LaMDA is a Transformer-based neural language model that has been pre-trained using 1.56T words of freely available conversation data and online pages. It has up to 137B parameters and is evaluated based on the criteria of Quality, Safety, and Groundedness. It is currently only accessible to a select group of AI developers through the AI Test Kitchen.
On the other hand, ChatGPT is based on the GPT-3.5 architecture and has 175B parameters. It includes three models: text-davinci-002, which is trained by supervised fine-tuning on human-written demonstrations and rated 7/7 by human labelers on overall quality scores; text-davinci-003, which includes reinforcement learning with human feedback (RLHF) and is trained on comparisons by humans; and code-davinci-002, which is the base model for code completion tasks. It uses LaMDA-like supervised knowledge, allowing human AI trainers to use model recommendations to create answers and train the model while simulating both the user and the AI assistant.
Some have speculated that LaMDA was never released because its conversational platform did not align with Google's current advertising strategy, which generates the majority of the company's income. There may also be reputational risks associated with the AI chatbot, as it has been known to mimic human speech from the internet and spread hate speech, fake news, and inaccurate information.
It is possible that Google may incorporate ChatGPT into its search engine in the future, as the company seeks to stay competitive and meet changing customer needs. However, there are also concerns about the need for human oversight to ensure that the data provided on the platform is accurate, trustworthy, and secure for the general public. It is also worth noting that LaMDA has certain advantages, including its metrics for evaluating answers based on groundedness and Quality, Safety, and Interest (SSI).
Overall, it is clear that both LaMDA and ChatGPT are highly advanced AI chatbots that have the potential to revolutionize the field of conversational technology. It will be interesting to see how they continue to evolve and how they may be used in the future.
Top comments (0)