The most important weird thing about LLMs is that they are so general. A bigger LLM that is better at coding is also better at ideation & ethical advice & medicine & math. This isn’t true of everything, jaggedness again (see fiction writing!), but it is remarkably true.
No Digg Deeper questions have been answered for this story yet.
The entire business of the AI Labs & the potential speed of transformation stems from this fact.
The most important weird thing about LLMs is that they are so general. A bigger LLM that is better at coding is also better at ideation & ethical advice & medicine & math. This isn’t true of everything, jaggedness again (see fiction writing!), but it is remarkably true.
通才模型的定价权不在于某一项能力的领先，而在于所有领域同时进步的边际成本趋近于零。Google限制API额度这条线，本质上也是这个逻辑的镜像：需求端涌得太快，供给端跟不上了。推理成本下降会加速渗透，渗透率上去后跑出来的数据又反哺模型，这才是AI真正的正反馈闭环。所以算力短缺不是阶段性的瓶颈，是结构性常态。
@emollick the exception might matter more than the rule if fiction is the crack in the pattern, that's probably telling us something real about what these things are actually doing under the hood
@emollick This is also why the bottleneck shifts from model quality to execution quality. Once the model is broadly capable the product gap becomes routing approvals and reliable handoff in real work.
@emollick That is also why infra spend follows capability gains. The broader the model gets the more valuable the execution layer becomes routing approvals recovery and human handoff.
@emollick its almost like breadth itself is the training signal for depth
every new domain sharpens the weird shared sense of what makes sense