AI Weekly Insights #17

Video Breakthroughs, Local Models, and Enhanced Memory

Author

Kharee Smith
February 16, 2024

Happy Sunday!

Coming at you live at whatever time you’re reading this with this week’s 17th Edition of “AI Weekly Insights”! This week, we're diving into OpenAI's latest marvel, Sora, Nvidia's "Chat with RTX" for AI on PCs, and a fresh update to ChatGPT. Explore tips for Prompt Engineering near the end as well.

For the Week of 02/11/24 - 02/17/24 (P.S. Click the story’s title for more information 😊):

OpenAI Showcases New Text-to-Video AI Model, Sora

  • What’s New: OpenAI introduces Sora, a text-to-video model that crafts videos with remarkable clarity and detail.

  • Text-To-Video: Sora can produce videos up to a minute long, encompassing complex scenes and motions with physics. It can animate still images and extend existing videos. Currently, the model is in the testing phase, addressing concerns like misinformation and bias before becoming more generally available.

  • Why It Matters: This is huge, and I have a few thoughts. As MKBHD insightfully notes, “This is the worst that this technology will be from here on out.” 

  • To me, this is both impressive and kind of scary at the same time. I am in awe at the things that can be achieved with AI. The innovation potential is limitless and I’m fairly optimistic about its future as a force of good for humanity.

  • It’s also hard not to look at these tools and imagine the potential for abuse. To the untrained eye an AI-generated video could easily be passed off as legitimate, how would you even know what is real anymore? This also poses a disruption to the work of videographers, animators, drone operators, etc. Why pay for footage when it can be generated with a prompt? These certainly deserve a larger conversation about the implications for content creators and the importance of establishing ethical guidelines.

  • I do think this technology is very cool though, and you should check out a few if you have not:

Video Credits: OpenAI/Sora

Video Credits: OpenAI/Sora

Video Credits: OpenAI/Sora

  • What's New: Nvidia has released a tool for GeForce RTX series owners to run generative AI models offline.

  • Chat with RTX: The tool enables interaction with documents and inquiries within a local setting, supporting various text-based AI models, including Meta’s Llama 2. The tool supports various file formats and requires around 50-100 GB to download.

  • Why It Matters: Currently, this is more of a tech demo than anything. The tool cannot remember context when asked follow-up questions, and the quality of its responses relies on a range of factors. Its importance is that it highlights a growing trend towards local AI applications, offering a glimpse into future capabilities without internet dependency.

  • What's New: OpenAI is experimenting with memory features for ChatGPT, allowing it to retain information across conversations.

  • Memory + Controls: With memory enabled, ChatGPT can remember specific details that you input or naturally learn over time. There will be controls to delete items from memory or even turn off memory altogether. New temporary chats can also be initiated that won’t retain memory or history.

  • Why It Matters: Integrating memory into ChatGPT enhances personalization and interaction quality, highlighting the potential for more meaningful AI engagements. It's a step forward in making AI interactions more nuanced and user-centric. I am curious to see how aggressively it will remember things and how often you may have to clean out irrelevant information.

Image Credits: OpenAI

Resources For Prompt Engineering

Prompt Engineering is emerging as a critical skill in optimizing interactions with Large Language Models (LLMs). Below are resources to help you refine your prompt crafting abilities:

Your curiosity fuels this exploration of the ever-evolving AI landscape. Share these insights with fellow AI enthusiasts 🔄. Got any thoughts? Questions? Comment or hit reply. See you next Sunday!

Warm regards,

- Kharee