AI

Perplexity AI CEO Denies Rumors Company Is Disintegrating Behind the Scenes

Perplexity AI insists that his company is working well, thank you very much.

In response to the user theory that indicates that the company suffers from a defect because it “does terrible financially”, the CEO of Aravind Srinivas took the home grass to the Hermieh – i. R/Perplexity_ai Subreddit – to record the record straight.

In the original publication that talks about garbage, the user “Noteinghapet”, according to the “campus strategy”, claimed that confusion “has stopped all financing for marketing and partnerships.” They also claimed that the Written Google competitor “is going publicly in the stock market” – a step they have proven “will happen only” if the confusion needs “huge cash injection”.

In addition, the user said they are suspected that the artificial intelligence research company may cut costs in other ways, including that it seems to belong to “automatic mode” regardless of the user’s preference-refer to the drop-down in Perplexity, which, like most other Chatbots, allows users to switch between his fastest and most essential model “deep search”.

Speculation of this type is not unusual once the company reaches an evaluation of billions, as happened in confusion-and the confusion gave people a lot of caffe through a happy search engine for hallucinations in hallucinations, which was highly accused of impersonating.

However, Srinivas chose to respond directly to the peanut gallery.

“The user should not learn a lot to use a product,” the CEO wrote. “This is the motivation with” Auto “. Let artificial intelligence decide for the user if it is a quick answers, or a supporting inquiry to search for a step in the step, or really deep search query.”

He continued, “Our goal is not to save money and fraud in any way.” “It is real to build a better product with lower chaos and simple specific options for customizing options for deprived and technically enlightened users.”

As for the optional protocol claim that the confusion is “in public”, Srinivas replied that the claim was not correct – and that the company has no plans to make a preliminary general offer (IPO) until at least 2028.

He said: “We have all the funding that we have made, and our revenues are only growing.”

Of course, Srinivas can present a comfortable vision for the future, since many believe that the artificial intelligence bubble is suitable for the explosion. However, the critics’ demands do not bear much water – and the unambiguous response is generally the best way to suppress investors, in any case.

More about artificial intelligence speculation: Amnesty International’s “Test” turns a catastrophe like Coreweave Flounders

2025-04-01 00:35:00

Related Articles

Back to top button