5 ms·

A better statement might be: Current agentic systems may be built from next-token predictors which are conceptually simple, but because of agentic framew

A better statement might be: Current agentic systems may be *built* from next-token predictors which are conceptually simple, but because of agentic frameworks, recursive invocation, tool use, and *heavy* investment in reinforcement learning in these contexts and for specific applications, they can no longer be thought of as "Merely" next token predictors. Modern agentic work is probably more of a "emergent system from simple rules and complex interactions" paradigm than a genuinely new technology.

danielmarkbruce 12d ago
Nope. Post training means even the raw model isn't predicting.
- jvanderbot 12d ago
  In what sense is the model not predicting?
  - danielmarkbruce 12d ago
    There isn't a truth to test against. If I predict the next word in a sequence is "sat", we can check against the sequence. If I predict the roll of a die will be 4, we can check against it. Whether i give 100% or give a probabilistic prediction, we can check against the truth. If I choose a specific move in chess, it's a choice. It's not a prediction. I might get a score 40 moves later given my choice, but I'm not predicting the next move. To compare - during pre-training, the model literally tries to predict the next token (probabilistically), the training loop checks against the "right" answer, and the weights are updated based on that check. It's optimized to predict the next token.
    - jvanderbot 12d ago
      Oh, that's somewhat philosophical. The prediction here is a mechanical thing, just random sampling along a pre-calibrated mapping of options given the tokens and their importance weights. That sampling can be called prediction if you use it that way, or it can be called thinking if you are into the psychosis too deeply. Or it can be called a CLI tool command if you're in an agentic framework
      - danielmarkbruce 12d ago
        No, it's not philsophical. Because if you optimize to predict, you are doing something different to optimizing for a reward. It's a different process - different objective function, different optimization, different set up.
        jvanderbot 11d ago
        [dead]
whateveracct 12d ago
the next token part is like the "clock" that drives it

A better statement might be: Current agentic systems may be *built* from next-token predictors which are conceptually simple, but because of agentic framew

danielmarkbruce 12d ago

Nope. Post training means even the raw model isn't predicting.

jvanderbot 12d ago
In what sense is the model not predicting?
- danielmarkbruce 12d ago
  There isn't a truth to test against. If I predict the next word in a sequence is "sat", we can check against the sequence. If I predict the roll of a die will be 4, we can check against it. Whether i give 100% or give a probabilistic prediction, we can check against the truth. If I choose a specific move in chess, it's a choice. It's not a prediction. I might get a score 40 moves later given my choice, but I'm not predicting the next move. To compare - during pre-training, the model literally tries to predict the next token (probabilistically), the training loop checks against the "right" answer, and the weights are updated based on that check. It's optimized to predict the next token.
  - jvanderbot 12d ago
    Oh, that's somewhat philosophical. The prediction here is a mechanical thing, just random sampling along a pre-calibrated mapping of options given the tokens and their importance weights. That sampling can be called prediction if you use it that way, or it can be called thinking if you are into the psychosis too deeply. Or it can be called a CLI tool command if you're in an agentic framework
    - danielmarkbruce 12d ago
      No, it's not philsophical. Because if you optimize to predict, you are doing something different to optimizing for a reward. It's a different process - different objective function, different optimization, different set up.
      - jvanderbot 11d ago
        [dead]

whateveracct 12d ago

the next token part is like the "clock" that drives it