It's honestly saddening to see companies calling their projects "fully free and open source" when in reality, they aren't fully open source. Of course, not all companies are contributors to this, as Microsoft Phi (One of my favorites!) is open source. But hey, at least we have good ol actually open source Llama 2!
Yes. Put it under a license that requires attribution (like MIT) and open source it. These kinds of software are better if open so that more people can analyze and implement similar technology to make software more accessible.
That’s exactly what I have in mind too. The only issue is avoiding the “freeloader problem” that most big corporations have embraced lately. For example, with my local navigation system, I don’t want Google scraping all my valuable data to feed into Google Maps—only for them to charge me $10,000 per API call when I try to query the same data they took.
If you’re a nonprofit or a small, local business, then by all means, use my maps—that’s who I built them for. But big corporations? No.
Open Maps is facing this exact problem right now. Every shady AI and LLM is scraping their data and making millions, yet contributing nothing back to the source.
I want to avoid all that by making it very clear: if you're a big corp and you want access to our data, you need to pay up. Otherwise, take a hike.
They’ll probably take the data anyway without permission, just because they can afford $500/hour lawyers—but at least I’ll have made the effort to stand my ground.
I've seen AI do this before, it's pretty funny lol.
When I was coding my Phi3 wrapper, I used some funky generation parameters and it would generate random sentences until it reached the max token amount.
Cyckle is an AI wrapper built in Cython, it runs local LLMs like Phi-3.5, and can analyze CSVs without sending anything to the cloud. All open-source.
Currently works on Debian (I recommend 13 as that was what I am running right now, but 12 should work fine. If it doesn't, let me know in the repo!)