Jason Wei (@_jasonwei)

Cognitive reward shapes in sports and career Sports are amazing environments to learn. When you play a sport for thousands of hours, you start to see the world through that sport. It is a simple fact—your biological neural network is being conditioned to respond to the behavior incentivized by the rules of the sport. The funny thing is that most people choose their sports for accidental reasons such as parents, geography, or school programs. People rarely think about how the particular sport you play influences how your brain thinks more generally. Going a step further, playing the right sport may even benefit your career. My two favorite sports are tennis and soccer. Tennis is one of the best sports for teaching consistency. In tennis, there are hundreds of points in a match, and each point is worth exactly one unit, regardless of whether your opponent made an unforced error or if you constructed the most beautiful point ending with a winner. Tennis is low-variance optimization—you win by reducing unforced errors, playing percentages, and grinding out small advantages. Tennis is also an individual sport, which teaches you to rely on yourself consistently. Tennis has a similar cognitive reward shape to professions like being a surgeon or a pilot. Surgery and aviation require consistency, self-accountability, and deep focus. And similar to how you can only win one point at a time in tennis no matter how spectacular it was, there is no extra credit for the best appendectomy or the smoothest SFO-JFK flight. Your craft is to provide consistency with very low tolerance for error. On the other hand, the tennis mindset transfers relatively little to entrepreneurship. Entrepreneurship is a high-variance, team game where failure is tolerated and occasional creativity gets rewarded exponentially. Minimizing unforced errors in tennis is a totally different mindset from deciding whether to make a moonshot business move that will likely fail but could potentially net a billion dollars. Obviously I am not saying that tennis players cannot be great entrepreneurs, but I do think it is a totally different cognitive reward shape. Being a forward in soccer has a much closer reward shape for entrepreneurship. What a forward in soccer learns is to create many small chances. It is a fact that most of the game, you are not scoring—even if you look at all the times that Mbappe got on the ball in one of his best games, most of those led to nothing! But all that matters is creating enough chances to score once (or a few times) and win the game. If you break down a 90-minute game for a forward, almost all the time is failure or noise, a few minutes will be leverage, and a few seconds will determine the fate of the game. I have not played soccer for thousands of hours, but I can imagine that being a lifetime forward in soccer would teach you to be comfortable with failure and asymmetric returns. In summary, I am claiming that there can be substantial value when the cognitive reward shape of your sport mirrors that of your career. I’ll admit that I’ve done some cherry-picking for illustration purposes—entrepreneurship also requires consistency and error avoidance; and goalies in soccer have reward shapes that are very different from strikers. But I think the point stands. If sports shape how we perceive risk, effort, and reward, then we should choose them wisely.

When language models first started using tools well, I was sympathetic to the narrative that instead of scaling up language models, all we needed was a strong enough "cognitive core", say 1B parameters, and anything else could be done with tool use, like browsing the internet or executing code. I think a lot of people were sympathetic to this argument, and indeed it is pretty hard to come up with a meaningful task that cannot be in principle achieved by a 1B model with adequate access to tools. For example, any esoteric fact that a large language model would know can be, in principle, retrieved from the internet and reasoned over by a 1B language model. However I now think this is totally wrong for one simple reason: doing tasks quickly and naturally without tool use matters a lot. The way that I internalized this reason was actually in my personal journey learning badminton this year. In badminton I am very much like a "1B cognitive core". While I can physically do every movement in a badminton shot that my coach teaches me, it requires a lot of work to mentally remember every cue and put it together. In practice I can do a shot almost perfectly, but I struggle to do it across a point and I definitely can't do it consistently in a game. This is obviously different from someone who has practiced a shot ten-thousand times and effortlessly executes it as a natural instinct. In the same way, language models knowing a fact internally, without tool calls, is meaningful. The first reason is that we obviously care about speed; you'd much rather get an answer immediately than have the model think a long time to be sure of its answer or browse the web. A second reason is that there are some things that are simply best learned via backpropagation over lots of data. If you ask about how people generally think of the Shambhala music festival, you'd rather a large language model give you an aggregate opinion based on all the data on the internet, than get a regurgitation of the first three reviews that show up in a web search. A third reason is that having to do a lot of work to find an answer is not as reliable as already knowing the answer. While this does not have to be true in theory, it is probably true in practice, at least for now. If you have to re-look up facts or redo a mathematical derivation all the time there is a higher chance of mistakes, which can compound in a long-horizon task. Once you buy that it is valuable to do things parametrically without tool use, then you must buy the argument that a 1B cognitive core is not sufficient. There is an information limit to how much knowledge can be internalized by a 1B model, and we will surely want AI to know more than that. Even 1T probably won't be enough. We will want the AI to know as much about our world as possible, we will want it to be updated with new information, and our expectations of what AI can do for us will continue to grow. In summary, tool use enables small models to do a lot more, but those who demand the highest quality intelligence will always want larger models. Bitter lesson strikes again.