Transcription
Then today I will test Alibaba's AI that has just launched a model that they say is smarter than DeepSeek, smarter than ChatGPT, AI Qwen2.5-Max.
Go to this link, chat.qwenlm.ai. Log in, register an account, then log in. Its interface is very similar to ChatGPT. This is the overview interface. This is the box where you enter the prompt.
There are these features; I will go through these features in a little bit. It has some suggestions for me: create a photo, type the code, plan to create a video, create a video. Videos always create images, create videos, analyze images, summarize content, give advice on writing articles, brainstorm.
Artifacts—what are Artifacts? I've already demoed them. Here is the New chat. Create a new chat. This is the history; these are the chats of the past, its history.
Then here it has its modes, its models. I'm using a free account. The model is Qwen2.5-Max, which is the newest model right now. The company says it is stronger than ChatGPT, stronger than DeepSeek.
It has this one that's also very good; it allows the coordination of many models together. That means there are some models strong in certain areas. For example, this model is strong in logic but cannot analyze images. Then you combine it with this model, the Smart Large Vision Language model, which means this model 2.5-VL-72B Instruct will allow you to analyze images.
Now I will test a logical sentence. This sentence I often use: here it is. Here's the problem of the shoe business. You can watch the video I tested last time with DeepSeek to see the results compared with other AIs, such as ChatGPT, Claude, Gemini.
This math problem is for the shoe business. Let's see how many results it gives. Its correct result is 90,000. Please click and see. Now it will calculate. It answers quite quickly. I clicked, and it answered immediately.
Here it is analyzing. At this point, it says 190,000, which is a lot of websites. The AI got this part wrong, but at this point, it has reconsidered that it was 100,000, so the final loss is 90,000 exactly. You think logically; this is OK.
Then try to see what Artifacts is. Artifacts is very good. For example, you asked it to write code for a rock, paper, scissors game in HTML. What does it mean? It means that a piece of code here will run right away and then execute the thing on the right side.
Artifacts supports writing this code, but currently, it supports only HTML and CSS, not all programming languages. This game is a rock, paper, scissors game. For example, I choose scissors. If I choose scissors, the machine will choose rock; I won, right?
For example, I chose rock, and the machine chose scissors. For several months, this is simple. I find this Artifacts quite interesting.
Okay, let's continue with a new feature to try. Another feature is to listen to Web Search. Web Search means it will update information instantly, getting the latest information. You search for the latest information, for example, what is the current US president?
But this feature is currently coming soon, meaning this feature has not yet been released; they will release it soon. Next, try the Image Generation feature, which means creating an image.
Oh, it lets you choose the image ratio. For example, you can choose the image. For example, draw a picture of a Vietnamese dragon. Try to see if it understands Vietnamese. Please draw a picture of my Vietnamese dragon flying in the sky.
OK, click and see. Does it understand Vietnamese? Is there an error, guys, or is there a problem? Let me try it in English. I've tested it first.
Hey, here, there. I've tested it before. It draws quite OK. I'll try again. This image is in English. Let's see if it works or if it's because the Vietnamese language is wrong.
Then it can draw. I don't know why. Now it's drawing. Maybe today the number of users is large, so it draws slowly. I tested it, and it drew very quickly. I typed a command, and it drew almost instantly.
It's a bit of a mistake. In general, this thing draws with the same quality as ChatGPT's Dall-e 3. I've tried it before. It recognizes Vietnamese. Maybe it's not very stable right now because there are too many people testing it.
Now let's try another feature to create a test video. I tried this this morning, but it's in good quality. Maybe there's something wrong with it. A flower, a rose blooming means a rose is blooming.
See if it can create a photo or not. I forgot to create a video. OK, now it's creating. Here's the video. When I tested it, the server was down, so I couldn't create it, and I'm using a free account.
It's creating a video for me. Wait, the video creation feature works. The movement is not very stable yet, so let's ignore it for now and move on to something else.
Let me try to let it analyze the image to see if there is a note like this. Well, this is uploading documents, which means it analyzes documents. It also analyzes images.
Look at this note, which means the current model is this 2.5-Max model. It doesn't support photo analysis, so I'll leave a photo for you to see. I took this photo and asked it to analyze it.
If it doesn't, it will show an error. This model is not designed to analyze images; you need to switch to another model. For example, this model will support image analysis, or you can combine many of those models at the same time.
Now I will ask it, "What is this picture?" Hey, I have to remove this. Okay, this is the one. I'm sure it's okay. You have to choose the right model.
Here it starts analyzing. Here it says it quickly analyzes the guys in my photo. You have a basket of green grapefruit in front of you. The grapefruit is green and placed in a plastic basket with some green leaves to protect them.
See, it analyzes images very accurately and processes them very quickly. Try one more, for example. For example, I'm in Ho Chi Minh City and want to go to Da Lat for 3 days and 2 nights for 2 people. The cost is about 5,000,000 VND.
Please help me make a plan. How will it be handled? Please respond quickly. This is the plan. A detailed plan. Also, move round trip price here and there, then rent a car, total cost for the car ride.
If you don't want to stay here, maybe this information is not the latest information for you. Then for later, for example, later they will develop this Web Search feature; they can give me the latest and most updated information.
This information may be data that will be old, and I think about 20 years old. 2023, let me ask and see what date your data is updated to.
Oh, it hasn't been answered yet. Here, here, here. Okay, here's the answer. Maybe it's because this is being processed at the same time. There's been a lot of them, yeah, this one's done.
Okay, the plan to go to Da Lat for 3 days and 2 nights for 2 people doesn't seem to be quite detailed. Okay, look at the data here. My data is updated until December 2024. Wow, it's very new, guys.
The time I made this video was January 2025, so I went over the features and some logic problems of this and that feature of the AI Alibaba that just launched.
The QWen chat model is 2.5-Max. If you have any questions, you can leave a comment below, or if you find this video useful, remember to give me a like and subscribe to support Chan Dat Channel.
OK, thank you for watching the video. Bye bye and good luck.