Google continues to push the envelope with its advancement in the field of AI thanks to divisions like DeepMind that have been entrusted with building new AI solutions. And recently the large language model (LLM) from DeepMind was able to solve a complex Maths problem which many thought was unsolvable.
The interesting thing about the results is that the LLM was not fed with data related to the problem, in fact, the AI model was able to derive new information which wasn’t there earlier, as mentioned by Pushmeet Kohli, vice-president of research at Google DeepMind, quoted in this report.
The DeepMind team has a new tool called FunSearch which has made solving these problems possible, and show people that AI can work out of the realm of the training data that are fed to the LLMs.
It is mentioned that FunSearch is in the business of solving fundamental Maths and computer science problems and DeepMind has it this purpose using its AI tech. The intriguing thing about FunSearch is that it uses existing data that is a mix of incorrect or other forms of rejected answers.
It seems the whole model has formed out of random observations and hypotheses which the researchers at DeepMind didn’t know whether it will work out or not. Many at Google’s AI division feel FunSearch has a lot more potential and advantage over AlphaTensor which gives it more room to evolve and grow as a language model in the near future.
The DeepMind team decided to test the capability of FunSearch further by throwing another complex Maths problem its way and the results were completely surprising to everyone at DeepMind, as FunSearch was able to solve it faster than most other models had done so far.
Google is bringing more AI tech to the market-ready product, and Gemini is getting its due attention this week as the AI model is now available for people on smartphones, their PCs through Bard and enterprises as well.
Comments
0 comment