There was a time when Microsoft had the only internet browser. Then came Firefox and Google Chrome, among others. I see the AI boom as the next frontier of the internet revolution and one that Google is poised to fill a need for.
Open Source AI has mostly been a realm of possibilities for people who know how to navigate with a terminal interface, or are programmers familiar with the full suite of tools like Visual Studio Code insiders with Codium. Microsoft seems to get that 95% of the population detests terminal interfaces and therefore limits access to these amazing programs to a select few people in the know. The learning curve if you don't know how to use terminal interfaces, or have to refresh your understanding from 30 years ago (like I did) to even operate some of these programs is steeper than most programmers are willing to admit. After all, they spend hours per day navigating these interfaces, so it can be easy to forget how cumbersome and awkward they are for your average internet consumer.
A few people have realized this great frustration from the masses and created GUI interfaces that "dumb down" the installation process for models on Github and Huggingface. A few that come to mind are GPT4ALL, H2O, and Pinokio... even H2O will direct the program to interface with the user through a browser, however. While these programs are OK, they can also be complicated without a lot of explanation and still require technical knowledge to function properly for certain language models to operate properly with the interface. Only GPT4ALL has a standalone GUI that I know of. Perhaps the most annoying thing about these universal GUI is that if you don't know anything about open-source AI, it can take weeks before you discover these interface programs. To my knowledge, only GPT4ALL is reasonably well-known among the beginner population. There really isn't a primer for enthusiast end-users to get in on the craze and this is where I think Chrome can fill a need.
Since these universal GUI programs tend to use Chrome anyway, why not make it official and incorporate the capacity for Chrome to download, install, and run LLM models direct from Github with a standardized interface that developers can begin to work their programs around that end users can become familiar with? Of course, there will always be some outlier and upstart models that may require a reinventing of the interface from time to time, but that's nothing new to Google. Making Chrome the easy first choice for GUI of brand-new Huggingface models is an obvious win for Google, especially if they work directly with models hosted on Colab. Then you have the best of all worlds, no downloads required, and an interface you're familiar with! It's a perfect entry solution for the general public.
My only disclaimer is that I'm not a program or developer so forgive me if I'm missing some obvious technical issue here. I'm speaking as someone who wanted to get excited about AI but after months of trying to make one do something productive for me, I am feeling a bit defeated. I can't be the only one, but I suspect this might be a solution for others like me.
Again, I'm not a programmer, so if y'all want to run with the idea then you are free to do so. Just give me first dibs! LOL
LinkedIn
Twitter
