Image source: Generated by Unbounded AI

Scientists who once used Google Scholar every day are turning to new AI tools.

As Google Scholar celebrates its 20th birthday, Nature is worried about publishing:

Can Google Scholar survive the revolution in artificial intelligence?

Yes, Google Scholar, which has more than 100 million visits per month, is being quietly subverted by AI tools.

In 2015, the Allen AI Institute launched Semantic Scholar; in 2022, OurResearch launched OpenAlex; recently it has launched scientific research AI assistants such as consensus and Emergent Mind;...

Jevin, a computational social scientist from the University of Washington West said that even though he uses Google Scholar every day,:

In view of the arrival of some new AI tools, now may be the moment when Google Scholar is overthrown as the main search engine.

Nature: Google Scholar is being subverted by AI tools

As you can see, several Popular AI tools are doing what scientists say “we want Google Scholar to do.”

Slightly further away, the Allen AI Institute launched Semantic Scholar in 2015, and OurResearch launched OpenAlex in 2022.

The former provides additional functions such as relevant paper recommendation and AI question and answer that Google Scholar does not have, and contains more than 200 million publications.

The latter combines a database of 45 million scientific papers with an 80 billion parameter LLM (large language model) to create a new species that surpasses GPT-4o in terms of factuality and citation accuracy.

In order to explain in detail the current status of AI academic tools, we select a few to take a closer look.

Consensus is an academic search engine designed specifically for scientific research. The two founders are alumni of Northwestern University and teammates on the football team.

This product uses LLM and vector search methods to extract more than 200 million peer-reviewed papers in Semantic Scholar to provide more accurate search results.

The homepage is purple and looks quite simple:

Without further ado, let’s just take the classic paper “Attention Is All You Need” as a sample:

Initial impression: Consensus is quite convenient because it can be used without logging in. In the search results, it displays the abstract of the paper, key insights (Key Insights), conclusions, and related issues from top to bottom.

After some digging, I found two interesting onesFunction:

One of them is that Consensus seems to have put in a lot of effort to ensure the reliability of the cited information. It provides multiple dimensions of information sources, including author background, publication period, journal reputation, citation times, funding support, and article content.

In addition, by clicking "Ask this paper", you can expand the entire paper with one click and ask questions for interaction.

The question box will first appear at the bottom of the page. Click on it to see the complete paper.

However, please note that at this time it will remind you that "Ask this paper" needs to be logged in before it can be used.

After logging in and unlocking, you can ask questions about the paper, and Chinese is also supported.

Currently, the free version of Consensus provides unlimited searches and limited GPT-4 summaries, and its premium version is $8.99 per month (about 65 yuan).

Similar to Consensus, Emergent Mind is also an AI research assistant designed for computer scientists, created by independent developer Matt Mazur.

Different from general models such as ChatGPT, it is highly focused on the field of computer science and can search and summarize the latest computer science papers.

After giving it the same paper, the result seemed to be "more professional".

First of all, it will display the paper pdf at the top of the results. Click on it to see the original text, which is very straightforward.

Then the overview, key contributions, experimental process and conclusion, research significance, etc. are laid out in one line. It can be seen that it basically conforms to the narrative outline of the paper, the logic is clear at a glance, and all the key points are extracted.

Unfortunately, this function of asking questions can only be unlocked by Pro users, and most people cannot do it (doge).

But, but, there is one feature that is amazing. At the end of the answer, the relevant comment addresses of major social media are provided, including Twitter, Hacker News, YouTube, etc.

Click to jump directly and even support preview.

Moreover, how many papers the author of the paper has published have also been compiled and summarized.

Currently, Emergent Mind allows 5 free searches per day.

Another representative one is Cambrian, which focuses on the field of machine learning and was developed by the Cambrian ML team (led by Xie Saining and LeCun).

Cambrian mainly helps people quickly discover the latest machine learning (ML) research, and has collected more than 240,000 ML papers since 2017.

Still the same set of questions:

An obvious difference is that although Cambrian provides the original text we want in the first position, it also displays other related papers in sequence.

In other words, it was not limited to our assigned paper in the first place.

Overall, it is also very simple. It even provides functions such as viewing PDF, chatting with the paper, asking questions, and jumping to the open source community at the bottom of the paper.

The number of citations is also intuitively displayed in the lower right corner of the paper, making it easier to filter related papers that everyone pays more attention to.

I tried chatting with the paper. After the page was expanded, I found two more interesting little points.

One is rare in other AI tools, you can directly view the paper outline.

