"Garbage in, garbage out" is a common concept in the computer industry, and the quality of the output is basically determined by the quality of the input.

The current common large model data comes from the entire network data, and the generated content often gives us a feeling of "you don't have to say it", and the questions we ask are just nonsense.

Especially regarding knowledge searches, for people working in specific industries, they want advanced information rather than basic knowledge.

Manufacturers are somewhat aware of this problem, so they have various large industry models that are produced after "pre-training" and "fine-tuning".

But for individuals, this kind of solution specially designed for Tob obviously cannot solve the problem. Not to mention whether there is a corresponding model in the industry to which he belongs, the knowledge and abilities required to work in different positions are also different. Big difference.

There is actually a special term for this type of demand -

Personal knowledge base.

Finally, Sijijun found an AI tool that provides similar functions, ima.copliot, which was recently released by Tencent.

For example, a colleague of Sijijun who focused on researching supermarkets resigned. He wrote a lot of articles about supermarkets, with detailed and original views, and received a lot of readings of 100,000+.

Now my new colleague wants to learn his in-depth opinions. It is too troublesome to check them one by one. I can import them all into ima and I can ask them casually.

For example, if you ask ima "Why can't Hema beat Sam?", the content generated based on colleagues' manuscripts is more specific and complete than the content generated based on the entire network data.

Compared with model fine-tuning, ima provides a more convenient method.

In addition, in addition to building a "second brain" based on the knowledge base, ima has introduced AI in the three areas of "search", "reading" and "writing".

In addition to retaining common AI search functions, ima adds AI search based on local knowledge base.

AI search can directly summarize the content and present it to users, improving the efficiency of information acquisition. However, the AI ​​search model has drawbacks. The results it generates are very dependent on the content of traditional search engines, and users cannot decide which content the AI ​​selects for reference. This also causes AI search to be more like a "whole-network public opinion translator", with insufficient content breadth and depth.

Ima’s AI search based on the local knowledge base can supplement the content of the entire network, as Si Jijun mentioned above.

In the construction of knowledge base, ima provides two ways of uploading local files and crawling web pages. Enter the knowledge base in the left sidebar of ima, and users can upload files in pdf and word formats in batches. In addition, users open web pages through ima and save web pages to the knowledge base.

However, the knowledge base cannot be used after offline, and the knowledge base and the entire network content cannot be mixed. These are two problems of ima. based onIt is understandable that the AI ​​search of the entire network cannot be used, and even the locally uploaded knowledge base content has disappeared.

IMA does not have to deploy local models. It is all in the cloud. If it is not connected to the Internet, there is no AI function. Therefore, ima cannot process local knowledge after the network is disconnected. The advantage is that cloud-based AI software does not require high computer configurations, but the privacy issues of locally uploaded files are worthy of attention.

Ima’s search function is actually not much different from Tiangong, Kimi, etc. It just emphasizes the concept of personal knowledge base search. In practice, the knowledge base storage space of ima is 1GB and can be saved for a long time. However, Tiangong and Kimi need to be in the conversation. The file upload size and number are limited, and they need to be re-uploaded in a new conversation, which is more like a file. parse.

IMA's "read" function can not only analyze documents uploaded by users, but also directly analyze web pages.

The difficulty in document parsing lies in accurate identification. Si Jijun uploaded a PDF version of the robot research report, and ima was able to identify the contents of a table in the research report very well. ima can also accurately identify pictures on web pages, such as the announcement time of NetEase Cloud Music.

When reading web content, ima provides translation, AI interpretation and more functions.

In fact, whether it is ima's AI search or document analysis, similar functions have been done by others, and the note-taking function is the key to ima.

For example, Sijijun uses AI to summarize an article, and ima will provide a note button to directly record the content generated by AI. Users can also underline and record good content in ima's web page analysis, and conversations in the sidebar also support one-click recording.

The organized notes can be added to the knowledge base again for easy use again. It can be said that ima's ubiquitous "notes" function really allows the knowledge of search and analysis to accumulate.

Building a continuously polished knowledge base is the most useful function that ima brings to users.

In addition, ima also provides AI writing assistance, mind map production and other functions. Compared with other AI software, it is nothing special. It is worth mentioning that ima is very restrained and does not make an intelligent agent, ensuring the smoothness of the interface. tidy.

Users can also use common AI software and take notes to achieve the functions of ima, but ima combines them to make them more natural to use.

However, like most AI software produced by major manufacturers, ima can only use Tencent's own large models, and you need to log in to your WeChat account before use, which seems a bit closed.

ima is a good personal AI note-taking tool, but the key to this type of tool lies in the user. If you are really willing to settle down every day, just give him a word.