If worried, run AI locally instead of training the machine. Data is the new oil.
There's a frisson around the idea that we are becoming replaceable, but it goes both ways. If you can replace yourself, you can also scale yourself. The company that would think to replace most of its workforce as a cost cutting measure is more likely to be a business model that customers will eventually replace with AI. The company figuring out how to leverage AI is hiring more people.
Do you have it explicitly pointed at certain sources? I think there is tremendous value in curating and designing a system that incorporates the models with structured data sets and more deterministic workflows as well. I also think a human judgment layer is still key for avoiding mistakes / hallucinations and also catching details that the LLM might see but not weight properly. I don't believe the human in the loop is going to lose value in the near term, in fact I think an expert human in the loop like yourself is actually MORE valuable now (even though the nature of the day to day work is changing)
What I see from most of the investor landscape right now is that LLMs are actually making their decision making more uniform / consensus. That's leading to worse crowding in certain stocks and therefore the resultant volatility is higher on both the upside and downside. To me this provides an amazing opening for human judgment and especially for those who can tolerate drawdowns
I am building several LLM enabled tools, let's chat sometime!
Very interesting. Maybe I'm asking the wrong questions, but I still think it misses some Qualitative stuff, and sometimes Quant stuff too (though I'm not using the latest versions, which may be much better).
My other argument on 'our' side would be - hedge fund managers ALREADY have a ton more IT and manpower resources compared to us lot, and there are STILL inefficiencies. It's simple (but not easy), isn't it - humans/crowds are emotional - so as long as humans are the end users, there should be inefficiencies.
Hi Andrew, I used Codex to create valuation model for $BHR and found the results quite sucessful. How can I share it with you? I plan to make more valuation models for the picks you mentioned and hopefully add some more value to the existing thesis.
The replacement question gets more oxygen than it deserves. AI compresses the part of the work that was always commodity, which is data gathering and first-pass synthesis. What stays is judgement under uncertainty, holding through drawdowns, and refusing to act when the model says act. None of that is automatable, because it is behavioural. The investors who get displaced are the ones who confused activity for skill. That gap is what Capital and Character is about.
This sounds like AI psychosis … the average person is using LLMs to ask if they look good in a various outfits and could not ask for an analysis of a 13-D because they have no idea what a 13-D is.
perhaps the wrong source, but i asked claude based on what and how i built what the right tool for me was, and oit told me cowork because im quite iterative with it
If worried, run AI locally instead of training the machine. Data is the new oil.
There's a frisson around the idea that we are becoming replaceable, but it goes both ways. If you can replace yourself, you can also scale yourself. The company that would think to replace most of its workforce as a cost cutting measure is more likely to be a business model that customers will eventually replace with AI. The company figuring out how to leverage AI is hiring more people.
Idea generation is great and all but in the end investing is about making money. Is your AI better than you at deciding what and when to buy/sell?
Do you have it explicitly pointed at certain sources? I think there is tremendous value in curating and designing a system that incorporates the models with structured data sets and more deterministic workflows as well. I also think a human judgment layer is still key for avoiding mistakes / hallucinations and also catching details that the LLM might see but not weight properly. I don't believe the human in the loop is going to lose value in the near term, in fact I think an expert human in the loop like yourself is actually MORE valuable now (even though the nature of the day to day work is changing)
What I see from most of the investor landscape right now is that LLMs are actually making their decision making more uniform / consensus. That's leading to worse crowding in certain stocks and therefore the resultant volatility is higher on both the upside and downside. To me this provides an amazing opening for human judgment and especially for those who can tolerate drawdowns
I am building several LLM enabled tools, let's chat sometime!
Very interesting. Maybe I'm asking the wrong questions, but I still think it misses some Qualitative stuff, and sometimes Quant stuff too (though I'm not using the latest versions, which may be much better).
My other argument on 'our' side would be - hedge fund managers ALREADY have a ton more IT and manpower resources compared to us lot, and there are STILL inefficiencies. It's simple (but not easy), isn't it - humans/crowds are emotional - so as long as humans are the end users, there should be inefficiencies.
Hi Andrew, I used Codex to create valuation model for $BHR and found the results quite sucessful. How can I share it with you? I plan to make more valuation models for the picks you mentioned and hopefully add some more value to the existing thesis.
The replacement question gets more oxygen than it deserves. AI compresses the part of the work that was always commodity, which is data gathering and first-pass synthesis. What stays is judgement under uncertainty, holding through drawdowns, and refusing to act when the model says act. None of that is automatable, because it is behavioural. The investors who get displaced are the ones who confused activity for skill. That gap is what Capital and Character is about.
This is a very interesting and thoughtful essay, AW. Set an alarm to go back and reread this to yourself in a year's time.
Andrew, I can name, off the top of my head, a dozen FinTwit dunderheads that I want AI to replace instead of you
This sounds like AI psychosis … the average person is using LLMs to ask if they look good in a various outfits and could not ask for an analysis of a 13-D because they have no idea what a 13-D is.
I've been doing something similar. Why did you choose claude cowork over code?
No coding background; claude code is basically hieroglyphics for me
Same with me. Give it a shot I managed to code an entire dashboard with no experience.
perhaps the wrong source, but i asked claude based on what and how i built what the right tool for me was, and oit told me cowork because im quite iterative with it