AI Tools Huang Renxun Uses Every Day That You Probably Haven't Heard Of#
Omnivore#
It is said that AI is a tool to improve productivity, but Silicon Jun is curious whether those big shots in the tech industry who are busy every day use AI to improve efficiency themselves.
The reporter from "Wired" magazine was also curious about this question and asked Huang Renxun during an interview: "How often do you use ChatGPT or Bard? Or other products?"
Mr. Huang said, I usually use Perplexity AI.
He also gave an example. For example, when he wants to understand computer-aided drug development, he would first ask the large model some framework questions and then ask more detailed questions specifically.
What kind of company is this that makes Huang Renxun use it every day?
Perplexity AI was founded in 2022 and launched a large model-based search engine product at the end of the same year. The number of monthly active users has reached 10 million so far.
This company does not develop its own models, but only calls the interfaces of other models and makes some adjustments. In simple terms, it uses AI technology to create a Google search without ads and bidding rankings.
Perplexity AI can automatically collect, analyze, and display relevant information from the Internet and other data sources, greatly optimizing the process of information retrieval and knowledge acquisition. It can recognize and respond to more ambiguous or abstract queries, providing users with accurate answers and related links.
Is Perplexity AI really that useful?
Intuitively, the web interface of Perplexity is very clean, mainly consisting of a left sidebar and a search box on the right. Users can search for anything they want in the search box.
For example, Silicon Jun searched for "How to view Li Yizhou". Perplexity AI first listed several information sources, and a total of 25 were listed under the question about Li Yizhou.
Mainstream websites such as 36kr, Huxiu, 51tcto, Sina, and Digital China, mainstream media such as Southern Weekend, and even Q&A communities like Zhihu are all within the scope of Perplexity AI's search.
Next is the answer to Silicon Jun's search query "How to view Li Yizhou".
After a quick look, the content answered by Perplexity AI is well-organized and has corresponding viewpoints. It covers who Li Yizhou is, what happened, why it happened, and the latest situation, and marks which article the content comes from.
Users can share the content answered by Perplexity AI, and if they are not satisfied, they can ask it to rewrite.
There are also some derivative topics. For example, "Does Li Yizhou's AI course have substantial content?" and "Does Li Yizhou's AI course have practical applications?".
When users click on them, Perplexity AI will generate content just like the previous question, including information sources, content, and derivative topics.
Moreover, in addition to textual content, Perplexity AI also displays some related images. For example, pictures of the two major AI giants in the Li Yizhou incident, Li Yizhou's Douyin and Video Number homepages, etc.
In addition to hot events, Perplexity AI also performs well in areas such as recipes and travel guides. For example, when Silicon Jun searched for how to stir-fry chili and meat, Perplexity AI also provided information sources and recipes.
In addition to search, Perplexity AI also provides a discovery page, displaying recently popular news.
For example, when Silicon Jun was writing an article, Lenovo's transparent notebook had a high number of views, Meta and LG were collaborating on head-mounted devices, and Reddit's IPO was opposed by users, and so on.
Most of the popular news recommended by Perplexity AI are related to technology.
Silicon Jun speculates that Perplexity AI determines whether news is popular based on its own user searches, and most people who use Perplexity AI are more concerned about AI and technology, so the popular news tends to be more technology-oriented.
Perplexity AI also provides different information source categories for content generation.
Users can treat Perplexity AI as a professional vertical search library, focusing information sources on literature, code, Reddit forums, etc.
Silicon Jun tried asking what a transformer is in the academic field, and found that the information sources quoted by Perplexity AI are academic websites such as arXiv and Semantic Scholar.
However, Perplexity AI does not perform well in Chinese academic searches. For example, when Silicon Jun searched for the latest "sora", the search results were in Japanese romaji.
Moreover, Perplexity AI can also be used for pure writing like ChatGPT.
In addition to basic search functions, Perplexity AI can also read local files. Silicon Jun uploaded the latest "AI Mobile Phone White Paper" released by IDC for Perplexity AI to summarize the content.
Perplexity AI also offers a paid version to choose from. The difference between the paid and free versions mainly lies in the usage limits of some features, such as reading local files, choosing large models, and using APIs.
Originally, I just wanted to write a topic, but unexpectedly, after trying it out, Silicon Jun was really impressed by Perplexity AI. Mr. Huang has a good eye.
Compared to ChatGPT, which answers in a long and verbose manner, sometimes the accuracy of the information needs to be verified because the content answered by ChatGPT is generated by AI based on data.
Perplexity AI is closer to the experience of a search engine, simply rearranging the search results. Those who often use AI understand the difference.
In addition, the UI of the answer page provided by Perplexity AI is much better than that of ChatGPT, and it has a great user experience.
Recently, Perplexity AI has completed its Series B financing, with a latest valuation of 520 million US dollars and a total financing amount of over 100 million US dollars, setting a record for the financing amount of start-up companies in the search field in recent years.
This is enough to show that it has gained market recognition. So how did such a popular AI software come into being?
This may be due to the different views of Perplexity CEO Aravind Srinivas on AI products.
Aravind Srinivas was born in India and graduated from the Indian Institute of Technology. He furthered his studies at the University of California, Berkeley. Before founding Perplexity AI, he worked at OpenAI, DeepMind, and Google.
At the beginning of the startup, he claimed that his product would surpass Google search.
Last month, Jia Yangqing released a demo of LeptonAI, which is also a large model-based conversational search engine. The CEO also tweeted that he was pleased to see the big shots "referencing" and "paying tribute" to their own products.
Aravind Srinivas believes that AI startups should focus on developing products first and then models.
After all, strictly speaking, Perplexity is also a literal shell product. In the face of such doubts, Aravind said:
"It is obviously more valuable to be a shell product with 100,000 users than to have its own model but no users."
As for why Huang Renxun publicly stated that he uses it every day.
It's probably because NVIDIA recently invested in this company.
End of the article.
References:
[1] Perplexity CEO: AI startups should focus on developing products first and then models | Overseas Unicorns
Author: Dong Daoli
Editor: Zhang Zeyi
Visual Design: Shu Rui
Editor: Zhang Zeyi
About "New Geek": We focus on all aspects of the technology field with AI as the center, and strive to explain how technology is changing the world in the simplest way possible. Stay tuned.