Last week, Zuckerberg opened the Meta AI chatbot to the public.
Users can summon it as a search or chat assistant on WhatsApp, Instagram, Facebook and other Meta-based social software. However, at present, only text can be entered to generate answers and pictures, except for a group of celebrity character custom robots, there is nothing too outstanding about it, and it has not caused much community response.
However, one person broke the defense as soon as possible and slammed Meta into a slash.
That person is Aravind Srinivas, CEO of Perplexity. On April 12th, he suddenly posted on the X platform, paying tribute to his company's designer, but the picture was the chat interface of Meta AI.
Aravind Srinivas then explained with a screenshot of the interface of Perplexity that Meta AI "plagiarized" its own products from the eye-catching logo, slogan, and inspiration tips with emojis.
Opinions in the comment area are divided into two factions, some say that Xiaozha can indeed win the Oscar by copying other people's things.
"It's the same as copying my final homework for senior year. ”
"Congratulations to the designers at Perplexity for creating a new industry template. ”
Other netizens think this is a bit far-fetched, "Then you have to say this, Perplexity is square in shape, and meta is arced..."
Someone also went back: "You mean pay tribute to the designers of ChatGPT, and although Perplexity came up with the good idea of moving the prompt question to the middle of the screen, congratulations." ”
Arvind Srinivas那边似乎并不服气,凌晨四点又继续发文说:
"Plagiarizing good stuff is the right thing to do. We should all do that. We borrowed a few things from Google (knowledge cards, components). Google also borrowed a few things from us (follow-up questions, attribution), and OpenAI borrowed from both (highlighting links, suggesting questions, quick query refactoring). Grok took away the news exposure feature that was first implemented on Perplexity Discover. When adopted by other smart people, it just confirms that these ideas are good. ”
The accompanying images are Grok on the left and perplexity on the right:
Good guys, this time Google, OpenAI, and Grok have all been shot by insinuation - all the companies that make chatbots have made Perplexity the first target of "plagiarism".
As a result, the eldest brother was directly pierced by a sentence: "We didn't know that there was such a thing as Perplexity Discover." ”
"Anyone can copy an idea and add your own twist. The world is like that, it's in nature, and we can certainly do the same. ”
Netizens told Srinivas to stop complaining. Isn't Discover essentially a News Feed? It wasn't invented by Perplexity, it's been in Google's search engine for more than a decade.
Speaking of which, this is not the first time that Aravind Srinivas has "accused" others of borrowing the fruits of his labor.
In January of this year, Jia Yangqing released a demo of a conversational search engine based on the LeptonAI cloud platform on X, which easily achieved a Perplexity-like effect with 500 lines of Python code. The backend of this "Lepton Search" is the Mixtral-8x7b model, which is connected to the Bing Search API, and users can enter questions and return answers, cited sources, and related questions.
With this demo, LeptonAI shows how easy it is to build an AI application today. In other words, it is equivalent to "peeling" the Perplexity product. Its front-end design does look fancy, but the technical threshold is actually quite easy to achieve.
This quickly attracted Srinivas to parachute in, retweet Jia Yangqing's tweet and said in his usual "euphemistic" tone: "It's great to see Perplexity become a benchmark for future fundraising actions, including former Meta and Alibaba executives! Perplexity's influence is not limited to the product itself, but radiates to the entire technology ecosystem and industry development, which is exciting!"
Fans on both sides also had their own opinions at the time. Proponents of Perplexity think that Lepton is plagiarism, opponents think that Lepton is just proving his ability through demonstrations, and that you Perplexity does not have the knowledge to search for patents, which is really an overreaction.
Jia Yangqing did not show weakness, saying that he was inspired to build this demo from a discussion about whether the effect of RAG came from search or a large model when he had a coffee with Microsoft technical experts@youwu_5u and hit back head-on with all the code being open source.
So now that Srinivas is "marching" Meta and Grok, we can also see that Perplexity is really in a state of combat readiness all the time, and it feels like the whole world is copying it!
But back and forth, it seems that the accusation is always inseparable from the plagiarism of "creativity, interface design, and function", and nothing else. Some netizens ruthlessly exposed, "Is this because, Perplexity understands that he really has nothing to copy except UI". There are even people who can generate similar front-end code with any AI chatbot, and with some ideas of their own, they can polish a whole new shell.
In terms of workflow, Perplexity first takes user input, reconstructs the query based on a real-time index, and then hands off the task of answering the question to the large language model, which is required to read all relevant links, extract relevant paragraph integrations from them, and return them to the user. In essence, it still relies on the search APIs provided by Google and Bing and LLMs such as GPT-4 and Claude 3.
As an application company and API interface provider, Perplexity does not have its own basic model, and the default two free self-developed models are fine-tuned from GPT, so there is no solid and insurmountable technical moat. Optimization and innovation in the search experience is the magic weapon that Perplexity initially attracted users, and it has naturally become something they hold in their hands and desperately defend.
Srinivas released the first whiteboard on the Perplexity product idea in X
Although for platform companies, this kind of "borrowing" from each other is all too commonplace, and there are many more examples of ride-hailing apps Uber and Lyft, food ordering apps DoorDash, Uber Eats and Postmates, and domestic Meituan and Ele.me. However, the young Perplexity has not yet formed a stable and extensive customer base, and its functions are relatively simple and have not yet fully blossomed, and the anxiety of worrying about being replaced at any time is even more serious when the cash flow needs to be supported by selling advertising. You even need to take every opportunity to "touch porcelain", even if it seems inevitably a little stressful.
Another undeniable reality is that there is a clear valuation gap among the various AI startups in the market. OpenAI, Anthropic, Cohere, Scale AI, etc., which have developed basic large language models or have core technologies, are always in the leading first echelon, while platform companies such as Perplexity or Poe are difficult to break through once they reach a certain position.
Poe is now rushing to integrate the latest large models on the market as quickly as possible, as well as the Poe bot creator sharing economy model, plus Quora as a backup, it can be regarded as finding its own way. For Perplexity, a company that currently relies only on conversational search engines, it must also be admitted that once an idea is made public, it is no longer a secret, and everyone has the right to learn from and optimize.
Finally, to be fair, to this day, as a Perplexity subscriber, its ability to organize trivial information and its accuracy in eliminating AI illusions are still the points that attract me to pay. Doing it the fastest way proves that the team is sharp and good enough, and doing the best requires the wisdom of continuous study. Rather than staring at competitors complaining online, Perplexity should let go of the obsession with ingenuity, consolidate its advantages and continue to focus on the market, polish the product, and prove itself with the next refreshing feature.