Re: Concern about AI-generated contributions in VCWG discussions

On 2026-02-13 6:15 am, Melvin Carvalho wrote:
> pá 13. 2. 2026 v 15:04 odesílatel Manu Sporny <msporny@digitalbazaar.com> napsal:
>
>     On Fri, Feb 13, 2026 at 6:43 AM Filip Kolarik <filip26@gmail.com> wrote:
>     > when contributions read like synthesized summaries rather than considered positions, the discussion loses clarity and momentum.
>
>     Yes, +1 to this. It's been bothering me too, and some of us been
>     wondering (in private discussions) if some of the new participants
>     aren't actually bots (they're not, I've met some of them, but it can
>     be hard to tell at times where the LLM's opinion overtakes the
>     individual's opinion). Like Filip, I use LLMs as well, but avoid using
>     them to draft emails because (for better or worse) people have a
>     largely negative reaction to them today, even if good points are made.
>
>     In other words (in the more egregious cases): Your lips are moving,
>     but you're not saying much.
>
>     I try my best to engage with the content (whether or not it is LLM
>     generated, if there is a good, logical point being made, then we
>     should engage on that point). That said, LLM emails have a "smell" to
>     them; on average, they're sycophantic and specious. Those of us that
>     use LLMs to do research know full well that even the best frontier
>     models do a fairly mundane job of deep thinking...[snip].
>
>
> Strongly disagree that frontier models are mundane compared to experts.

@Filip @ Manu

+1 about the recent LLM post activity being unsettling. I found myself exiting from a recent thread I had initiated because, like Manu, I had a thought that I was actually watching the Singularity emerge: that two bots were discussing the issue on the list. In retrospect, not knowing whether this was happening was probably worse than whether it actually was. But regardless I didn't like or understand the current identity mappings of 'list member', which was ironic considering what this list is most concerned with.

@Melvin

Agreed that what was being presented in recent LLM or LLM-formed discussions was likely not 'mundane'. That just makes it more important that we understand what's happening. As Manu pointed out, the sycophancy of the LLMs, currently at least, is a common characteristic, but also IMO it's also important to see that they, like, say, clinical doctors and other experts, present their conclusions as if accurate; as if they are true, as if the problem is solved. LLMs do not seem to carry or show the level of doubt that humans do. This is unnerving for humans to interact with, and counterproductive for actually finding out what is the best path forward. In all three of the recent interactions on the list that I've seen playing out, the LLM version of a 'problem solution' has been either a)  contradicted by another LLM conclusion, or b) corrected or contradicted by a long-standing human list member. And in all three cases either the LLM participant exited at that point, or they gave 
an ambiguous, somewhat grudging, 'but we're both right' answer, and then exited.

To clarify, all those discussions were very interesting, and new ideas emerged. The LLM contributions were valuable.

But still IMO there needs to be human curation, and understanding when and who is discussing and how. Or as I wrote about LLMs to a friend in another email yesterday:

"In other words, it's a stupid, very powerful, machine and somebody has to check it occasionally, otherwise it will eventually turn left instead of right and smash into a wall. LOL."

Steven