The Heart of the Machine Original
Authors: Yang Wen, Zhang Qian
After ChatGPT became popular, many professionals learned AI with the mentality of "join if you can't beat it". But unfortunately, many people have taken a detour, and even been cut leeks by some "199" courses.
Actually, it's not that hard to learn AI. Reading articles, watching videos, and chewing on papers are all good ways to learn.
Of course, there are still many readers who complain that learning AI is like learning English back then, and there will always be some headaches "new words", reading papers is more troublesome than doing reading comprehension, and video courses don't know which one is suitable for them, wasting a lot of time.
These "stumbling blocks" encountered on the road to AI learning are due to the lack of a learning partner.
We randomly open an article in "Machine Heart", and someone may get stuck in the first sentence: What is AlphaGo? Don't worry, just a light stroke, and the learning partner will be explained clearly.
To learn a certain level of English papers, but the English level is average? Don't worry, this partner will not only translate, but also be waiting for you to ask questions.
There are so many learning videos available online, don't know which one to choose? It's okay, let's summarize the learning partner first.
How about using this learning partner, is the AI learning path much smoother?
This is a learning and working artifact that we recently discovered - the computer version of the bean bag. With it, the efficiency of the Machine's Heart Editorial Department in reading papers and learning things has been improved, and the key is that it is free, and there is no need to ask the editor-in-chief for reimbursement.
Next, let's reveal how this efficiency artifact helps us learn AI.
In the class of white prostitution at station B, can there still be a teaching assistant?
For most AI beginners, watching videos is the quickest way to get started. In this regard, Station B can be said to be a "paradise" for AI enthusiasts, with many learning resources, such as Pokémon master Li Hongyi and Li Mu, who is known as "Mu God" in the rivers and lakes, and you can find them with just a pull.
However, finding doesn't mean fitting. Many of the videos are actually difficult and require a bit of programming and math to understand. Many beginners may click on a dozen videos in a row, and experience full of frustration and waste hours before finding the right one for them.
New students in the editorial department also encountered this problem when they were learning AI. Therefore, when you discover the "AI Watch Station B" function of the computer version of Doubao, everyone is like a treasure.
Taking Li Hongyi's recent course "Introduction to Generative Artificial Intelligence" as an example, before opening it, we were worried about whether this course would be very demanding. As a result, after reading the summary of the first lecture summarized by the bean bag, this concern was dispelled.
In order to further confirm, we also launched a round of follow-up questions to Doubao:
Since the threshold is not high, let's learn to see. However, you always have to take notes when studying, and the time for workers to watch videos is squeezed out, and there is no time to organize it slowly.
When we saw the second lecture, we found that this job can be done with bean bags. Not only can it organize the main points of each lesson, but even draw a brain map. You know, Mr. Li Hongyi speaks a thick Taiwan accent, and the bean bag can be summed up so accurately, which is somewhat unexpected.
Some people may say that the content mentioned above is too simplistic, the course is 26 hours in total, and in order to save time, I just want to see what I have not learned. The PC version of Doubao makes this so easy, because whether it is the "original web page" mode or the "AI watching video" mode, it has a local summary arranged according to the timeline, and you can pick and click on the part you are interested in to watch.
Sometimes, though, these summaries may not answer our questions, or we may not have heard of a certain concept from the teacher. Before this happens, we need to switch web pages, such as opening ChatGPT to check it, and then switch back after understanding it. Watching a video often involves jumping back and forth between various web pages.
In the PC version of Doubao, these problems can be solved in one interface, because it has a "continue in chat" function, you can ask questions while watching videos, just like listening to the kind of online classes configured with teaching assistants, no matter what you ask, someone will respond to you.
We also tried Li Mu's course with this tool, and the effect was just as silky:
In general, I used to watch AI videos at station B, and after searching for eligible courses, I was very confused, I didn't know which one to watch, and my mind was full of question marks after watching it. Now with a bean bag, it takes a few seconds to understand the approximation, and in the process of reading, the problem can also be solved, and the learning efficiency is gradually improved. Friends who like to watch videos to learn can try this function.
Reading AI articles is not stuck, and you don't know where to draw
It is best to watch the video and read a few more articles to consolidate the learning results. However, many readers reported that they always encountered some troubles when reading articles, such as not knowing what a word meant every few sentences, just like doing English reading comprehension in the exam room back then. If only there was something like a "dictionary pen" it would be perfect.
On the computer version of Doubao, this "dictionary pen" is really there, which is your mouse. To read which paragraph, you only need to swipe with the mouse, and the toolbar of the bean bag will come out. And, according to what we've tested, this feature doesn't restrict browsers. In other words, no matter which browser you are reading the article, you can call up the bean bag through "scratching words", which is as if it is built into your system.
Beginner AI, being made by various professional terms, has tricks! Continue to tick the word and click "Explain" or "Search".
If I can't read the original text quoted in the article, I still mark the words and click "translate".
A section in the middle of the article is too long, and I don't have time to read it slowly. Then draw the whole section, and then click "Summary", and the highlights and essences in the original text will be laid out in detail.
As we explored, we also discovered that the toolbar is actually customized, and you can prioritize a few of the features you use frequently.
If you want to talk about the overall feeling of using this function, that is, it turns out that when you read an article, you will be confused after reading a few paragraphs, and you don't understand and don't bother to check some places, because switching pages is a distracting thing. Now, with this "dictionary pen"-like cross-stroke function, you can learn more from reading an article than you would with several of them before, something you haven't experienced before with other large AI models.
As a translator and consultant, "AI Companion" accompanies you in reading your papers
I've seen almost the introductory videos and articles, and if you still want to advance on the road of AI, you have to "go through the calamity" in the paper.
