The last few months have been very interesting from an AI perspective. I have written frequently about my exploits and explorations with ChatGPT 3, but I have also been using OpenAI’s Davinci and have tested several related systems built off it (such as Copy.ai, Simplified, Paragraph AI) as well as image generators (Stable Diffusion, DALL-E) and audio related systems like lalal.ai (which can be used to strip sounds from a recording to create an acapella, for example).
Google have now announced “Bard”, a new generative AI that is powered by LaMDA. LaMDA is a large language model that has been trained on a massive dataset of text and code. This allows Bard to generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
LaMDA (Language Model for Dialogue Applications) is a more recent language model than GPT, and it is designed to be more informative and comprehensive. It is also trained on a wider range of data, including dialogue data, which makes it better at understanding and responding to natural language.
Here is a table that summarizes the key differences between LaMDA and GPT (generated by Bard):
|Training data||Dialogue data, text data, code data||Text data, code data|
|Capabilities||Question answering, summarization, translation, dialogue||Text generation, translation, summarization|
|Strengths||More informative, comprehensive, better at understanding and responding to natural language||More creative, better at generating text|
|Weaknesses||Less creative, not as good at generating text||Less informative, comprehensive, not as good at understanding and responding to natural language|
Bard can understand and respond to natural language and this means that you can ask Bard questions in a conversational way, and it will be able to understand what you are asking and provide a helpful answer. For example, you could ask Bard “What is the capital of France?” or “What is the meaning of life?” and it would be able to provide you with a comprehensive and informative answer. The documentation around Bard seems to indicate they think this is an improvement over GPT based systems, although my experience is they both do a pretty good job of understanding your intent.
- Answer customer service questions
- Generate creative content, such as poems, scripts, and musical pieces
- Translate languages
- Write different kinds of creative content
Apparently you can use Bard directly from the Google Assistant (so Bard says), but I have not got this working yet.
Unlike ChatGPT, whose training data is not being actively updated (it cuts off in September 2021 currently), Bard’s dataset is being constantly updated from a variety of public and private sources, including Google Search, Wikipedia, and the collection of books, code and news articles that Google already indexes as part of those existing products. This means that Bard is currently much better equipped to answer questions relating to the last couple of years.
It’s not all plain sailing though, while Bard claims that LaMDA is “better” than GPT, my experience across a variety of tasks is that ChatGPT and Davinci currently both out perform Bard in several tasks – especially generating code. Asking ChatGPT to generate a PowerShell script to connect to a GSuite tenant will result in code that is nearly correct and can be fixed up easily. Asking Bard the same question generates code to connect you to Microsoft Exchange Online…