Last call for AI Lunch Club

Want to go deeper into the subjects we discuss in our workshops? We’re doing just that in our new AI Lunch Club virtual events. The final of our first batch of four events will take place over Zoom from 12:30-1:30 p.m. CST on Wednesday:

• Next Steps in Prompting: Effectively prompting a large language model is an important skill for getting the best outputs from tools like ChatGPT and Google Gemini. In this session, you’ll learn about prompting concepts and techniques beyond the basics that you can use across a range of tasks.

Sign up for events here. Use the code SAVE10 to save $10 off the price for each event. Or bring 3+ people and save more.

Generative AI News

AI will reduce workforce, say 41% of execs in a survey

Yet another executive survey on the adoption of AI was released this week. The survey, given to executives at 2,000 companies worldwide by recruitment agency Adecco Group, found:

• 41% of executives expect to have a smaller workforce in five years (the number is 36% in the US).

• 57% think that human skills will continue to be more valuable in the workplace than AI skills.

• 78% hold that generative AI will play a "critical role in providing upskilling and development opportunities."

You can read the full results here.

AI business is booming: ChatGPT Enterprise now boasts 600,000+ users

In January 2024, ChatGPT Enterprise had 150,000 subscribers. Since then, the number has quadrupled, with more than 600,000 subscribers. Readers may recall that ChatGPT Enterprise is the highest tier of the various ChatGPT account levels, offering “improved security, longer context windows, unlimited high-speed GPT-4 access, advanced analysis capabilities, and more.”

OpenAI makes it easier for developers to fine-tune and build AI models

Last Thursday, OpenAI announced updates to its Custom Model Program and its API for fine-tuning their language models (a model is fine-tuned when it is modified to incorporate new data beyond the data it was originally trained on, thereby producing a customized version of the model). The updates include assisted fine-tuning, which allows for more advanced fine-tuning options for organizations to implement, and custom-trained models, which will allow organizations to "train a purpose-built model from scratch." New options for fine-tuning GPT-3.5 are also newly available. Details can be found here.

OpenAI transcribed over a million hours of YouTube videos to train GPT-4

This week the New York times dropped three stories on the race to accumulate new training data for AI models:

• How Tech Giants Cut Corners to Harvest Data for A.I.

• What to Know About Tech Companies Using A.I. to Teach Their Own A.I.

• Four Takeaways on the Race to Amass Data for A.I.

One of the ways this desired data was harvested was through transcribing YouTube videos, which OpenAI pulled off using their audio transcription tool Whisper, as highlighted in this piece. Reportedly, the transcriptions were subsequently used to train GPT-4 under the guise of fair use. In response, a Google spokesperson stated that this would constitute a violation of YouTube’s terms of use. YouTube’s CEO Neal Mohan made a similar claim about the potential use of YouTube videos to train OpenAI’s text-to-video tool Sora.

Quick Hits

Tool of the week: DALL-E’s new editing tool

Open AI announced last week that paid users can now edit parts of their AI images using text prompts. Users can highlight the part of images they want to change, and write a text prompt of what they want to see in that space. Here’s a demo.

AI-generated image of the week

Does adding the words “award-winning” to the beginning of your image prompt make for a better image? You be the judge. We made the following two prompts (see the tip of the week below for more on this). Do you think adding award-winning helped?

Personally, I think it added a little more detail.

Prompt: award-winning photo of a blue dog wearing orange pajamas and red glasses

prompt: photo of a blue dog wearing orange pajamas and red glasses

Generative AI tip of the week: Permutation Prompts

Midjourney Permutation Prompts let you generate variations of a prompt with a single /imagine command. Simply include a list of options separated with commas within curly braces {} in your prompt.

For example, the prompt, “award-winning photo of a blue dog wearing {pink, blue, orange} pajamas and red glasses,” will create three different prompts, one with each color of pajamas.

Get starting with Generative AI

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

• NightCafe image prompting tips

• Learn ChatGPT prompting

• Best AI image generators

• AI terms everyone should know

• Prompt Engineering: What to Know and Why It's Important (NEW LINK!)

What we found

Last week Stable Diffusion released Stable Audio 2.0, an updated version of its text-to-music generative AI tool. Users can create 3-minute-long audio that can be used commercially. Try it here.