Liquid Neural Networks: The End of the Era of Giant Models
Liquid AI
Liquid AI has introduced Liquid Foundation Models (LFMs), a new type of neural network that is more efficient, explainable, and adaptable than traditional large language models. These models, inspired by the brain of the C. elegans worm, use far fewer parameters and can be run on edge devices, potentially making AI more accessible and sustainable.
Researchers are increasingly developing liquid neural networks (LNNs), which were originally created at MIT, as a more efficient alternative to traditional LLMs. In late 2024, the company Liquid AI released its Liquid Foundation Models (LFMs), which claim to surpass ChatGPT, Claude, and Gemini in some respects while using far fewer resources. LNNs are based on the principles of the nervous system of the C. elegans worm, which has only 302 neurons, and they are adaptive and explainable, unlike traditional AI models that act as black boxes. Liquid AI's LFM, with 1.3 billion parameters, performs comparably to models like Llama or Mistral, but can run on devices with limited computational power, making it suitable for edge AI. This could significantly reduce the cost and energy consumption of AI, potentially leading to more sustainable and accessible AI systems. The company is actively working on deploying these models in various applications.
- Abbreviations
- LFM = Liquid Foundation Model — жидкая фундаментальная модель
- LNN = Liquid Neural Network — жидкая нейронная сеть
- LLM = Large Language Model — большая языковая модель
- MIT = Massachusetts Institute of Technology — Массачусетский технологический институт
Source: AI-News.ru —
original