ChatGPT knows everything

For simple topics or "teach me about X" for sure... but AI has lost hype in the workplace because it makes SO MANY errors. Most people in my firm hardly use AI because you have to fact check it meticulously and that takes more time than finding it yourself.

Can it automate simple tasks - sure. I think at the current level, it is not there yet to replace anyone except maybe call center employees.

If you think ChatGPT knows everything you haven’t been fact checking it closely enough.

It’s also a matter of macro vs micro.

I find it very useful to structure thoughts, blueprint phase, high level industry validation. But the details really matter in complex asset environments. More importantly judgement calls. 

Maybe it’s crazy fast desktop research is sufficient to give you a glipse of what a broad industry may be growing at or being disrupted by. 

But in more opaque markets, there’s elements at a firm level that really make the difference: who’s talking to who at trade shows for B2B deals, hirings/firings that industry people know about way before journalists or even regulators, you know your product is a killer app because you’ve seen it in product demos, your middle management sucks and can’t manage a cash flow forecast. Business has an inherent bias towards secrecy where possible because that information is in itself an edge. That will create a limit on what the models can be trained on without deeper access into firm level information and therefore a limit to its threat to real world business.

Then of course there’s the issue of hallucination, although in my experience this is more of a nuisance than a fatal flaw. I recently wrote up a teaser with mostly internal input leaning on GPT for restructuring and copy editing. Only found one straight up hallucination (which was a very understandable one and in fact I feel like 1-2yr analysts make similar mistakes all the time.)

Tldr: won’t know everything. What that means for alpha in my opinion is a further deterioration in edge at public markets level (w/o questionable practices) but in PE I think it favors those closest to the metal (their operating assets) where desktop research ceases to be a meaningful driver of value creation.

I've found the best way to avoid hallucinations/mistakes (that aren't super niche & specific) is the next the models. e.g. take the output from GPT and feed it to Grok, then take that output and feed it to Gemini asking each to fact check through the other's work, streamline it where possible, and to add relevant insight/deeper analysis. It works wonders. 

I know there will be tons of people flooding into this and will tell that it is all BS and just pulled together randomly ... but I also think that you are right. 

I've ran numerous DD's and did hundreds of expert calls and these days, before I hop on the call, I always lay out the land with ChatGPT. I'm still baffled how it gets it about ~80% right and tells me a lot of the exact content that industry experts tell me. 

Best of many examples is just competitive landscape: looked at a very niche pharma related business and problem-solved with ChatGPT the industry landscape. Grouped players into archetypes and drilled down what the key difference is. Then went in with a solid understanding into the expert call and validated most of it. 

It is also very smart in modelling markets ... like that super-smart college friend that you always texted when you needed help on a problem. Again I use it more for sparring, but it got me started on so many market sizing driver trees (e.g., start with variables xyz, multiply by Y to arrive market size for X if you want to validate use following sources, here is also what you can experts to get certainty around it).

let's say you don't know shit about investing, and GPT tells you that it would be a good idea to allocate $1m in X stock. Will you be ready to put all of this or you'll be somehow hesitant? What if it missed some detail like the usual "oh, yes, you're right... this is the updated answer considering ABC" or what if it misses another perspective that must be considered for a specific industry  - "based on the updated input for this situation, investing in C would be more appropiate as the initial stock may cause several issues". 

Also, an example maybe more in line with what you're trying to say is that Renaissance also uses the same data that each one of us has access since decades. We could replicate 100% the entire infrastructure of Renaissance: PhDs, terminals, data, STEM guys, etc. Will we deliver the same "alpha" as them? Probably not. Because everyone has access to the same chart/the same information, but making out something of all this complex system is the hard part/the "brain" power, which very few can

At first you may think, well, this still means that the average investors will be wiped out, but you can't see the blind spots/what's missing/exercise judgement on what GPT is vomiting back to you. So you need still someone who has a strong grasp of finance to challenge whatever AI is delivering, and the strongest his grasp of finance, the better will be his final idea made with AI. It's a tool at the end of the day which follows the same idea of weak inputs = weak results.

And realistically, I can also imagine a future where you have a comparison on AI portfolios who may have 100% the same access to information/etc. but are still delivering different results, which again will prove that it's not the information that leads to alpha, is what you do with that information/how you interpret it/how you invest based on it, which again will leave plenty of room for alpha if you truly have great ideas