The other is the note-taking function. After selecting a certain paragraph or word, you can ask for an explanation or make annotations. You can also see all the annotations in the sidebar.

To summarize, the vast majority of AI academic tools currently focus on two directions:

The results display on the preliminary search page interacts with the paper

The general direction is basically the same. Below, each company is slightly different in some small details.

However, compared with similar domestic tools we usually use, there is one function that seems to have not been seen - full text translation.

For example, the Doubao plug-in allows you to compare left and right translations on the same screen.

In addition to the above forms, AI tools represented by Agents have also begun to emerge.

For example, Undermind uses a more complex Agent-based search. Although it takes longer (a few minutes) than traditional search engines, the answer quality is better.

(You need to register an account at the beginning, and you need an institution or company email address)

At a glance, it organizes the reference sources in a very clear and detailed manner, and the overall focus is on serious scientific research.

I won’t go into details about more similar tools here. We can see that, as Nature is worried about-

Google Scholar is being subverted by AI tools.

Google Scholar: About to be established for 20 years, with hundreds of millions of monthly visits

Who would have thought that Google Scholar, founded in 2004, would also break into the field of literature retrieval as a "subversive".

Before its emergence, researchers mainly relied on libraries or paid databases (such as Web of Science and Scopus) to search academic literature. Not only is it cumbersome and time-consuming, but it also requires paying a fee to access the full article.

Even the same month that Google Scholar was launched, Elsevier launched the paid service Scopus, a large database containing a large number of scientific references and abstracts.

Facing the situation at that time, Google Scholar launched the advertising slogan "Standing on the shoulders of giants".

To put it simply, the two co-founders of Google Scholar, Anurag Acharya and Alex Verstak (both Google engineers), used powerful web crawler technology to crawl variousforms of academic information.

For example, book chapters, reports, preprints and web documents, even non-English works, cover a variety of subject areas, including the natural sciences, humanities and social sciences.

According to Anurag Acharya’s vision, its goal is to:

Make scholars around the world more efficient and help everyone stand at the common frontier of science.

In November of that year, Google Scholar launched a Beta version, focusing on fast and free search.

Later, they upgraded and defeated monsters along the way, constantly enriching and improving the functions of Google Scholar.

In the early days of launch, the team mainly solved copyright issues.

They actively seek support from academic publishers and agree to have their content indexed by Google Scholar.

After a lot of efforts, JSTOR, the world's largest online journal library, finally agreed to provide users with a scan of the first page of the article. This is very important to users because they can at least see the abstract and decide whether to Further reading of the entire article is required.

In the mid-term, with the collection of a large number of high-quality academic documents, the team focused on launching new features.

For example, they have launched a personal library function that allows users to save articles of interest to their personal accounts for subsequent review and management.

In addition, a citation tracking function has also been launched, allowing users to understand the academic influence and citation trends of a document by viewing its citations.

Until recently, Google Scholar continued to optimize its search algorithms and services, introducing more intelligent features.

In short, after nearly 20 years of development, Google Scholar has become "the largest and most comprehensive academic search engine."

Some netizens even enthusiastically confessed that they have never used anything other than Google Scholar.

As of now, according to data from the network traffic statistics website Similarweb, Google Scholar has more than 100 million visits per month.

This is mainly due to the advantages that Google Scholar has accumulated over the years:

Free access, and tends to display free versions of documents in search results; wide resource coverage; advanced search options; …

In the view of Wharton School Professor Ethan Mollick, Google Scholar plays a vital role in modern research.

It's free and absolutely beats every other academic search engine, including specialized search engines.

However, due to the free nature of Google Scholar, people have always been worried that Google will one day "shut down."

In this regard, the founder of a startup company wrote an article to discuss the business model behind Google Scholar:

Google Scholar may actually have internal business value to Google

Photo His point is that Google Scholar helps Google provide insights into the latest researchintelligence and can mine its database to find potential employees.

Both factors are difficult to translate into money, but in a company the size of Google, the long-term benefits could be worth hundreds of millions of dollars annually.

But at present, it seems that these advantages of Google Scholar are being replaced by AI.

Undermind: https://www.undermind.ai/home/consensus: https://consensus.appCambrian: https://www.cambrianml.org/Semantic Scholar: https://www. semanticscholar.org/OpenAlex: https://openalex.org/Emergent Mind: https://www.emergentmind.com/

Reference link: [1]https://www.nature.com/articles/d41586-024-03746-y[2]htt ps://x.com/emollick/status/1587295619449864192[3]https://x.com/rohanpaul_ai/status/1858970411658342407