Confluence for 11.26.23
Amazon announces AI skills training for all. Anthropic releases Claude 2.1. Meta introduces Emu Video and Emu Edit. A video explainer of large language models.
We hope our U.S.-based readers have enjoyed spending time with friends and family this past week. With all the noise surrounding OpenAI, ultimately resulting in Sam Altman’s reinstatement as CEO, we want to draw attention to some other developments that may have been lost in the shuffle. With that in mind, here’s what has our attention at the intersection of AI and corporate communication:
Amazon Announces AI Skills Training for All
Anthropic Releases Claude 2.1
Meta Introduces Emu Video and Emu Edit
A New Video Explainer of Large Language Models
Amazon Announces AI Skills Training for All
How Amazon aims to help meet the demand for AI skills.
Amazon recently announced “AI Ready,” an ambitious initiative dedicated to equipping two million people with AI-related skills by the end of 2025. This is a significant commitment and a signal that larger players are taking the human and labor aspects of the equation seriously. No matter how powerful the latest models and applications become, there will be a pressing need for skilled individuals capable of harnessing the potential of Generative AI effectively.
This announcement brings to mind a piece published in The Atlantic earlier this year, in which Charlie Warzel suggested that understanding how to work with AI will be the most important skill we can teach people. Nearly 10 months later, Warzel’s argument certainly seems prescient, and we expect more companies, institutions, and governments to announce initiatives addressing the human side of the equation over time.
Anthropic Releases Claude 2.1
A reminder that OpenAI and Microsoft aren’t the only games in town.
Anthropic, one of OpenAI’s leading competitors and recipient of a $4 billion investment from Amazon, announced an update to its leading large language model (LLM), Claude 2.1. We’re still experimenting with this update, but thus far have found a few key updates worth noting:
Expanded context window: Simply put, Claude can now work with much larger pieces of content. Anthropic claims that Claude 2.1 can summarize, analyze, and answer questions about pieces of text as large as the Iliad and the Odyssey.
Enhanced developer tools: Anthropic is focusing on simplifying the developer experience and making it easier to build applications with Claude.
System prompts: Claude now allows for the introduction of custom instructions for individual chat sessions. While not as intuitive as ChatGPT’s custom instructions or GPTs, it’s a notable acknowledgement of the value of providing context for chatbot interactions.
For all the attention on ChatGPT, other companies continue to push forward and improve their own models and applications. Claude remains one to watch, and we’ll continue to highlight updates from other competitors as they roll out.
Meta Introduces Emu Video and Emu Edit
There’s a new foundational model for video generation and image editing.
Meta recently published a blog post that details the latest developments for Emu, their foundational model for image generation. The post itself gets fairly technical, describing the functionality of Emu Video and Emu edit in detail, though one particular quote stood out to us for its implications:
Emu Edit is capable of free-form editing through instructions, encompassing tasks such as local and global editing, removing and adding a background, color and geometry transformations, detection and segmentation, and more.
This quote and Emu’s focus on fidelity to user instruction portends a future where LLMs overlay software capable of doing much more. Instead of users needing to navigate complex editing software for tasks like enhancing contrasts or adjusting color balance, they can simply instruct the software on the desired outcome. This same functionality could extend to other types of software beyond video and image editing. With LLM-powered interfaces like these, we’re moving toward a future where we tell software what to do in natural language rather than navigating complex interfaces.
A Video Explainer of Large Language Models
OpenAI’s Andrej Karpathy shares a “busy person’s introduction to LLMs.”
The better you understand generative AI technology, the better you will understand where it excels and where it struggles — and thus have a better grasp on what to use it for, where the risks are, and so on. When we come across items that we find particularly valuable for those purposes, we’ll continue to share them here. One new item worth viewing, in our opinion, is Andrej Karpathy’s Introduction to Large Language Models, a one-hour talk available on YouTube. It’s longer and slightly more technical than previous pieces we’ve referenced here1, but we believe it’s worth the hour. Karpathy demystifies the key concepts and nuances underpinning LLMs with illustrations and examples that non-technologists will understand.
Karpathy provides foundational information on how developers train the current generation of models and forecasts what we might expect for their applications (e.g., increasing use of tools and multi-modality, both of which exist in ChatGPT today) and continued development (i.e., where the next leaps in capability might come from). Particularly interesting is Karpathy’s attempt to “tie it all together” and lay the conceptual framework for how what an LLM operating system might look like — which, suffice to say, is much more than the chatbots and language generators with which most of us are familiar today.
Now we’ll leave you with something cool: fal has an image generator that creates images as you type. It’s pretty fun to see the graphic take shape as your prompt develops (and change as your prompt changes).
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.Some explainers and primers we’ve referenced to date:
Mudhamita Murgia, “Generative AI Exists Because of the Transformer” in the Financial Times
Cal Newport, “What Kind of Mind Does ChatGPT Have?” in The New Yorker
Steven Rosenbush, Isabelle Bousquette, and Belle Lin’s overview of AI terminology in The Wall Street Journal
Thank you for sharing insights on Big Tech's focus on AI.