2024 AI Predictions - Part 2

By Steve Pfrenzinger

My last post was the Top Ten for “major” predictions for 2024.? Here’s a lightning round for “Part 2” (basically an “AI Rant”) for my Top Ten on lessor, but still important, predictions and a must read for those following AI, and wanting to be a step ahead of other non-AI-experts. ??

I promise some surprises.? If not, let me know.??

Here goes, hang on to your hat:

1. Legal issues will plague AI vendors for copywrite and IP infringement.? NY Times lawsuit against OpenAI is an example.? Closed models are more at risk than open source model (which can’t really effectively stop developers from such infringement).? No one really knows how this will turn out.? Could cost OpenAI and others many many millions.??
2. Personalized chatbots will be the big thing for specialists, coaches, advisors, doctors, lawyers, healthcare, and more.? Each trained to be as if you’re interacting with the real person or their spokesperson, if it's a company chatbot.? Expert advice will be available 24/7.? Monetization of such chatbots will begin to happen as well. ?Note - Creating “legacy chatbots” will become a big thing as older people want to consolidate their personal and professional histories for their families and future heirs.??See #4 for chatbot dev tools.
3. If AGI is coming (Artificial General Intelligence) and AI can fully simulate human behavior, the question will be what personality type (style or trait) will it show in its behavior (e.g., Introvert vs Extravert, Thinker vs Feeler, …)?? Well, which one do you want, as it will likely become an option, based on the purpose of the AI your using or building?? Personality type is basically “bias”, a tendency to act a certain way, and is well known to identify differences in behavior in humans and AI will be no different.? Be it Big 5, Myers Briggs, DiSC, Hogan, we all have preferences, and AI will be no different.??
4. No-code (programmer-less) development is real and major progress will be made this year.? Programmers will have lots of old code to maintain, but the new code bases won’t need them (or far fewer of them).? No-code development comes in two basic varieties, fully NLP (natural language driven – like GPT’s from OpenAI) and forms-based entry of specs (aka fill in the blanks).??As for chatbots mentioned above, I've built 10 of them and I haven't programmed in decades. I use this forms-based tool https://gpt-trainer.com/. Amazing. Two more https://chatwith.tools/ and https://buildchatbot.ai/
5. Alternatives to mega expensive LLMs with hundreds of billions or even trillions of parameters, via cost-efficient prompt engineering (RAG and Mixtral rock!).? ****** RAG will grow! ?Domain specific LLMs are recommended for those that want to provide guardrails to the larger foundational LLMs (like ChatGPT, Google Bard, BingChat, …), but are way too expensive to train for most.? Using a “vector database” and RAG (Retrieval-Augmented Generation) is an inexpensive way to create a domain-specific-like model and to focus?larger LLMs on what you want as output.? RAG is a form of “prompt engineering” and allows the prompt to be examined in real time for what it needs as data, the vector DB then finds that data in sources you provide (typical training data like files, URLs, websites, and such) and then includes that subset of data in the prompt, again in real time, to provide just the searchable data needed (hope that made sense).? Crazy fast and very cheap.? This is a big deal and core to the personalized chatbots I mention above!? Just remember RAG is a cost-effective way to simulate an expensive domain specific LLM.? ****** Mixtral will emerge.? If you are prompting on healthcare topics, do you really need a mega, trillion-parameter LLM to answer your questions?? No you don’t.? With this mix-and-match concept, one or more smaller, domain specific LLMs (with as few as say 2-10B parameters) can be selected real-time to dynamically utilize the most relevant sub-models (domain expertise), based on your prompt’s context. ?Makes total sense to me and could have huge impact on the AI market.? This could be the entire future of AI architectures, using just the data you need.??
6. Augmented reality, after a slow start, will expand rapidly with Apple and Meta leading the way.? So, if you like wandering around in virtual worlds, rather than the real world, your opportunities will abound. ?And Apple’s gear could cost 5-10X more than Meta’s.??
7. Video to text for summarization services will make listening to long podcasts/presentations old school.? Who would listen for an hour of video/audio (typically nonfiction) when they can get the essence (a detailed summary) instantly and absorb the key points in just a few minutes?? This already exists and will grow in popularity for long videos, text-based articles and papers as well.?Checkout Chrome extension “YouTube Summarizer” by https://syllaby.io/
8. AI is growing exponentially and vendors are leapfrogging competitors almost daily.? The result will be that for every new entrant, there will likely be a loser.? If you decide to “buy rather than build/modify” and use a start up’s product, you have to ask yourself, “what if they fail?”? Eventually, nearly all existing applications will be retired and replaced by AI-based tools (in future years).??
9. Cold chain (a subset of the supply chain) will be revolutionized with AI-based IoT connectivity of ambient sensors for both “item and case-level” tracking using cheap, postage stamp sized, stickers that are “battery free” sensors (via energy-harvesting) for tracking products from “field to fork” or “manufacturer to customer”.? New FDA temperature-related, traceability requirements on food, medication, chemicals and more will drive this end-to-end tracking, monitoring and transparency trend.? And AI will help make it happen.??Checkout the leaders in this space https://www.wiliot.com/ and https://montage-connect.com/
10. Lastly and sadly for content providers, AI will help create a massive amount of new online content (100+X) that will begin to run out of audiences.? Expect fewer “likes, views and comments” for your posts.? LinkedIn and other content providers will be negatively impacted.??

The 20 predictions in my two 2024 prediction posts are obviously not all encompassing, but just a list I created as I did my near-daily AI research and with the insights of others I follow on this topic.

See #pfrenzingerAI for my other AI posts.

Comments welcome.

Here’s a link to my previous post of the Top Ten AI predictions for 2024, related to the biggest AI players and issues.? Again this could be considered 11-20 in that list! https://www.dhirubhai.net/posts/steve-pfrenzinger_pfrenzingerai-agents-2024predictions-activity-7150146644565446657-nO3G?utm_source=share&utm_medium=member_desktop