If you haven't checked out the "Surpise me" button on https://wiby.me I strongly suggest you do. These digital gardens are still out there even if sadly stuck in time.
How odd. My 3rd click, it brought me to news.ycombinator.com - exactly what brought me there in the first place and open in another tab. I was very confused as I assumed the tab had closed but no, it just appeared to be luck of the surprise.
I wonder how many of these things have data exports, would be great to get a list of all the URLs. I'd like to peruse them, and save them to archive.org.
I've used that button so many times to find things. I wish there was another directory that had the same unique feel but a bit more modernized. Wiby seems to be only one era or feel.
Nah, just been working on low level stuff and I haven't had the time to do a lot of the manual quality tuning that's always been necessary once in a blue moon.
Downside of being a one man show is I only have so much attention to spend on tasks, so sometimes quality tuning gets a bit neglected.
A really good topic that ties in with this is the need for deterministic sampling (I may have the terminology a bit incorrect) depending on what the model is indended for. The LLMWare team did a good 2 part video on this here as well (https://www.youtube.com/watch?v=7oMTGhSKuNY)
I think dedicated miniture LLMs are the way forward.
Disclaimer - Not affiliated with them in any way, just think it's a really cool project.
I have one personal niggle: I get annoyed when we end up lying to ourselves. Regarding the 101 section in video 1 - People forgot this the day LLMs came out. I felt this was too generous with the benefit of doubt.
This basic point was and remains constantly argued - with “Emergence” and anthropomorphization being the heart of the opposing argument.
I've been building upon the LLMWare project - https://github.com/llmware-ai/llmware - for the past 3 months. The ability to run these models locally on standard consumer CPUs, along with the abstraction provided to chop and change between models and different processes is really cool.
I think these SLIM models are the start of something powerful for automating internal business processes and enhancing the use case of LLMs. Still kinda blows my mind that this is all running on my 3900X and also runs on a bog standard Hetzner server with no GPU.
Don't forget to give some credit to llama.cpp which actually runs the models here and does all the things you're praising it for. This project is more about building a platform on top of it with RAG and function calling.
Oh yea 100%! llama.cpp and the opensource community in general is truely awesome in getting AI models into the hands of as many people as possible. I think these platforms are the key things to inspire people and get them to see the power of local LLMs in just a few minutes. Can't wait to see what other opensource platforms crop up in 2024 as well.
> The startups I worked with that experienced explosive growth were ones where their initial UX was mediocre or even hostile, but the demand for the thing was so high they had customers anyways. Improving the UX unlocked more growth.
Thank you for this comment. I've been stressing over UI elements a bit and this was the kick I needed to just go and get it out there.