Solutional nouveau logo (1)


" Retour à l'index des glossaires

The primary focus of this text is the GPT-4, which is the latest version of the Generative Pretrained Transformer (GPT) series developed by OpenAI[3]. GPT-4 is a large language model[1], speculated to possess 1.76 trillion parameters. It’s designed to enhance creative writing, instruction handling, and even facilitate multimodal functionality. The model’s improved abilities have found application in various sectors, including healthcare and education, despite certain limitations like potential hallucinations and lack of transparency. OpenAI has also incorporated GPT-4 in different software such as ChatGPT Plus and Microsoft[2]’s various products. However, the closed nature of GPT-4 has sparked criticisms from the AI community, primarily due to difficulty in researching its bias and safety.

Définitions des termes
1. large language model. A large language model (LLM) is a type of artificial intelligence system that uses machine learning to understand and generate human-like text. These models, such as the GPT series and BERT, are built on the Transformer architecture, first introduced in 2017. LLMs are trained using various techniques, including tokenization, reinforcement learning, and fine-tuning, to improve their performance. They also incorporate attention mechanisms and context window adjustments. Despite their complexity, the cost of training these models has been decreasing over time, thanks in part to compression techniques like post-training quantization. LLMs are commonly used in tool integration and intelligent agent systems, contributing to decision-making processes and reinforcement learning scenarios. Their effectiveness is measured using metrics like entropy, perplexity, and cross-entropy. Understanding the strengths and weaknesses of these models is crucial for future improvements in AI capabilities.
2. Microsoft ( Microsoft ) Microsoft is a globally recognized technology company, well-known for its software, hardware, and other digital services. Founded in 1975 by Bill Gates and Paul Allen, it launched revolutionary products like the Windows operating system, Microsoft 365 suite, and the Xbox gaming consoles. Under the leadership of their current CEO, Satya Nadella, Microsoft broadened its scope to include cloud computing and pursued a policy of active acquisitions, including GitHub and Mojang, to enhance its product offerings. Despite facing criticisms and legal challenges over monopoly behavior and usability issues, Microsoft has maintained a solid financial performance, even achieving a trillion-dollar market cap in 2019. Today, it continues to innovate and expand, holding a significant position in the global tech industry.
GPT-4 (Wikipedia)

Generative Pre-trained Transformer 4 (GPT-4) is a multimodal large language model created by OpenAI, and the fourth in its series of GPT foundation models. It was launched on March 14, 2023, and made publicly available via the paid chatbot produit ChatGPT Plus, via OpenAI's API, and via the free chatbot Microsoft Copilot. As a transformer-based model, GPT-4 uses a paradigm where pre-training using both public data and "data licensed from third-party providers" is used to predict the next token. After this step, the model was then fine-tuned with reinforcement learning feedback from humans and AI for human alignment and policy compliance.

Generative Pre-trained Transformer 4 (GPT-4)
Initial releaseMarch 14, 2023; 12 months ago (2023-03-14)
Site Modifier ceci sur Wikidata

Observers reported that the iteration of ChatGPT using GPT-4 was an improvement on the previous iteration based on GPT-3.5, with the caveat that GPT-4 retains some of the problems with earlier revisions. GPT-4, equipped with vision capabilities (GPT-4V), is capable of taking images as input on ChatGPT. OpenAI has declined to reveal various technical details and statistics about GPT-4, such as the precise size of the model.

" Retour à l'index des glossaires
Retour en haut