If it includes my work, how bad can it be???😆

Ja, interessant, maar zie nog niet waarom text mining en statistische methodes daarvoor niet meer geschikt zijn. Wellicht kan een LLM langere trends beter doorzien dan text mining. Er zijn inmiddels wel steeds meer omvangrijke sets data die langere periodes beslaan. Dus dat geeft vast mogelijkheden.

If it is fed enough material, I think it can. An advantage of using a locally hosted model is that it does not censor the material and the output in the ways that the public models do.

At least the Open WebUI makes it feel very easy to create "knowledge" for the model.

Should be really easy. We have the txt files made with HTR of a large number of colonial archives. Must load even easier than the pile of pdf's that I've fed it now.

Nice, I'm using Ollama 3.2 now. And to be honest don't really notice a difference between models (yet).

Haha, ok, so I've fed it my publications from the past 15 years. I guess I've created a machine version of my academic brain.

...and I can now train my local LLM with my own relevant data and literature. This might actually become fun/useful

Installed the web user interface (Open WebUI) during morning coffee. Really don't see why people who like to use AI for daily tasks don't use it locally. ollama.com

Strange how wel it can reason. I've asked it to reflect on certain solutions to complex issues. It managed to do so quite well. And when asked to give counter arguments it did so successfully, in my view.

And it works surprisingly well. I just asked it to summarise several full text articles and I don't think it did better or worse that chatGPT. I don't personally see a need for it atm, but I don't think that most people who do want to use some AI models need to rely on the popular US-based services.

Or just download the freely available pdf! library.oapen.org/bitstream/ha...