With Kosmos-1, Microsoft is developing an AI system that can see
With Kosmos-1, Microsoft has developed an artificial intelligence that can see and interpret the world. Common IQ tests don’t seem to be a problem for the system.
The topic of artificial intelligence is still in vogue. Hardly a day goes by without corresponding headlines. Facebook is researching LLaMA, Google made a fool of itself with Bard and Microsoft implemented ChatGPT in its own Bind search engine. The Windows Group sets with cosmos-1 but now apparently one more.
Because the new model is an artificial intelligence that can analyze images, solve visual puzzles and recognize texts. But the model now also passes IQ tests – at least in part. Kosmos-1 is an AI system that consists of a seeing eye and a thinking brain. First results impress.
Kosmos-1: Microsoft is already achieving good results
Because after the developers subjected the system to an IQ test, the accuracy of the AI was around 22 to 26 percent. That sounds small at first, but it is a remarkable achievement for an algorithm. This in turn is made possible by training data from the Internet and information from moderated training data sets.
It is also worth mentioning that there is no dependency on OpenAI and ChatGPT. In order for the machine eye to be able to interpret the data correctly, Microsoft first converted images into character strings. The company then fed the algorithm with the information and taught it to interpret the world. The results already beat chance.
The AI model should also be able to speak in the future
Microsoft says that the model performs significantly better than alternatives that are already available. Because current systems usually do not manage to beat chance. So you might as well roll out results. Kosmos-1 therefore seems to be quite mature at first glance.
In the future, however, Microsoft would like to significantly increase the model size of the AI and teach it how to interpret language. However, it is still unclear exactly where the system will be used. Perhaps Microsoft is trying to make itself a little more independent of OpenAI for the future.
Also interesting:
