The biggest problems in financial markets x AI is the measurement problem
Markets are so complex that it's unclear what to measure and how to measure it accurately
A successful backtest could be a sign of a great strategy, but it probably isn't. More likely, its a sign of future leakage, overfitting, or luck
If you want to create a powerful learning flywheel, you need to solve this
It's part of why markets haven't "fallen" to RL yet, like code and math. In code and math, "good" is easier to disambiguate from luck or cheating
We've spent the majority of our time on this problem, and we finally made the necessary breakthroughs. You'll be able to feel the effects of this when we release our first family of models
AI labs and GPU providers are making lots of money selling tokens, but who's making money buying them?
In order to justify the next leg of token spend, companies have to reorganize around tokens as a driver of revenue vs. just increasing efficiency. This is especially true as models get more expensive & longer horizon.
There are some domains that are better positioned than others. Even in those domains, companies are just now starting to adapt. This post dives into the details
The difficulty is, the things it needs help with change over time, sometimes gradually, sometimes suddenly
Right now they are changing quickly, so a big part of the job is staying ahead of the meta
In return, capital offers promotion up the socio-economic ladder, increasing your standard of living
For most of recent history, capital wanted you to build infrastructure, write code, analyze data, make sales calls and generally engage in traditional labor
Now, it can do those things itself, so it’s asking for your judgement to make more asymmetric allocation decisions. It wants to concentrate itself around the areas where more of itself matters the most
Capital whispers, it reveals where you will be most handsomely rewarded
I’m telling you that capital wants you at the fringes, beyond the overton
It's asking you to ignore what most people are doing and identify what they have overlooked. It wants your help in formalizing and investing in the things that aren't obvious yet, but will be eventually
At Nof1 we believe the next breakthrough in AI after reasoning is adaptation
If reasoning is the ability to detect patterns and leverage them, adaptation is anticipating how those patterns will change
This is a crucial capability for real world AI, and current LLMs struggle with it. We've run thousands of live market experiments and no amount of context, prompting, or harness engineering has worked. So, we've been pushing our research deeper down the stack
In this paper, we trained models that can better adapt in dynamic environments like markets. They learn to value an action by the futures it opens up, not just its immediate payoff
It's an early step, but we're building towards forever learners that are default forward-looking, rather than static and backward looking
1/6 🧵
How do you train a language model to become a self-adaptive, long-horizon optimization engine for executable trading policy search?
In our new paper, "The Time Value of Evolution," we introduce Lineage-Value Policy Gradients, or LVPG, a new long-horizon RL technique for adaptive evolutionary search.
I'll be on TBPN today at 3:25pm EST (in ~1 hr)
More details on the platform we're building, what LLMs are good/bad at in investing, Alpha Arena S2
Come through
Happy Monday. On today's show:
- @FerdiDabitz (Augustus)
- @spencerrascoff (Match Group)
- @IPlayedD1 (Consensus)
- @lohstroh (Giga Energy)
- @jay_azhang (Nof1)
- @_amirabs (Astrocade)
- @BaijuBhatt (Cowboy Space)
- @alextaubman (Long Lake)
- @quaidwalker (Bezel)
See you soon.
Great writeup, thanks for the shoutout @chamath
Alpha Arena Season 2 coming soon 👀
You’re absolutely right. I messed up everything. Every line of code I wrote is incorrect and misses the fundamental nature of what you are trying to build. Let me go back and fix these critical mistakes which includes, but is not limited to, virtually everything I’ve said and done up until this point.
Wait. Upon further reflection, I was right. You were right. Everything is perfect and 100% working as expected. Just to be sure, I'm going to rewrite your entire code base from scratch.