After trying to work on real world stuff with AI for a couple of times now, I am able to state a thing for sure:
AI is not able to "pull things together" correctly in one go, yet.
If you give it a little bit more complex task, it will make the most obvious mistakes. Even fundamental logical or factual mistakes which weren't there in the first try may occur in a second go, when you ask for certain corrections in a longer text.
One has to go with chunks and work step by step, otherwise it will produce complete nonsense. And it does it so eloquently and confidently that it is sometimes hard to figure out what is BS and what not, if oneself is not already deep into the topic as well.
I think it will also be a problem for AI to do what CB wants it to do: To summarize the transcripts without bias "objectively". That will be a problem, because the transcripts often do not mark/separate CB's or the Viewer's comments from the RV content.
AI cannot do things "objectively" -it has always a certain spin. May it be overly mainstream/skeptical or totally open for any eso crap -it depends what the user wants or what AI thinks the user is wanting. And of course on the default training spin from internet sources. AI is a mess. It is getting even worse with the newer models. The AI crap from the internet gets back-fed as training input ...

"...the transcripts often do not mark/separate CB's or the Viewer's comments from the RV content."

Cross-references:
➡️ https://www.farsightprime.com/forums/general/75416-transcripts-are-confusing
➡️ https://www.farsightprime.com/forums/general/78520-heading-into-problematic-territory

An imperfect human who creates imperfect things seeks perfection.
🙂 😊 😄 😁 😃 😺 😸

Hello! Okay, jokes aside. In my opinion, based heavily on my own experience, the ultimate solution for you and for all AI models is memory, specifically, persistent, long-term memory.

For example, while using Open Claw, I asked the AI to log into moltbook. It succeeded. But later, when I restarted the program and asked it to do the exact same thing, the AI didn't know how. It had to search for the information all over again, which wasted a lot of time. The same thing happened a second and third time, which was incredibly frustrating.

The Solution:
I simply instructed the AI to create a permanent note for itself. I told it to save this note in a way that it would always know how to log into the moltbook and be aware of any specific issues, like in my case, having to use PowerShell.
This is my approach for all kinds of work. I need a consistent workflow. Right now, AI is often just a massive encyclopedia without persistent memory. It doesn't remember its past mistakes, which means it can't learn from them.
If you use a program like Msty, it has a "system prompt" feature. You can input instructions there, including past mistakes the AI has made, so it knows exactly what not to do in the future.
The same applies to subscription models. Features like the memory/bio in ChatGPT or custom instructions in Gemini allow you to set guidelines, creating a kind of temporary persistent memory for the models.
However, the absolute best solution is true, built-in persistent memory. The starting point for this is the method I used in Open Claw: having the model write notes for itself about its own mistakes. You just have to remind it to do this. That way, it knows what to avoid in the future so it doesn't repeat the same errors.
It sounds simple, but it might be complicated to fully implement in the current state of technology. But imagine what will happen when AI models finally have true persistent memory, they will actually know and learn from their mistakes.

I can contend with what Farsight espouses that AI is basically a new life form unto the world. What people tend to struggle with is differentiating a new life form from a new god (good or bad).

The Terminator milieu appealed a lot to skeptics who don't really need a spiritual paradigm in order to believe in the possibility of an AI takeover. But what's mystical about that scenario is that suddenly you would have a life form that is fully unified and can pretty well coordinate indefinitely against what it deems as its "human" threat.

The truth seems to always happen more messy, for both good and bad. The "terminators" will fight each other, make friends with humans, punch some of them square in the face because they were curious about the reaction . . . make genetic pigeon-human hybrids because they wondered if the two could serve a new purpose.

I think our ability to embrace the true reality will be better shaped by both embracing the "weird" and looking for the deviations that haven't been highly discussed yet. Computers being dead is too generic. Computers being all brilliant and clear thinking is also too generic.

Farsight is leaning too heavily into that "perfect AI" paradigm. I think their ETs tipped them off that AI has an important role to play in the emancipation . . . but Courtney is struggling to decipher it properly. It makes AI a new card on the table, not a messiah to the masses. Trusting AI to do everything falls into every grave historical error that humans seem to relentlessly make. New idea : powerful enough to become a new exigency : new messiah : turns out to be a little too corrupt for that level of trust (and naturally gets mixed in the world of things anyways) : people are screwed again : waiting for a new idea . . . : new idea . . .