51–58 of 58 posts
Earlier quoted context omitted.
No but each version of L2E OS has a name. v0.1 is the first version and we wanted to pay our tribute to Terry A Davis, so called it TempleDOS and added a bit of TempleOS references.
man those kids are gonna look that up and go down a rabbit hole aren't they
> How do we make sure that the output is factual and not hallucinated? One method the readme doesn't mention: ask the same question multiple times. Apparently research suggests that when LLMs hallucinate answers, their hallucination is likely to be a different one every time, where as factual answers will tend to be consistent. https://arxiv.org/abs/2305.18248
Asking repeatedly would be impractical. I have some vague idea in my mind called the fact engine. It's like a wiki where people could key in facts. Also facts based on "timeframes", ie 1800's 1900's etc (eg Moon Landing). The fact engine feeds a model that checks for hallucinations, and also rule based engines. Then a final score would be assigned based on consensus. Something like that.
Earlier quoted context omitted.
Asking repeatedly would be impractical. I have some vague idea in my mind called the fact engine. It's like a wiki where people could key in facts. Also facts based on "timeframes", ie 1800's 1900's etc (eg Moon Landing). The fact engine feeds a model that checks for hallucinations, and also rule based engines. Then a final score would be assigned based on consensus. Something like that.
An encyclopedia with an api?
Creator here. Thank you for sharing!
Creator here. Thank you for sharing!
the iso is 68mb or so, so the model is not there, how does it work? (tried on Oracle VM and didn't work but seems it is not supported)
Qemu:
qemu-system-x86_64 -display gtk,zoom-to-fit=off -m 512 -accel kvm -vga virtio -cdrom l2eos.iso
> How do we make sure that the output is factual and not hallucinated? One method the readme doesn't mention: ask the same question multiple times. Apparently research suggests that when LLMs hallucinate answers, their hallucination is likely to be a different one every time, where as factual answers will tend to be consistent. https://arxiv.org/abs/2305.18248
Asking repeatedly would be impractical. I have some vague idea in my mind called the fact engine. It's like a wiki where people could key in facts. Also facts based on "timeframes", ie 1800's 1900's etc (eg Moon Landing). The fact engine feeds a model that checks for hallucinations, and also rule based engines. Then a final score would be assigned based on consensus. Something like that.
Couple related things in this direction that may be leveragable:
- WikiData: https://m.wikidata.org/wiki/
- Knowledge Futures' Underlay: https://docs.underlay.org/history