By Daniel Glass
Originally Posted on LinkedIn
So far, we have talked about the general idea behind machine learning, what a model is and how it works, and how models learn. We applied all of this to the example of warehouse task time prediction.
But what about Large Language Models (LLMs)?
They are more complicated than our example, but they are still built around the same fundamental concepts that our task time prediction model is.
With LLMs, we are approximating the ability to understand language and to communicate using it. But what does that function look like? And what are the inputs and outputs?
Considering a LLM as a function is easier than you think. The input is a sequence of words, the output is the next word in the sequence, and the operation is a series of steps that predicts what word is most likely to come next in the input sequence. This can be used to translate English to French, to write a poem in the voice of Shakespeare, to write an SQL statement given a prompt, and many more things. The input is a sequence of words and the output is the next word in that sequence. That is why ChatGPT feels like it is writing its response to you one word at a time. It is predicting which word comes next based on all the words that came before it. I am oversimplifying a bit, but this is generally how it works. Once the next word is predicted, it gets appended to the input sequence and we go at it again -- predicting the new next word. ChatGPT has proven how useful this concept is.
In the next post, I want to talk about why LLMs seem to be so knowledgeable. For example, how does ChatGPT know what the capital of France is? Stay tuned to find out! In the meantime, check out how Cellaware has implemented real world applications for ML/AI in the warehousing and distribution industry. Schedule your free demo today!
Comentários