This week we had China’s AI lab DeepSeek making headlines for different reasons. But that’s not the only company causing a stir in the world of tech.
Alibaba just rolled out its latest family of AI models called Qwen2.5-VL. This has the capability of carrying out a series of picture and text analysis functions. The models can parse files, comprehend videos, and even serve as a counter for objects inside pictures. This comes with full control of the PC.
In case you’re wondering, it’s being compared for its similarities to OpenAI’s Operator which was launched recently as well. As per Qwen, the model has the capability of beating out GPT-4o from OpenAI as well as Claude 3.5 Sonnet from Antropic. It’s even said to go head to head in design and functionality with Google’s Gemini 2.0 Flash on a series of prompts like math, analysis, and video comprehension including answering queries.
The latest model is up for grabs for testing through the Qwen Chat platform and you can even install it from Hugging Face. So far, it’s making a revolution with its services for chart analysis, graphics, and extracting data through invoices and other forms. If that’s not enough, it can analyze and comprehend videos lasting long hours.
It also has the tendency to highlight IPs found in movies and television series and a host of other products. This means the models could be trained partly on the likes of projects with copyright protections.
However, the fact that it’s made by a company in China means it does come with several restrictions on topics for discussion. For instance, a chat on a politician’s mistakes might give you an error so you need to be selective on your prompts.
It’s not a major surprise as Chinese internet regulators benchmark the models to make sure the replies are a reflection of core socialist thinking. This is why you’ll more often find systems declining replies to certain topics that raise concerns amongst regulators.
Another interesting feature of the latest Qwen model is the capability to interact with software found on computers and smartphones. One video shared on X showed how it could launch the booking app for making flight reservations for two Chinese cities.
Meanwhile, less sophisticated model systems are also up for grabs under a certain license that’s permissive. The flagship model is under the custom license of Alibab which needs companies as well as developers having 100 million MAU to request permission from the company before deploying it on a commercial level.
Read next:
• Elon Musk’s X Launches Digital Wallet and Peer-To-Peer Payments by Visa
• Meta in Panic Mode? DeepSeek’s AI Breakthrough Sends Shockwaves Through Silicon Valley
Alibaba just rolled out its latest family of AI models called Qwen2.5-VL. This has the capability of carrying out a series of picture and text analysis functions. The models can parse files, comprehend videos, and even serve as a counter for objects inside pictures. This comes with full control of the PC.
In case you’re wondering, it’s being compared for its similarities to OpenAI’s Operator which was launched recently as well. As per Qwen, the model has the capability of beating out GPT-4o from OpenAI as well as Claude 3.5 Sonnet from Antropic. It’s even said to go head to head in design and functionality with Google’s Gemini 2.0 Flash on a series of prompts like math, analysis, and video comprehension including answering queries.
The latest model is up for grabs for testing through the Qwen Chat platform and you can even install it from Hugging Face. So far, it’s making a revolution with its services for chart analysis, graphics, and extracting data through invoices and other forms. If that’s not enough, it can analyze and comprehend videos lasting long hours.
It also has the tendency to highlight IPs found in movies and television series and a host of other products. This means the models could be trained partly on the likes of projects with copyright protections.
However, the fact that it’s made by a company in China means it does come with several restrictions on topics for discussion. For instance, a chat on a politician’s mistakes might give you an error so you need to be selective on your prompts.
It’s not a major surprise as Chinese internet regulators benchmark the models to make sure the replies are a reflection of core socialist thinking. This is why you’ll more often find systems declining replies to certain topics that raise concerns amongst regulators.
Another interesting feature of the latest Qwen model is the capability to interact with software found on computers and smartphones. One video shared on X showed how it could launch the booking app for making flight reservations for two Chinese cities.
Meanwhile, less sophisticated model systems are also up for grabs under a certain license that’s permissive. The flagship model is under the custom license of Alibab which needs companies as well as developers having 100 million MAU to request permission from the company before deploying it on a commercial level.
Read next:
• Elon Musk’s X Launches Digital Wallet and Peer-To-Peer Payments by Visa
• Meta in Panic Mode? DeepSeek’s AI Breakthrough Sends Shockwaves Through Silicon Valley