Tag: Language model

. ChatGPT, OpenAI's advanced language model, creates code that has been gaining a lot of attention. Researchers conducted a test to check the security risk. ChatGPT's code failed all minimal security standards. Hence it is not encouraged to be used without manual review by a coder. Awareness of ChatGPT's security flaws should be raised to prevent catastrophic consequences.

Stability AI recently released an open-source language-model tool, StableLM, to contribute to the fast-growing large language-model ecosystem. StableLM is an alternative to OpenAI's ChatGPT, which was causing some concerns due to Microsoft, OpenAI's primary investor, having too much power in the industry. With its alpha version, StableLM features 3 and 7 billion paramater models, with 15, 30 and 65 billion parameter models currently in development. Try the 7 billion parameter StableLM live model on HuggingFace.

ChatGPT is the ultimate go-to for finding answers quickly and easily. An advanced language model developed by OpenAI, ChatGPT uses vast amounts of training data and its supervised and reinforcement learning techniques to provide human-like responses to questions. Invested in by big tech companies such as Microsoft and Alphabet, ChatGPT is paving the way for a new generative AI. However, recent cheating scandals involving the technology have made proper implementation and regulation increasingly important.

StableLM, an open-source chatbot from Stability AI, is a new rival to ChatGPT. With only 3 to 7 billion parameters, it is 2 to 4% smaller than ChatGPT's 175-billion-parameter model. Access to StableLM is free, plus its use of EleutherAI’s 1.5 trillion token data set facilitates quality language processing for chatbot developers and hobbyists. StableLM Alpha accommodates lower-cost GPUs and can be tried at the Huggins Face demo page.

Explore the world of technology and business with O'Reilly's resources, content and courses on the new large language models (LLMs) and ChatGPT, the OpenAI-developed AI chatbot. Join the "ChatGPT: Possibilities and Pitfalls" online event, sponsored by deepset and featuring Microsoft's Jordi Ribas and GitHub's Thomas Dohmke to learn more about disruptive potential of these technologies. On-demand and live courses on the topic are also available to O'Reilly subscribers.