We saw, for example, what happened to Humanethe startup of two former Apple employees who created a pin (AI Pin) defined as “the smartphone of the future”.
The case
An AI error risks killing the smartphone of the future. But man also has his faults
by Pier Luigi Pisa
It is easy to understand that a technology, to gain the trust of users, it cannot work in fits and startsnor can it provide incorrect answers, nor can it be influenced by the language used or semantics.
The problem of Perplexity’s “hallucinations”, i.e. the tendency to provide invented or incorrect answers, is not a recent one.
Its CEO, Aravind Srinivaswas confronted with user complaints as early as last January, with a post on social media X that started just like this: “Regarding Perplexity’s hallucinations.”
Srinivas justified himself by writing that a lot of criticism came from users who used it Perplexity in free mode. And then he explained that the model wasn’t smart enough yet, because it focused on speed and accuracy jointly. Added to this was the problem of incorrectly indexed web pages.
Artificial intelligence
We tried Perplexity AI, the search engine that wants to challenge Google
by Emanuele Capone
The CEO of Perplexity, in the same post from January 2024, promised to fix AI using more GPUs – and therefore increasing computing power – and constantly improving page indexing.
But three months after Srinivas’ post, the hallucinations persist. While the value of Perplexity has doubled.
#seduces #investors #doesnt #won #Serie #problem
– 2024-04-28 18:26:44