It's no secret that AI papers are notoriously tricky, and they're not only intimidatingly long, but they're also in English, interspersed with all sorts of ambiguous charts and graphs. Just looking at these dense English words, the enthusiasm of AI learners is half gone.
Coincidentally, the computer version of Doubao has an AI companion reading mode, and after a period of trial, we found that it has a "miraculous effect" for finishing papers.
Take Meta's latest Llama 3 paper as an example, we threw it to the bean bag and enabled the "AI companion reading" mode, and the full text came out with a "whoosh" translation of the translation. Therefore, if you want to learn AI these days, it is not a big deal to not pass English.
However, this paper has a total of 92 pages, and even if you read the Chinese translation, you will not be able to read it for a while. Luckily, the bean bag is waiting in the sidebar, waiting for you to assign it tasks, such as summarizing the main points of the paper.
For the details that are not covered in the summary, you don't need to look for them wide-eyed, just ask the bean bag directly.
Some people don't like to read the paper description and only care about the model score, so adjust the whole page to chart mode, and the charts they are interested in can be reached directly with one click.
In this way, a senior AI researcher reading a dozen papers a day is just a sprinkling.
Write a blog in a bean bag and don't have to be in a daze with a blank document
After reading the paper, many AI learners have the habit of writing it as a blog, which is not only a consolidation of learning results, but also an effective way to spread knowledge.
However, we've observed that a lot of high-quality blogs have bloggers who update less frequently. This may be because blogging is very time-consuming, and it is quite tedious to conceive the framework and fill in the details; Or maybe you have a desire to express yourself, but in the face of a blank document, you give up after a while.
This situation may improve after the bean buns are installed. Because we found that Beanbao not only produces an outline, but also contracts many tasks such as writing, polishing, and illustration, so that the blogger can focus on the communication of ideas, rather than some minutiae.
Let's take Llama 3's paper as an example, we feed it to a bean bag and let it come up with an outline. To make things more important, we also limited the prompt a bit, and the bean buns followed our instructions.
The skeleton of the article is supported by ridges, and the next step is to fill in the "flesh and blood". In order to get the most up-to-date and accurate information for Doubao, we ask for the "deep search" function to be enabled in the prompt, and then write in sections according to the above outline.
Worried that the article smells of AI? The word-marking function of the bean bag is once again "out of the mountain", where you are not satisfied. For example, if the increment of information is too small, use the cross-word expansion function.
If you don't think the language is dry, we can also use the custom function of marking words to make it easier to understand.
Now that the text part is done, the next step is to add pictures.
The "image generation" function of the computer version of Doubao focuses on an intimate one, and the official gives 7 style tags such as "portrait photography", "art", and "national style illustration", while providing a lot of image examples, and even attaching prompts.
We asked it to draw a "robot reading a book" in 3D. Let's take a look at the effect:
In addition, it can upload reference images. For example, let's upload a humanoid robot with a futuristic tech style, and after entering a prompt, we get 4 similar illustrations at a time.
Re-effect:
That's it for a blog post about the open-source AI model Llama3.
If you have friends who find this troublesome, you can also enable the "Help Me Write" function of the computer version of Doubao, 34 genres and styles to choose from, what official account articles, papers, reports, Weibo, Xiaohongshu...... It can all be played.
For example, if we write a Weibo about Llama 3.1, as long as we tell it the topic and reference information, it can automatically convert the language into Weibo style, and even the emojis and tags are generated, just waiting for you to copy them to Weibo with one click.
After the whole process is down, the PC version of Doubao can indeed save a lot of time for students who want to write some AI knowledge content.
Netizen: Doubao is like an operating system with AI functions
As soon as the computer version of the bean bag came out, many netizens went to try it.
Some netizens said that Doubao has a lot of ambitions, and it has also built up a browser, and it can even be said to be an operating system with AI functions.
There are also netizens who planted the "AI to watch station B" function, and they want to weld it to death on Xiaopo station.
Of course, Rome wasn't built in a day.
In June 2023, Byte began testing the AI conversation project, codenamed Grace, before the Lark model was born.
In August, Byte released its self-developed lark model, and took the first train for the filing of domestic large-scale model products, and also renamed Grace "Bean Bag" for official public beta.
After more than half a year of hibernation, at the Volcano Engine Motive Power Conference in mid-May this year, new news came from Doubao - ByteDance also changed the name of the lark model, unified called the "Doubao" model, and also released 9 large model products in one go, covering general models, role playing, voice recognition, Wensheng diagram and other fields.
However, no matter how fancy the features are, no one will buy them if they are not easy to use. After all, there are too many large models on the current market, and if you don't have any real skills, you really can't retain users.
In this regard, bean bags are confident.
According to official data, the Doubao model processes an average of 120 billion tokens of text and generates 30 million images per day.
As of May this year, the Doubao app has exceeded 100 million downloads in the Apple App Store and major Android app markets, ranking first among AIGC apps, with more than 26 million monthly active users and a total of 8 million agents.
What do these people do after downloading the bean bag? We flipped through Xiaohongshu and found that in addition to studying, there are many people who use bean bags to prepare for classes, manage money, and even start a side hustle to make money.
It is certainly inconvenient to use only a mobile phone to carry so many usages, so the computer version of the bean bag came.
From the experience of the heart of the machine, this form of bean bag is more efficient for work and study, especially in scenarios that require a computer, such as reading papers, watching learning videos, and taking notes.
As more features are developed, it is believed that this PC version will become more popular. The Heart of the Machine will also continue to be updated with the latest experience results, so stay tuned.