To paraphrase a famous quote, "AGI is already here, it just isn't evenly distributed."

The accelerating advances by multiple companies are showing GenAI capabilities that individually are each amazing. When they are combined, they are beyond what most people still consider possible outside of science fiction. We need to prepare our businesses for leading GenAI providers combining all of this to enable the next wave of Generative AI very soon.

?? We already know tools like ChatGPT can analyze and generate nearly any style of text. I previously tested it to consume all of Shakespeare's writings which is 3706 pages of text, and it did it in 37 seconds. No person can do that. And GenAI was able to answer nearly any question about the content across all of those works. Imagine this being used to read and analyze every contract, invoice, RFP, and policy in your business.

?? ChatGPT and Claude have also expanded in capability to handle complex math as well. A year ago, GenAI could not do even basic math. Today, the leading models can solve nearly any math problem, including calculus, that you can toss at them. They can even write and run code to solve the problem and create a visualization of the solution. This is moving GenAI into finance and operations.

?? Image and video generation have progressed so rapidly in the last 12 months that it has gone from a novelty to producing commercials and short films. Tools like Sora, Midjourney, Dall-e, Runway, Luma Labs and many others are enabling not just text-to-image and text-to-video, but some also allow control of the content for consistent style and characters, in-painting to regenerate parts of the content, and keyframes to control the start and end points of video. These are transforming marketing and all creative content.

?? Sound and voice generation have also made huge leaps. In the last year, not only has voice generation allowed cloning of nearly any voice, but also the generation of music, singing, and sound effects. A voice AI generator recently added sound effect generation from text. Suno creates entire songs, in any style, from a simple prompt. They are often amazing. Tools like D-ID, Heygen, and Hedra also have the ability to synchronize text and voice to images and videos, creating very convincing avatars of people. These can be used to teach, to share ideas, and to provide customer service.

?? Code generation is a well known Generative AI capability. Tools like Github Copilot and CodeWhisperer are all capable of software development assistance and code generation. Recently, Claude 3.5 introduced Artifacts which allows you to describe an application and it will generate, run and let you interact with the application. You can change the requirements at any time, and it just regens the application with the new requirements. This is much faster than agile devops. You can expect text-to-application to become a significant capability in all platforms and to transform IT and product development.

?? Speed and latency have been a challenge in GenAI. Typically, to get the best answers, you need the most advanced models. But, it can take 30-60 seconds to get output, even longer for images. Less capable models are often much faster but they are more limited. Recently companies are demonstrating near real time generation. Voice solutions like GPT-4o, and Moshi are showing voice response in less than 250ms which is faster than people respond. And, new hosting platforms using an LPU (Learning Processor Units) technology rather than GPU's are optimized for incredible inference speed. These allow the latest models to respond faster than anyone could even read. Once GenAI is nearly instantaneous, then the 'cost of rework' goes to zero and you can iterate on solutions, images, text freely to get the best possible results. When you combine this speed with the use of new agent frameworks, entire workflows can operate at speed that people can not match on their own.

NET: When the leading models combine all of these advancements, they enable entirely new ways of working and new business models for companies. These capabilities are so powerful that we also need new ways to ensure trust and compliance with GenAI and to protect intellectual property. And, we have to build up our own skills in using these new tools to change the way our companies work.

The best people working with the best GenAI will have substantial competitive advantage.

#ai #genai #artificialintelligence #generativeAI