1. Open the chat, GPT-4 is stronger!
When I wake up, the long-awaited GPT-4 is here!
OpenAI boss Sam Altman introduced straight to the point: This is our most powerful model to date!
2. GPT-4, a list of new features
How strong is it?
GPT-4：openai.com/research/gp…
01. Large multi-modal model, accepting image input and text output
Compared with the previous ChatGPT (relying on the GPT-3.5 model), GPT-4 does not have much innovative progress in daily dialogue performance. However, as a "multimodal" model, GPT-4 not only applies to more languages, but also adds a powerful image recognition function in addition to language text.
OpenAI said that when users input text and images at the same time, it can generate text such as natural language and code, while the previous GPT-3.5 only accepted text.
According to the official demonstration, let's look at a wave of GPT-4 image generation functions:
Randomly draw a sketch of a website on paper:
Take a photo and upload it to GPT-4, it can immediately generate the HTML code of the website!
This is pretty awesome! Being able to better process image content or recognize handwritten text is a big leap for computers to better understand the way humans communicate.
02. Can reason and judge, and the ability to test is better than "doing a questioner"
Due to its strong reasoning ability, GPT-4 has a substantial performance improvement over ChatGPT on standardized skill tests.
According to OpenAI's official website, in the simulated US Bar Qualification Examination, the results of ChatGPT can only be ranked in the bottom 10% of all candidates, while GPT-4 can be ranked in the top 10%. In the more difficult International Biology Olympiad, GPT-4 can even beat 99% of human players.
03. Will "role-play" different personalities
After watching GPT-4's "Revealing Lushan's true colors, it is more "empathetic" in the fields of pictures and text than the "unhuman" AI in people's impressions before, and can even play different personalities and communicate with users.
Specifically, users can ask the model to pretend to be various roles, such as the host of a TV station conducting interviews, a fund manager facing elderly users, and so on. According to the case, GPT-4 can skillfully switch language styles to adapt to its role positioning. For example, when talking to the elderly, use a calm and slow tone, but when acting as a manager to give orders, take a decisive and resolute attitude.
summary
This article attempts to explain the diversification and superiority brought about by technological progress. Low-code tool artificial intelligence was once an efficient office tool created by technologists. Various application management systems can be developed with only a small amount of code or no code. Since most of them are developed using components and encapsulated interfaces, development Efficiency is greatly improved.
Official website: www.jnpfsoft.com/?juejin
The above is the JNPF experience official website, whether you are studying/observing/selecting models, you can use it as a reference.
As a one-stop application development platform, JNFP low-code supports the complete life cycle management of the application, that is, from the beginning of design, through development, construction, testing and deployment, to various operation and maintenance after going online (eg monitoring alarm, application up and down line) and operations (eg data reports, user feedback).
In this closed loop, various roles such as product managers, requirements analysts, architects, developers, testers, operation and maintenance personnel, operation personnel, and technical support personnel are also involved. With a low-code platform, enterprises will not need to spend a lot of manpower to develop applications, effectively avoiding this high-cost requirement and making application development easier.