blogs

I finally got around to writing down my key takeaways from #MAICON23. It was a really good conference, and the LLM advancements were remarkable (compared to last year). I learned a lot from the sessions and networking with colleagues who are doing some interesting things in AI, ML and predictive analytics. I am looking forward to testing some of the new tools and deploying several new ideas that I gained from the conference. Here are my key takeaways:

1. AI will NOT replace Thinking and Creativity (for most jobs). Humans are needed for creativity, subject matter expertise, decision-making, and judgement.
2. AI tools have passed an inflection point, are now affordable, and get better EVERYDAY. Today’s AI tools are the least capable that you will ever see or use. Ready for prime time applications are: content generation (text, video, and code); information extraction; summarization; rewriting; classification; and answering questions.
3. Forward thinking leaders are implementing AI, transforming their companies, and seeing exponential returns and productivity gains – RIGHT NOW.
4. LLM have real bias, IP, data security, and privacy issues. AI developers are NOT protected like social media companies under Section 230.
5. The Enterprise choice: proprietary (Open AI/Chat GPT, Google/BARD) vs. open source (Meta/LLaMa 2). Some enterprises may adopt a hybrid model. New security policies, procedures, and controls will have to be developed.
6. Equality of Opportunity (focuses on individual merits) vs Equality of Outcomes (aims for everyone to reach the same result, regardless of starting conditions). Governments and economic blocs will determine AI usage, regulations and implementations. AI benefits will NOT be evenly distributed.
7. AI deployment at enterprises is at minimum a 2-step process: (1) AI incubation team (or AI council); and (2) AI implementation team. Also, quick “wins” where results can be seen in a relatively short time frame should be the starting point. 
8. LLM will be foundational in business like CRM software.  
9. No ONE knows where AI development is going. It is not like the physics that defined Moore’s law and was predictable with the doubling of compute power every 18 – 24 months.
10. AI – just because we can do something, does that mean we should do something? – see the comment from Jeff Goldblum in Jurassic Park.
11. Native English speakers and developed countries have the early AI advantage.
12. AI makes mathematical decisions not emotional decisions.
13. AI/LLM will require A LOT of compute power.
14. AI will require fewer writers to produce more content that will call for more editors.
15. GPT-4 is a reasoning engine – compare concepts, make counterarguments, or generate analogies, or evaluate the symbolic logic in code. Trained on humanity’s concepts and thoughts.
16. No ONE knows where AI development is going. It is not like the physics that defined Moore’s laws and was predictable with the doubling of compute power every 18 – 24 months. The Open AI creators didn’t even know that GPT-2 could translate languages, or that GPT-4 could be 10x more powerful than GPT-3.
17. LLM are best suited for automating high-volume repetitive tasks. This will result in an explosion of entrepreneurship.