AI Top-of-Mind for Dec 1
Top-of-mind is a good one-year AI review by CNN covering where we’ve been and possibilities for the future.
Maybe one last note on the Q* saga? Additional perspective in ‘The Generator’ on what OpenAI may or may not have created, with some observations:
Like the human brain, that would give the new model the ability to reason in an intuitive and creative way (using its “jumble of wires”, in the same way, that today’s LLMs do), but also to switch into a symbolic reasoning mode and solve problems that have a single correct answer, like math problems.
A model that is able to reason both intuitively and symbolically would be hugely important in fields like natural language processing, drug discovery, and mathematics. These fields require a blend of creativity and logic that LLMs currently struggle to deliver.
We’ve all heard of the cost of LLM training and queries. ‘The Information’ on how customers find creative ways to reduce their spend. From the article:
Here’s how it works. Startups are using high-end and expensive LLMs such as GPT-4, but only for a narrow purpose, such as generating marketing copy for a particular product. They take the results and use them to customize — or fine-tune, in industry parlance — a free, open-source model like Meta Platforms’ Llama 2 to handle those same tasks going forward.
To counterbalance costs, a four-step guide in the ‘Pareto Investor’ on how to get started creating custom GPTs, and the potential financial rewards. From the article describing the opportunity:
App Store boasting 1.8 million apps with a staggering $910 billion revenue.
On the creative front, a good perspective on ‘Leonardo.ai’ as the most sophisticated graphics tool. A snippet from the article below, and it also speaks to the simplicity of the API.
At MidJourney? There is no such thing. You are limited to prefabricated models and have little scope for individual adjustments. In short, while Leonardo.ai gives you the tools to create your own AI art gallery, MidJourney just lets you stroll through the exhibition.
Also on creative, a simple way to create simple gif animations within ChatGPT through the use of sprite-sheets and feedback.
Finally, on the marketing front, an overview in ‘AdAge’ of the best AI-assisted campaigns in Nov, and in the same publication, lessons learned using Google’s AI-powered Performance Max.
