I was skeptical that I would like it as well, but it turns out I use it quite a bit.

If I could summarize AI its sort of like this....imagine if you had a super helpful intelligent 15 year-old to help you out with any/all task. He is often wrong about things, sometimes spectacularly, but more often he is either right or 80% right...but he has a special power: He can read and digest long complicated manuals you have no interest in reading (say a Service and Repair Manual for your stove) and provide you meaningful data about it.

So you can upload a PDF of the service manual and ask: Please read this and tell me how to change the front LCD panel....and he will summarize the procedure in an easy to digest form.

The space is rife with gimmicks, the signal will become clearer over time.

Made my nostr profile pic Made cover art Was thinking about making a comic book The beefy start9 has dalle that you can run locally

I find Claude better for straight technical questions. However Chat-GPT4 seems better at finding info on social sites.

One thing I forgot to mention, is it has a "Focus" mode that basically skews search results towards discussion forums, reddit, etc. So if you ask a question is "What is best framework for blah...blah...blah" it avoids all the vendor specific websites and focuses on what others are saying about the framework. ChatGPT seems better for that.

Super helpful. Thanks so much.

Take it as is: a model (like Llama) is a Large Language Model (LLMs). LLMs can take a various sort of inputs (text, images, audio waves, whatever) and return some response. The response can be new text, an image, a suggestion of text based on what the input text was, etc. At their core, they simply take the input data, transform it into a format that they consider readable (encoding) and throw this data into an enormous washing machine that is the model itself. The model shuffles data, tries to understand relationship between data, tries to find order and patterns into what it was given as input.

Once the model is done, it presumably has found some sort of order, pattern in the data...it developed an understanding of what the provided data are and what to return as output.

This process can be helpful for a ton of tasks, particularly text processing (creation, summarisation, rephrasing). Also other form of input can be handled, but they usually require bigger washing machines (aka models) because there's more stuff to elaborate. For example, a picture is way more heavy than a text sentence, thus models that handle pictures are generally more computationally demanding (they are usually Convolutional Neural Networks, if you're interesting it that).

Models running locally are usually text processing models, because it requires relatively less power to run them.

This is what it is, at its core.