AI hasn't killed anyone...yet.

The AI safety discussion is largely overblown, and as colleague of mine once called it, "safety theater." All of the supposed AI safety "experts" have popped out of the woodwork and are selling snake oil to companies. Will AI end up killing someone? 100%. When you do something millions or billions of times there will be errors. You'll hear about the outlier but not the millions of times everything went well. Yet we need to remember that you're more likely to get struck by lightning or killed by a shark than get harmed by AI.

When you go down the rabbit hole of AI safety, eventually all paths lead to who wants the power and control over AI. In the AI safetyist's mind, only a select few should have control and everyone else should need permission. Too bad though, as LLMs are already out of the toothpaste tube and they aren't going back in. AI wins and is a powerful and positive force for humanity when anyone can use it. That is where the true innovation happens. Don’t let the self-appointed gatekeepers tell you otherwise.

Ignite CEO lays off 80% of staff — Ignite is betting on AI automation to handle core operations. Digging in though, this was more of a culture change that was needed, and of course the article is much more nuanced. Read more →

Write your own scary legal letter with heavyweight — A new tool lets you generate cease-and-desist letters that sound like they came from a top law firm. Try it out →

Create your own pop song with AI — AI music platforms are making it possible to write, record, and publish songs in minutes. Try it here →

The biggest challenge to using ChatGPT or other LLMs is literally just starting. People are afraid they will mess something up. Newsflash: make as many mistakes as you need to learn. This prompt below uses ChatGPT to help you become more proficient in prompting:

"Create a beginner's guide to using ChatGPT. Important topics should include prompts, priming, structure, and personas. Include tangible examples. The guide should be no longer than 500 words."

Applying AI to moonshot ideas is fun, but what about the boring things? Possibly an empathetic letter that can be sent to customers?

Allstate is using generative AI to draft replies to their customers, and are finding that AI is more empathetic than a frustrated claims agent. Most of their communications are now being created by AI and they are being verified by actual people before being sent out. They are also integrating into their customer service process, making it more seamless and ensuring less back and forth between their claims agents and customers. It's a practical case of using AI to enhance and improve workflow, while also ensuring there is still a human-in-the-loop.

AI, and now generative AI, feel like they are in a perpetual hype cycle. We overestimate what we can happen in a year, and underestimate what can happen in 5 years. AI is following this same trajectory. The human mind has difficulty visualizing and understanding exponential growth curves. Because AI is in the digital world, there aren't many barriers to scaling, and the only physical barriers are energy and data centers.

Take the GPT models for instance. The parameters went up exponentially through the different GPT versions (think of parameters like the adjustable knobs on a giant control board, where more knobs are better for the model). GPT-1 to GPT-2 saw a 13x increase in parameters, GPT-2 to GPT-3 saw a 116x increase, etc. This is the real power of AI. You can't scale any physical business (retail, manufacturing, etc) the way you can scale things digitally. AI is doing this on steroids, and it is essential for you to get on the bus if you haven't already.

-Ylan