“What AI Doesn’t – AU Does!”
Artificial Intelligence (AI) models are constantly evolving. I dedicate this page to one of the most exciting and controversial trends in the world of computer science. Since the beginning of computing, man has been attempting to build “intelligent” digital systems. Systems that attempt to replicate human thought and offer the potential to enhance our abilities in areas that only science fiction could imagine. The true possibilities are mind boggling. I will be constantly adding links and descriptions to many of the most popular models that people and companies are using and exploring daily in their quest to improve how they learn, work, and communicate.
ANTHROP\C
Anthropic’s Claude models—including Claude 3 and its successors—are based on proprietary large language model (LLM) architectures developed in-house by Anthropic. These models are not built on top of existing engines like OpenAI’s GPT or Google’s Gemini; instead, they are unique to Anthropic and incorporate specialized training methods such as “Constitutional AI” to guide safe and reliable outputs.
The Claude engine is continually updated, with recent versions like Claude 3.5 Sonnet and Haiku introducing advanced capabilities and improved performance benchmarks compared to other leading models.
ChatGPT
ChatGPT is based on OpenAI’s proprietary Generative Pre-trained Transformer (GPT) models, specifically large language models like GPT-4o and its predecessors. The core engine is the transformer neural network architecture, which enables ChatGPT to generate human-like responses by analyzing context and predicting the next word in a sequence. The model is further fine-tuned for conversation using supervised learning and reinforcement learning from human feedback (RLHF).
Google Gemini
Google Gemini is based on a family of multimodal, transformer-based large language models (LLMs) developed by Google DeepMind. It is not based on a single “engine” in the traditional sense, but rather on a series of neural network architectures optimized for processing and generating text, images, audio, and video. The models are successors to LaMDA and PaLM 2, and are designed to be natively multimodal—meaning they can understand and generate content across multiple data types.
Gemini runs on Google’s custom AI hardware, including Tensor Processing Units (TPUs) and the latest Trillium TPUs, which power its training and inference. The architecture is enhanced with efficient attention mechanisms to process long sequences of mixed data types.
Microsoft CoPilot
Microsoft Copilot is primarily based on OpenAI’s GPT-4 large language model, with additional integration of Microsoft’s own Prometheus AI model. The Prometheus model combines GPT-4 (and sometimes GPT-4o and DALL-E 3 for image generation) with proprietary Microsoft technologies, including Bing search data, Microsoft Natural Language Processing, and Retrieval Augmented Generation (RAG) for enhanced context and grounding. For more complex reasoning tasks, Copilot can also use OpenAI’s o1 reasoning model.
https://copilot.microsoft.com/
Perplexity.ai
Perplexity.ai is based on large language models (LLMs) and advanced AI technologies. The free version primarily uses OpenAI’s GPT-3.5 and integrates with web search engines like Bing to provide up-to-date answers. For Pro subscribers, Perplexity offers access to more advanced models, including GPT-4 Omni, Claude 3.5 Sonnet, Mistral Large, and Llama 3, allowing users to choose the model best suited for their needs. The platform combines these LLMs with real-time web indexing and natural language processing to deliver concise, sourced answers in a conversational format