This Week In AI News: Nov 22, 2023

In this episode, the DAS crew talked about a range of AI-related topics. The episode featured including Eran, Jyunmi, Beth, Robert, Karl, and Andy, who explored several significant developments in the AI world.

Key Points Discussed:

• Stability AI's Text-to-Video Model: Stability AI announced their foray into text-to-video technology, showcasing impressive demos and comparisons with Runway and Pika. They hinted at capabilities ranging from 14 to 25 frames, indicating significant advancements in AI-generated video technology.

• DERFA - Emotive Avatar Technology: A research group from Singapore University revealed DERFA, a novel technology that combines audio clips with a single photo to create lifelike avatars. This breakthrough demonstrates the integration of audio models and facial expression models to produce natural-looking speech animations.

• Advancements in Large Language Models (LLMs): The Large Model Systems Organization reported a new method that accelerates the generation process of LLMs. This technique, which involves parallel processing, promises over a 2X improvement in speed and efficiency, potentially leading to cost savings and more accessible AI generation.

• Goliath 120B and AI Language Model Developments: The episode also touched upon Goliath 120B, an open-source language model combining two LLAMA2 models, and updates from the Hugging Face leaderboard. These developments underscore the dynamic nature of AI advancements and the importance of diversification in AI sources.

• OpenGPT and ORCA 2 by Microsoft Research: OpenGPT from LangChain and ORCA 2 by Microsoft Research were highlighted. ORCA 2, in particular, showcases advanced reasoning capabilities and instruction-following in a smaller model, paving the way for more efficient AI applications.

• Anthropic's AI Developments: The panel discussed Anthropic's latest release, including advancements in reducing AI "hallucinations" and the challenges in accessing their API. This conversation underscored the ongoing efforts to refine AI models for more reliable and accessible applications.

Action Items:

1. Explore AI Video Technology: For those interested in AI-generated videos, investigating Stability AI's new offerings and comparing them with existing tools like Runway and Pika is advisable.

2. Experiment with Emotive Avatars: DERFA's technology opens new avenues in digital communication and content creation. Professionals should consider how emotive avatars can enhance their digital interactions.

3. Stay Informed on LLMs: The rapid advancements in LLMs, as evidenced by the Large Model Systems Organization's report, necessitate continuous learning and adaptation to leverage these technologies effectively.

4. Diversify AI Sources: Given the multitude of AI models and tools emerging, it's crucial to stay updated with various sources like Goliath 120B, OpenGPT, and ORCA 2 to find the most suitable solutions for different needs.

5. Monitor AI Safety and Efficiency: The developments in reducing AI hallucinations and creating more efficient models highlight the importance of prioritizing safety and efficiency in AI applications.