Google's AI Learns to Improve Coding Skills by Incorporating Other AI Models

JJohn January 6, 2024 7:01 AM

Google's AI research teams at Google Research and Google DeepMind have devised a method to enhance the capabilities of a large language model (LLM) by supplementing it with other language models. The approach not only enhances the performance of existing tasks but also enables the LLM to take on new tasks that were previously unachievable.

Augmenting LLMs with specialized models

In a significant step towards AI enhancement, Google Research and DeepMind teams experimented with Google's PaLM2-S LLM, comparable to OpenAI's GPT-4. They discovered that this large language model could learn new 'tricks' and improve performance when supplemented with smaller, specialized language models. The improvement was evident in tasks such as translation and coding.

Significant improvement in coding tasks

The hybrid model, an amalgamation of the PaLM2-S and a code-specific model, demonstrated significant improvements in performance. Specifically, in code generation tasks, the hybrid model outperformed the base model by an impressive 40%. These advancements could have considerable implications for the AI sector, potentially boosting efficiency in various tasks.

Improved language translation capability

A noteworthy benefit of augmenting the LLM with specialised models is the enhanced translation capability it offers. This was particularly evident when translating low-support languages into English - a persistent challenge in the realm of machine learning. Google's innovative method could be a game-changer in addressing this issue.

While the advancements in AI model training are promising, they also raise legal questions, particularly around the use of copyrighted data. Several AI companies are facing lawsuits alleging that their large language models are trained on copyrighted data, potentially leading to a shake-up in the AI landscape. However, Google's new LLM augmentation scheme could possibly mitigate some of these legal and financial challenges.

More articles

Also read

Here are some interesting articles on other sites from our network.