Abstract: Recent months have seen the emergence of a powerful new trend in which large language models (LLMs) are augmented to become autonomous language agents capable of performing objective oriented multi-step tasks on their own, rather than merely responding to queries from human users. Most existing language agents, however, are not optimized using environment-specific rewards. Although some agents enable iterative refinement through verbal feedback, they do not reason and plan in ways that are compatible with gradient-based learning from rewards. This paper introduces a principled framework for reinforcing large language agents by learning a retrospective model, which automatically tunes the language agent prompts from environment feedback through policy gradient. Specifically, our proposed agent architecture learns from rewards across multiple environments and tasks, for fine-tuning a pre-trained language model which refines the language agent prompt by summarizing the root cause of prior failed attempts and proposing action plans. Experimental results on various tasks demonstrate that the language agents improve over time and that our approach considerably outperforms baselines that do not properly leverage gradients from the environment. This demonstrates that using policy gradient optimization to improve language agents, for which we believe our work is one of the first, seems promising and can be applied to optimize other models in the agent architecture to enhance agent performances over time.
* Outlier values are used to prune values.
* Transformers seem to undergo a "phase shift" in how outlier features are treated around 6.7B parameters. This could complicate research on removing them.
Maybe you and Tim Dettmers would have a lot to talk about :)
I'm not sure SF is a good example. It's not a healthy city, but it's problems go way beyond drug use and it doesn't have the same policies as what Oregon adopted.
A lot of comments are discussing the difficulty in estimating range accurately or how all EPA estimates are inflated. But the article claims Tesla knowingly uses an algorithm with inflated numbers and swaps the rost estimate out for a more accurate estimate at 50% charge. That's different than a good faith attempt at estimating range and a dark pattern.
The article says it was a mandate from Elon Musk. Not sure I believe the claim, but I'd also be surprised if he wasn't aware just how optimistic the EPA estimate is.
I get that there are multiple endpoints being tested here and some of them may have been satisfied, but I feel like I'm living in a bizarro world on hacker news where people are arguing that multiple failed engines, a failed stage detachment and an exploding rocket worth millions of dollars and and carrying a large number of limited supply raptor engines is a "massive success". Can we just call it for what it is? Mixed results maybe? Is that not a fair assessment?