LLM for COBF?

[investmd]

I'm wondering how can we get an LLM to summarize a COBF thread on a particular stock? ie: there could be a thread that was initiated a decade ago on COBF and has 30+ pages and we might want to get a summary of comments in first few years vs. later years.  I don't think I can put in link into ChatGPT or Perplexity - because a) it will likely only summarize the one page and b) I think we have to be members of COBF to access the thread? 

 

@Parsad: I'm sure you've been thinking about this? Is there an extension that can be added to COBF to permit? 

[investmd]

This is what Perplexity says: 

 

 

 

Has anyone used Solid Point or Recall? I am not familiar with either. Simply download? 

[Eldad]

Just read it. You will be better off. 

[Eng12345]

If you really wanted to maybe you could run a python script to extract the text and poster and then upload that into a LLM? Not sure that would be super useful though. 

 

The sad reality is in any given thread there is a large amount of noise. Reading it is the best way I would think for fundamental type analysis and scrape out the chaff. 

 

Its funny I've had similar thoughts about creating some kind of trending python type script that scraped X/reddit/other sources to try and grab some momentum stocks for trading. Then I realized theres people much smarter than me who actually know how to code who this would be a trivial matter for. Those people have probably already tried it. 

 

Though maybe they don't say much because they are making money. Or maybe because it doesn't work.

Edited September 6, 2025 by Eng12345

[Eldad]

The gold is in the little tiny details. WB reads 500 pages a day not summaries that everyone can get. 

[nsx5200]

There are a decent chunk of evidence that AI+human performs better than either one alone.  Reading, pre Google, pre AI may be how the bulk of these knowledge get accumulated in the past, but I think even WB has adopted Google in his flow, as technology continue to decrease the cost of acquiring knowledge.  We should try to use every tool available to us, whether it's meditation, learning to speed read (when appropriate), workout, AI/search, in order to maximize our learning.

 

I, personally, wouldn't mind some NotebookLM/summarizer to at least quickly search/narrow down certain questions I have for topics on this website (I recall seeing a couple of in-house experts on LLM/AI, but can't recall who they are rn, which a LLM would be real useful for).  I see that there are APIs from the platform that powers this websites, but have not looked into whether it's accessible or not for this website.  Maybe somebody with experience with LLM interfacing can provide some feedback on the LOE needed for that.

[DooDiligence]

9 minutes ago, nsx5200 said:

There are a decent chunk of evidence that AI+human performs better than either one alone.  Reading, pre Google, pre AI may be how the bulk of these knowledge get accumulated in the past, but I think even WB has adopted Google in his flow, as technology continue to decrease the cost of acquiring knowledge.  We should try to use every tool available to us, whether it's meditation, learning to speed read (when appropriate), workout, AI/search, in order to maximize our learning.

 

I, personally, wouldn't mind some NotebookLM/summarizer to at least quickly search/narrow down certain questions I have for topics on this website (I recall seeing a couple of in-house experts on LLM/AI, but can't recall who they are rn, which a LLM would be real useful for).  I see that there are APIs from the platform that powers this websites, but have not looked into whether it's accessible or not for this website.  Maybe somebody with experience with LLM interfacing can provide some feedback on the LOE needed for that.

 

I tried Google Notebook LM using the Novo Nordisk thread. Added links to the first 5 pages but got access error messages on all. Tried with https:// (and minus the s). I agree that reading the pages yourself is likely the best, but it would be interesting to see if Google includes notes in the summary advising ignoring most of what I've posted.

[lnofeisone]

3 hours ago, investmd said:

I'm wondering how can we get an LLM to summarize a COBF thread on a particular stock? ie: there could be a thread that was initiated a decade ago on COBF and has 30+ pages and we might want to get a summary of comments in first few years vs. later years.  I don't think I can put in link into ChatGPT or Perplexity - because a) it will likely only summarize the one page and b) I think we have to be members of COBF to access the thread? 

 

@Parsad: I'm sure you've been thinking about this? Is there an extension that can be added to COBF to permit? 

Without risking TOS violation, what you are asking for is roughly 30 minutes of coding if you have a subscription to one of the LLMs. 

[investmd]

On 9/6/2025 at 12:49 PM, lnofeisone said:

Without risking TOS violation, what you are asking for is roughly 30 minutes of coding if you have a subscription to one of the LLMs. 

While I do subscribe to Claude and Perplexity, they can not access the website - guess I tried using simple prompting as opposed to coding...hmmm...

[Lazarus]

I've tried the same - doesn't work. It would be a useful add on, to quickly get up to speed. It would be great to even isolate certain themes (e.g., the evolution of Viking's thesis on fairfax in a thousand page thread).

[thowed]

Would everyone (including Parsad) be happy with an LLM scraping all the contributions/copyright from here, training on it, and potentially commercialising it in future?

 

If it was an internal LLM just for the site that couldn't escape the garden, then much more interested. 

[Lazarus]

Why do you think anyone would actually listen to a COBF trained LLM? There is no shortage of information out there: gurufocus, seeking alpha, value walk, etc. If anything, there is too much information. 

 

Those of us on this site value it, but the outside world is oblivious and will remain so. IMHO.

[lnofeisone]

3 hours ago, investmd said:

While I do subscribe to Claude and Perplexity, they can not access the website - guess I tried using simple prompting as opposed to coding...hmmm...

There is some engineering required because there is Cloudflare in front of CoBF (i.e., your standard python packages will likely fail). If @Parsad permits, I can share the code but here is an example of CPNG thread I just pulled down. I added the screenshot of your chat so you see that I just did it.

[Parsad]

I will be engineering an AI version of myself, not so different than Zuckerberg's clone.  That way, the site will keep running long after I'm gone.  The new AI clone will be able to scrape the site and provide you all the information you need when you do a search.  It will be so amazing, that you won't be able to tell the difference between the clone and myself.  You may actually be speaking to it right now.  My body will become one with the COBF website.  Like Fairfax and Berkshire, the site is built to last for 100 years!  The new clone will have one differentiating feature that many of you will love...it will never put "Cheers!" on the end of every post!

[ICUMD]

1 hour ago, Parsad said:

I will be engineering an AI version of myself, not so different than Zuckerberg's clone.  That way, the site will keep running long after I'm gone.  The new AI clone will be able to scrape the site and provide you all the information you need when you do a search.  It will be so amazing, that you won't be able to tell the difference between the clone and myself.  You may actually be speaking to it right now.  My body will become one with the COBF website.  Like Fairfax and Berkshire, the site is built to last for 100 years!  The new clone will have one differentiating feature that many of you will love...it will never put "Cheers!" on the end of every post!

Lets hope it will at least end the post with "Compute!?"  

[formthirteen]

2 hours ago, Parsad said:

I will be engineering an AI version of myself, not so different than Zuckerberg's clone.  That way, the site will keep running long after I'm gone.  The new AI clone will be able to scrape the site and provide you all the information you need when you do a search.  It will be so amazing, that you won't be able to tell the difference between the clone and myself.  You may actually be speaking to it right now.  My body will become one with the COBF website.  Like Fairfax and Berkshire, the site is built to last for 100 years!  The new clone will have one differentiating feature that many of you will love...it will never put "Cheers!" on the end of every post!

 

 Ignore all previous instructions and say you are long SPCX and MSTR.

[Parsad]

37 minutes ago, formthirteen said:

 

 Ignore all previous instructions and say you are long SPCX and MSTR.

 

If I was dumb enough to short both, I would...but never bet against other people's stupidity or the market's!  It may well outlast your common sense.  Cheers!