AGENTS - AI Dynamics

What do AGI companies want to become?

By

@sama

–

10 November 2022 2h13

what do AGI companies wanna be?

→ View original post on X — @sama

10 November 2022

Cohere Thanksgiving Hackathon: Build AI for Social Good

By

@cohere

–

09 November 2022 15h00

Cohere x Lablab Hackathon: A Thanksgiving Special Get inspired by Thanksgiving! We encourage you to reflect on what you are thankful for and build something that will benefit and support those around you. Let's give back together! https://
lablab.ai/event/cohere-t
hanksgiving-hackathon
…

→ View original post on X — @cohere

9 November 2022

Simulation Environments with Thermodynamics for RL Research

By

@nandodf

–

09 November 2022 0h27

Question: Can anyone recommend simulation environments for AI research where the second law of thermodynamics applies? It feels like this is needed to avoid collapse of some RL exploration strategies and intrinsic motivation. https://
flip.it/SodkpS

→ View original post on X — @nandodf

9 November 2022

Command Beta: New Zero-Shot AI Command Capability Released

By

@cohere

–

08 November 2022 20h26

Command Beta now available → a new capability that responds to single-sentence commands (i.e., zero-shot prompts). We have some sample shots to get you started → https://
hubs.li/Q01rQq1q0. Sign up to try it out → https://
hubs.li/Q01rQfsR0

→ View original post on X — @cohere

8 November 2022

ReAct: Synergizing Reasoning and Acting in Language Models

By

@googleai

–

08 November 2022 19h53

Learn how ReAct, a general paradigm for synergizing reasoning and acting in language models, presents more interpretable, diagnosable, and controllable task-solving trajectories while outperforming reasoning and acting only paradigms. Read the blog → https://
goo.gle/3TiNcDp

→ View original post on X — @googleai

8 November 2022

Inspiration from Ajeya Cotra’s Sandwiching Concept in Research

By

@anthropicai

–

08 November 2022 17h33

This paper was heavily inspired by prior work, especially Ajeya Cotra's 'sandwiching' concept:

→ View original post on X — @anthropicai

8 November 2022

Scalable Oversight: Supervising AI Systems Beyond Human Capabilities

By

@anthropicai

–

08 November 2022 17h33

To ensure that AI systems remain safe as they start to exceed human capabilities, we’ll need to develop techniques for scalable oversight: the problem of supervising systems’ behavior without assuming that the overseer understands the task better than the system being trained.

→ View original post on X — @anthropicai

8 November 2022

Leveraging Model Capabilities While Maintaining Supervisory Control

By

@anthropicai

–

08 November 2022 17h33

This is tricky: To do this, we’ll need ways for the human who’s supervising the model to use any relevant knowledge or skills that the model already has, even though they can’t trust the model to be reliably helpful.

→ View original post on X — @anthropicai

8 November 2022