From Noah Smith: Generative AI and the finance industry. (Unfortunately, most of the article is paywalled, which has become an annoying and worsening trend with Substack and online journalism overall lately.)
I am not concerned AI will make my job obsolete or supplant my income. More computing power does not mean better strategies. This is because, first, the search space grows exponentially even as the amount of data and parameters grows linearly, so even with unlimited GPUs you will still run into the constraints of hardware. Most of the time, more computing power is not as helpful as it seems because whoever is using it is looking in the wrong areas (e.g. looking for life on Mars or casting a larger fishing net in the Dead Sea), or has the wrong conceptual or mental model of the world (the garbage in, garbage out problem). Life is not like chess, in which more computation generally equals more wins.
Yes, people have actually lost money insider trading and gotten into trouble anyway, as evidence that more information is not always better, that dynamic systems such as financial markets are dominated by outside and secondary effects, or that misapplied information is worse than no information. [Let’s assume you know that a company will beat its EPS estimate. Great! So you buy the stock or call options anticipating a large positive movement, but it fails to beat the ‘whisper number’ or the CEO makes negative remarks, and the stock crashes instead. This happens a lot.]
Third, macro-economic fundamentals and firm-specific fundamentals will likely supersede AI. Profits and earnings are a floor for how low a stock can fall, and produces an upward drift in asset prices over the long term (which is why equities tend to always outperform commodities). The book value acts as a sort of floor for what a stock is worth. Low interest rates and other factors conductive to rising (or falling) stock prices will also prevail over AI. Having more traders, including fully automated traders, may increase liquidity and market efficiency, at least for short-term timeframes.
AI has been used in finance for quite a while, of course. After all, AI is really about prediction, and that’s how you make money in the trading world. Renaissance Technologies, arguably the best quant trading firm in the world, has reportedly had some success using machine learning techniques, including natural language processing; two recent CEOs of the company, Robert Mercer and Peter Brown, were AI researchers. Other firms have followed suit, and there has generally been a lot of interest in the idea of using AI to trade. But as the WSJ’s Gregory Zuckerman recently reported, AI tools have been of surprisingly limited use so far:
I am running a one-man Renaissance Technologies. Like Renaissance, my fund/methods is only limited to myself and family (been trying to get my dad to run a more conservative/unleveraged version of the Bitcoin method). But unlike Renaissance, no need for complicated algos or having to make thousands of trades/day. I only make four trades/day with my Bitcoin method: two trades to open the positions and two to close it at the end of the day, repeat.  I also have other strategies, like 3x funds and Meta on other accounts. This involves hardly any trades. Simpler strategies tend to prevail over more complicated ones, but the huge success of Renaissance obviously shows that complicated strategies can work too. I think AI will add unnecessary complexity and hurt returns. Renaissance works so well because its models are superior, not because of AI. Again: garbage in, garbage out.
AI will probably never be able to replace true creativity, talent, or ingenuity. Talent will always command a premium even if there are cheaper alternatives or more content being produced overall. Photoshop has been around forever, and it’s very cheap to hire overseas labor to produce graphics, yet top artists still hold on, and some contemporary artists earn millions (such as modern art, not just masterpieces from antiquity). Since 2021, the dozen or so Substack writers who I follow are seeing record subs and traffic (as measured by ‘likes’ and comments), such as Noah Smith above, despite how LLMs are optimized for that very task.
I also posit AI cannot compete with a human endowed with superior IQ and pattern recognition abilities, because humans can have ‘epiphanies’ in way that an AI cannot, or tie together seemingly unrelated or disparate pieces of information. A few days ago I identified the so-called ‘compressed candlestick pattern’ when BTC was at $29,400, which portends to much more downside. BTC has since fallen $2,500 points in under a week, with no end in sight to the selling even as the Nasdaq keeps going up. What are the odds that even the most powerful array of GPUs running 24-7 would have noticed this, but I did? 2023 has been a record year for me with trading just shorting Bitcoin and long QQQ, as described above. As discussed yesterday, despite hedge funds having unlimited resources to find strategies, including AI, I still found a big edge. There will always be opportunities and profits for people with an intuition of how things work, and know where to look.
Another hugely profitable day today, having shorted BTC at the open at $27,400 and closed at $26,900.