3 ms·

Damn that's cool. Can we implement a version of System 1 / System 2 [1] thinking with this? From the future work section: > better determining the number of <

Damn that's cool. Can we implement a version of System 1 / System 2 [1] thinking with this? From the future work section: > better determining the number of <pause> tokens (perhaps using model confidence) If the number of pause are learned, using model confidence + regularisation (so that the model doesn't always use the maximum number of pauses), then we effectively have the System 1/2 switch. If it's a task the model has seen tons of times before, it just goes with the first inference pass. If it's low confidence, then it keeps expending more inference passes until it reaches an acceptable confidence threshold. - [1] https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow

striking 3y ago
Would that be in an effort to more accurately model human thinking? If so, I hate to be the one to tell you that > Readers of “Thinking: Fast and Slow” should read the book as a subjective account by an eminent psychologists, rather than an objective summary of scientific evidence https://replicationindex.com/2020/12/30/a-meta-scientific-perspective-on-thinking-fast-and-slow/ https://replicationindex.com/2020/12/30/a-meta-scientific-pe...
- silveraxe93 3y ago
  I actually read that one before! Kahneman is overconfident, as we all were before the replication crisis. While most of the psychology field is crumbling and filled with bullshit (even though most people didn't catch on yet). There's still _some_ truth lying in there. It might look bad in isolation, but Kahneman's work is one of the few that actually holds up to scrutiny. This post by Scott Alexander is pretty good in summarising the few good bits left. https://www.astralcodexten.com/p/heres-why-automaticity-is-real-actually https://www.astralcodexten.com/p/heres-why-automaticity-is-r... --- But regarding the initial point. The goal is not to fully emulate human thinking. It might sound wishy-washy, but there's _obviously_ _some_ truth to the system 1/2 model. It's not perfect! But I think it's useful. We as humans do most decisions without thinking (hard), but we have a way to _switch_ into a more reliable but expensive mode. We see some evidence that artificially inducing LLMs to 'think' more improves their output. So it stands to reason that adding this capability to a model would make it better. And what better way to do it than swallowing the bitter pill and have that decision be made by the model itself, on a case-by-case basis by learning it from data.
  - airstrike 3y ago
    Just don't give them a bicameral mind and we should be safe.

striking 3y ago

Would that be in an effort to more accurately model human thinking? If so, I hate to be the one to tell you that > Readers of “Thinking: Fast and Slow” should read the book as a subjective account by an eminent psychologists, rather than an objective summary of scientific evidence https://replicationindex.com/2020/12/30/a-meta-scientific-perspective-on-thinking-fast-and-slow/ https://replicationindex.com/2020/12/30/a-meta-scientific-pe...

silveraxe93 3y ago
I actually read that one before! Kahneman is overconfident, as we all were before the replication crisis. While most of the psychology field is crumbling and filled with bullshit (even though most people didn't catch on yet). There's still _some_ truth lying in there. It might look bad in isolation, but Kahneman's work is one of the few that actually holds up to scrutiny. This post by Scott Alexander is pretty good in summarising the few good bits left. https://www.astralcodexten.com/p/heres-why-automaticity-is-real-actually https://www.astralcodexten.com/p/heres-why-automaticity-is-r... --- But regarding the initial point. The goal is not to fully emulate human thinking. It might sound wishy-washy, but there's _obviously_ _some_ truth to the system 1/2 model. It's not perfect! But I think it's useful. We as humans do most decisions without thinking (hard), but we have a way to _switch_ into a more reliable but expensive mode. We see some evidence that artificially inducing LLMs to 'think' more improves their output. So it stands to reason that adding this capability to a model would make it better. And what better way to do it than swallowing the bitter pill and have that decision be made by the model itself, on a case-by-case basis by learning it from data.
- airstrike 3y ago
  Just don't give them a bicameral mind and we should be safe.