Innovation Profs - 12/5/2023

Your weekly guide to generative AI tools and news

Dec. 15 Generative AI Event

Join the Innovation Profs at noon on Dec. 15 for HOLID-AI: A holiday-themed introduction to AI. Professors Porter and Snider will share practical uses of generative AI tools such as ChatGPT, Bard, Midjourney, D-ID and more - all with a holiday twist. The event is free and open to the public. Sign up here.

Generative AI News

Looking back at a transformative year for AI

We now have more details on what the heck happened at OpenAI

Apparently, the fiasco with Sam Altman and the OpenAI board wasn’t due to worries about the too-rapid development of artificial general intelligence. Instead, it looks like Altman was trying to have Helen Toner, AI safety expert, removed from the board and was misrepresenting conversations with various board members about support for ousting Toner, at least according to a story published by the New Yorker late last week.

Amazon unveils Titan image generating AI: what you need to know

At a conference last week, Amazon announced a number of generative AI tools, including Titan, an enterprise-level image generation tool, Titan Text Lite, a smaller language model that can be used for “lighter” text generation tasks, and Titan Text Express, a larger language model that can be used to power conversational AI applications. Even more generative AI tools!

Stable Diffusion XL Turbo can generate AI images as fast as you can type

Last week Stability AI released Stable Diffusion XL Turbo, which they are billing as “real-time” image generation. “SDXL Turbo's primary innovation lies in its ability to produce image outputs in a single step, a significant reduction from the 20–50 steps required by its predecessor.“ There is a trade-off, though, as the resulting images are less detailed than those produced with a higher number of steps.

Perplexity AI unveils ‘online’ LLMs that could dethrone Google Search

Perplexity AI has released its own LLMs after previously using existing language models: pplx-7b-online and pplx-70b-online (where the names express the number of parameters in the respective models). These models were produced by fine-tuning two open source models, Mistral’s mistral-7b and Meta’s llama2-70b. The key features of these models is that they are web-connected with no knowledge cut-off, capable of providing “helpful, factual, and up-to-date information.”

GPT Store Set To Launch In 2024 After ‘Unexpected’ Delays

OpenAI sent a letter out to GPT builders that the GPT Store, originally expected to launch this month, would be delayed until early 2024. Enhancements to the GPT Builder tools were also announced in the letter.

Google has quietly pushed back the launch of next-gen AI model Gemini until next year, report says

Yet another delay: the launch of Google’s LLM Gemini, billed as rivaling GPT-4 in performance, will be delayed until 2024. Apparently issues in how it handles non-English prompts need to be addressed.

Quick Hits

Tool of the week: Pika Labs

AI video is growing fast. Pika Labs rolled out Pika 1.0 last week. The idea-to-video platform works like image creation tools, but allows you to create and edit videos. Join the waitlist (or join the Discord to get access right away) and check out what the tool can do.

Innovation Profs Homework

Get up to speed on what may be the next big thing in AI… This week, ByteDance revealed MagicAnimate. The tool uses a single image and a reference motion video to produce an animated video. Read about it and see examples here.

There is a working model here, but the servers are a bit overloaded currently.

This comes at the same time as the release of similar tool Animate Anyone by Alibaba Group’s Institute for Intelligent Computing.

AI-generated image of the week

We tested out Pika Labs idea-to-video platform (mentioned above) to create a three-second video of a bulldog on Drake University’s campus. I think @drakeugriff’s job is safe for now.

Prompt: a bulldog in a blue shirt runs down the sidewalk 

Generative AI tip of the week

Learn how to create data visualizations with ChatGPT as your assistant.

Get starting with Generative AI

New to generative AI? Here are some places to start…

What we found:

Runway, an image-to-video tool, was fed a bunch a memes and came up with…well, you judge for yourself.

This is much more of an indictment of the free version of Runway than the capability of AI to animate images, given the much better results produced by Stable Diffusion’s image-to-video tool linked below.