AI Chatbot Mania & AI Agents: Part 1

A Bursting Bubble - I Don't Think So

NOTE: There are personal opinions expressed in this newsletter about AI chatbots, AI agents, generative AI in general.

An article hit CoinTelegraph last week detailing that during the first half of 2023, over $25 billion was invested in AI startups. It was even titled "The absurd AI mania is coming to an end", and it went on to note that signs of vulnerability are emerging within the AI space. Let me paint a picture of how I perceive the "AI space" right now. Generative models and diffusion models took over the scene in early 2023. There are really only three camps when it comes to large language models (LLMs) which are the crux of this recent AI surge. They are:

1. Companies that build and train LLMs (machine learning focus) (OpenAI, Google, Anthropic, Meta, X, smaller entities that might be building open-source models, etc.) These are models like GPT-4, PALM 2, LLaMA 2, Falcon, BLOOM, Stable-Diffusion, etc.
2. Companies that supply hardware, GPUs, and processors for the development of AI models. These are companies like NVIDIA, Intel, Microsoft, IBM, etc.
3. Companies & individuals that fine-tune pre-trained LLMs and/or those that utilize LLMs in their products via API (everyone else).

This sets the stage as most of the $25 billion invested was sent in the direction of companies falling into camps number one and number two. Plus, I always said this at the beginning, I never believed that ChatGPT was meant to be their main product. I still don't. It certainly grew legs. More so, I believe it was meant to be a showcase of what is possible with their models, and then everyone started using it and it became a revenue stream. However, one truth in the article is that usage has fallen and the market has become saturated with ChatGPT lookalikes.

As for hardware shortages due to model training, just think, Elon Musk bought 10,000 GPU units to train X. The Technology Innovation Institute bought 384 GPUs to train Falcon but dedicated 2800 compute days to its pre-training. I would like to think that X would take anywhere from 30 to 80 days to have trained their V1 model. While this is just a comparative example, it doesn't dismiss the ongoing supply and demand challenges in the realm of silicon.

You get a chatbot, you get a chatbot, and you get a chatbot (said in the voice of Oprah)

Maybe there was and still is some mania and maybe it is absurd how it came about. If you noticed, I titled this "AI Chatbot Mania & AI Agents". I did this because companies offering their application programing interfaces (APIs) to their LLMs almost made it so that anyone who had ever touched a code editor before was able to create a chatbot or create a company that produces chatbots as a product.

Many new startups seem to offer a generative AI tool and 9 times out of 10 it is in the form of a basic chatbot. These chatbots are usually rate limited to a certain number of requests or words/tokens. I believe that the web will always trend toward more of a conversational interface than a point and click interface. However, chat is really just an interface in which the user interacts with some sort of "application". Q&A and customer support are a great fit for a chat interface but there has to be more behind these applications than just the chat feature. Otherwise, nothing more than a conversation between a person and an AI actor happens.

What I will say is that I think the "chatbot mania" is cooling off and if that was the bubble, it will likely get consumed by a much larger bubble that is the next wave of generative AI. What the chat mania did, is it opened doors to a different perspective of how we do use and could use the internet as a whole. It opened doors to more complex applications in that it productized and made LLMs accessible to the masses through APIs. That created an onslaught of chat products, yes. Just like the early web days and early crypto days though, most of those entities will disappear just as quickly as they emerged. This is a natural cycle for innovation and adaptation.

AI Agents and Part 2

Part 2 will be released as a free article on Substack and be the first edition of Future Frontier: Blue later this week. I will detail what AI agents are, what they mean to businesses and individuals, and even a few use cases. If you have not already checkout my premium publication called Future Frontier: Black. This publication is aimed at developers who are looking to get started building applications with generative AI (chat features included ??).

------- ?? -------

Hey, I'm Dave! I'm a former digital agency owner, now co-founder of Bold Crow. I'm all about helping businesses implement customized AI solutions in a responsible and ethical way. I've built cool tools for nonprofits too, helping them gather and leverage social proof. Let's connect:?Dave Norris

#futurefrontier #ainews #technews