5 ms·

I don’t think the issue of Google is a lack of advanced AI technology but the transfer and realization of products. So many Google’s products are half-backed, o

I don’t think the issue of Google is a lack of advanced AI technology but the transfer and realization of products. So many Google’s products are half-backed, only half-heartedly developed and then be buried at the end. For a giant like Google, they must be stuck in a rut and to get free need all the help they can get. If they continue the business as usual, I afraid they will experience a never-seen before landslide. I believe calling back the founders at this critical time is the right move.

westurner 4y ago
I disagree with your assessment of their comparative performance. "Beyond Tabula Rasa: Reincarnating Reinforcement Learning" https://ai.googleblog.com/2022/11/beyond-tabula-rasa-reincarnating.html?m=1 https://ai.googleblog.com/2022/11/beyond-tabula-rasa-reincar... : > Furthermore, the inefficiency of tabula rasa RL research can exclude many researchers from tackling computationally-demanding problems. For example, the quintessential benchmark of training a deep RL agent on 50+ Atari 2600 games in ALE for 200M frames (the standard protocol) requires 1,000+ GPU days. As deep RL moves towards more complex and challenging problems, the computational barrier to entry in RL research will likely become even higher. > To address the inefficiencies of tabula rasa RL, we present “Reincarnating Reinforcement Learning: Reusing Prior Computation To Accelerate Progress” at NeurIPS 2022. Here, we propose an alternative approach to RL research, where prior computational work, such as learned models, policies, logged data, etc., is reused or transferred between design iterations of an RL agent or from one agent to another. While some sub-areas of RL leverage prior computation, most RL agents are still largely trained from scratch. Until now, there has been no broader effort to leverage prior computational work for the training workflow in RL research. We have also released our code and trained agents to enable researchers to build on this work. Feed-Forward with Prompt Engineering is like RL; which prompt elements should remain given objective or subjective error?

westurner 4y ago

I disagree with your assessment of their comparative performance. "Beyond Tabula Rasa: Reincarnating Reinforcement Learning" https://ai.googleblog.com/2022/11/beyond-tabula-rasa-reincarnating.html?m=1 https://ai.googleblog.com/2022/11/beyond-tabula-rasa-reincar... : > Furthermore, the inefficiency of tabula rasa RL research can exclude many researchers from tackling computationally-demanding problems. For example, the quintessential benchmark of training a deep RL agent on 50+ Atari 2600 games in ALE for 200M frames (the standard protocol) requires 1,000+ GPU days. As deep RL moves towards more complex and challenging problems, the computational barrier to entry in RL research will likely become even higher. > To address the inefficiencies of tabula rasa RL, we present “Reincarnating Reinforcement Learning: Reusing Prior Computation To Accelerate Progress” at NeurIPS 2022. Here, we propose an alternative approach to RL research, where prior computational work, such as learned models, policies, logged data, etc., is reused or transferred between design iterations of an RL agent or from one agent to another. While some sub-areas of RL leverage prior computation, most RL agents are still largely trained from scratch. Until now, there has been no broader effort to leverage prior computational work for the training workflow in RL research. We have also released our code and trained agents to enable researchers to build on this work. Feed-Forward with Prompt Engineering is like RL; which prompt elements should remain given objective or subjective error?