4 ms·

>The method signature at the end doesn't matter you are right there but what the model is trained to do matters a ton. You're being loose with semantics and dr

>The method signature at the end doesn't matter you are right there but what the model is trained to do matters a ton. You're being loose with semantics and drawing bad conclusions. The training describes the nature of the feedback given to the system. Being trained to predict tokens means it gets feedback on the quality of its predictions. This does not constrain or determine the structure of the learned model. The learned model is a function of the training time, quality and amount of data, as well as the architecture that is biased towards capturing certain features or entails limits on what can be captured. The point is, the space and properties of solutions to the problem of "predicting the next token" is not constrained by the nature of the feedback.

Jensson 3y ago
> This does not constrain or determine the structure of the learned model It absolutely does! A human can learn to perform an action by reading a manual. But an LLM is only trained to predict tokens, if you feed it a manual it will learn to recite the manual, it wont learn how to perform the actions in the manual. This makes the training inherently flawed from a learning perspective, and that flaw is impossible to fix without moving past the token prediction way of training models.
- hackinthebochs 3y ago
  >if you feed it a manual it will learn to recite the manual, it wont learn how to perform the actions in the manual. This claim doesn't track with LLM-based agents. For example: https://towardsai.net/p/machine-learning/meet-webagent-deepminds-new-llm-that-follows-instructions-and-complete-tasks-on-websites https://towardsai.net/p/machine-learning/meet-webagent-deepm...
  - Jensson 3y ago
    The whole topic is about training time, not the tokens you feed it during runtime. Don't think that it can reason about the stuff it was trained on the same way it can reason about tokens you give it. They can't. Any arguments based on what they can do with runtime token input are therefore not relevant.
    - hackinthebochs 3y ago
      Your objection is getting less and less clear. The issue is the LLM's capabilities at runtime given that it was trained to predict tokens. LLM agents demonstrate that such a training regime induces a system that can capably follow instructions. Nothing you've said is counter to this.
- K0balt 3y ago
  Mmmmm… llms appear to be able to apply the information gleaned from textbooks, in my experience.

Jensson 3y ago

> This does not constrain or determine the structure of the learned model It absolutely does! A human can learn to perform an action by reading a manual. But an LLM is only trained to predict tokens, if you feed it a manual it will learn to recite the manual, it wont learn how to perform the actions in the manual. This makes the training inherently flawed from a learning perspective, and that flaw is impossible to fix without moving past the token prediction way of training models.

hackinthebochs 3y ago
>if you feed it a manual it will learn to recite the manual, it wont learn how to perform the actions in the manual. This claim doesn't track with LLM-based agents. For example: https://towardsai.net/p/machine-learning/meet-webagent-deepminds-new-llm-that-follows-instructions-and-complete-tasks-on-websites https://towardsai.net/p/machine-learning/meet-webagent-deepm...
- Jensson 3y ago
  The whole topic is about training time, not the tokens you feed it during runtime. Don't think that it can reason about the stuff it was trained on the same way it can reason about tokens you give it. They can't. Any arguments based on what they can do with runtime token input are therefore not relevant.
  - hackinthebochs 3y ago
    Your objection is getting less and less clear. The issue is the LLM's capabilities at runtime given that it was trained to predict tokens. LLM agents demonstrate that such a training regime induces a system that can capably follow instructions. Nothing you've said is counter to this.
K0balt 3y ago
Mmmmm… llms appear to be able to apply the information gleaned from textbooks, in my experience.