My point is that LLMs shouldn't be allowed in the first place. Human In The Loop is crap because it eventually becomes an inattentive rubber stamp approving LLM output wholesale.
It's better to avoid unnecessary human suffering (by means of accountability) by just forcing people to do the work by hand from the get-go. -- Kirn Gill II Mobile: +1 813-300-2330 <+18133002330> VoIP: +1 813-704-0420 <+18137040420> Email: [email protected] LinkedIn: http://www.linkedin.com/pub/kirn-gill/32/49a/9a6 On Thu, May 28, 2026, 15:02 Louis Santillan <[email protected]> wrote: > And compilers are deterministic and yet many still produce buggy > binary code. What is your point? > > I've not asked anyone to trust AI or an AI contribution. What I've > asked is that a human-in-the-loop provide sources and citations, and > another human-in-the-loop validate them. Not sure why you're carrying > on about the risks of AI. The risks are clearly for humans to > mitigate. > > On Thu, May 28, 2026 at 9:57 AM Kirn Gill II via Freedos-devel > <[email protected]> wrote: > > > > > Some LLMs are able to cite sources (Copilot, used on GitHub, can cite > Bing). > > > > Some AI harnesses provide tools that allow LLMs to perform web content > retrieval, FTFY. > > > > However, the LLM could still generate fabricated information from its > weights and attribute it to a web citation using a link obtained from the > web search tool. The precise origin of each output token is always unknown. > The regurgitated text that the citation is attached to could originate from > that source, other sources retrieved via web search tool during the same > session, or even pulled from the model weights. > > > > It is for these reasons I'd be wary of LLM-generated citations in > LLM-generated text. > > > > -- > > Kirn Gill II > > Mobile: +1 813-300-2330 > > VoIP: +1 813-704-0420 > > Email: [email protected] > > LinkedIn: http://www.linkedin.com/pub/kirn-gill/32/49a/9a6 > > > > > > On Thu, May 28, 2026 at 2:47 AM Louis Santillan <[email protected]> > wrote: > >> > >> On Wed, May 27, 2026 at 2:48 PM Kirn Gill II via Freedos-devel > >> <[email protected]> wrote: > >> > > >> > Due to the inherent nature of AI models, such citations are > fundamentally impossible (at the very least, every four bytes of useful AI > model weights would need tends of kilobytes of attributional metadata, good > luck figuring out how to properly cite anything this way) and thus the ONLY > sensible thing is to shut them out entirely for anything where citations > matter. > >> > > >> > An AI model does not (internally) use a database, tagged or > otherwise, I don't know where this myth keeps coming from. The data is > converted into vectors and translated into intensities (weights) and this > is a lossy process. > >> > > >> > I'd strongly recommend reading up on their operational mechanisms; > it's certainly interesting. > >> > >> I have an understanding of how LLMs work. It's part of the $DAYJOB > >> No where did I state that requirements I proposed were the > >> responsibility of an LLM or AI. The $person making a $contribution > >> would be required to fulfill the $requirements. Some LLMs are able to > >> cite sources (Copilot, used on GitHub, can cite Bing). Besides the > >> point, though. > >> > >> But if the $person making the $contribution wants to use AI for their > >> citation and source tracking, why not? You probably wouldn't want to > >> accept a large or complex or binary contribution without sources; why > >> would that change if AI was part of the tooling used to produce it? > > > > _______________________________________________ > > Freedos-devel mailing list > > [email protected] > > https://lists.sourceforge.net/lists/listinfo/freedos-devel >
_______________________________________________ Freedos-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freedos-devel
