reward signal
Feedback that indicates whether behavior was good or bad, enabling improvement; without it, one keeps repeating the same approach.
As defined in AMA: career advice given AGI, how I research ft. Sholto & Trenton
Where it comes up · 3
- #416: The Relentless Missionary Creating AGI: Demis Hassabis2 claims
- Richard Sutton – Father of RL thinks LLMs are a dead end1 claim
- AMA: career advice given AGI, how I research ft. Sholto & Trenton1 claim
Most important claims · 4 of 4Search all claims for “reward signal” →
- AlphaGo Zero was trained exclusively through self-play using random moves and a reward signal rather than expert human games, and it outclassed its predecessor by a wide margin, discovering strategies unknown to human players.#416: The Relentless Missionary Creating AGI: Demis Hassabis
- DeepMind started with Atari video games as its testbed because the simple graphics kept computing needs affordable, the dozens of games offered many opportunities to prove generality, and constantly updating scores gave the agent the feedback it needed to learn.#416: The Relentless Missionary Creating AGI: Demis Hassabis
- In reinforcement learning, there is a right thing to do because the right thing to do is the thing that gets you reward; we have a definition of what's the right thing to do, so we can have prior knowledge provided by people about what the right thing to do is, and then we can check it against the reward signal.Richard Sutton – Father of RL thinks LLMs are a dead end
- The most important feedback loop for the podcast is not audience compounding or repetition, but that the work is good enough to earn access to interesting people who become friends and teach the host things, who then enable better episodes and access to new fields—a relationship flywheel that recently netted a whole China network from a single blog post.AMA: career advice given AGI, how I research ft. Sholto & Trenton