There is an insane number of AI projects being developed and ".ai" businesses being spun up. AI has been present in the enterprise sector for years in various forms, but now, with large language models (LLMs), there is a highly accessible and functional model for people to interact with AI.
Many of these models are open-sourced, and with their widespread availability, it's unlikely that organizations like OpenAI will slow down their efforts. Competing tools will quickly emerge and offer new functionality. Projects like LangChain are already working in areas that OpenAI has been cautious about due to safety and governance concerns.
This situation resembles the early days of crypto coin wars; where developers could clone a coin, make small enhancements, and target a new niche. This trend is common in digital frenzies and is unlikely to stop with AI. In recent weeks, the cost of training new models has decreased dramatically, disproving predictions that it would take years for prices to drop. Quotes from Cathie Woods saying it would take 7 years to drop the price from $4m to $400k to train these models have proven wrong as clever folks at Stanford have done it in 5 weeks for $600 for Alpaca AI.
This rapid development has significant implications for various industries, including white-collar jobs. Experts believe that the economics of digital services compared to those requiring physical labor will change as AI automates tasks in fields like copywriting, branding, and teaching. Even mechanical engineering jobs may be impacted in the future. This mind blowing on many levels!
Looking ahead, the market will likely expand by using edge computing to bring AI to the edge, similar to how Google distributed search. This development could benefit companies like Cloudflare and Fastly. Training AI models currently requires tens of thousands of high-end processors, such as those from Nvidia, and this demand will likely continue unless more affordable solutions using technique's like self-instruct as shown with Alpaca become available. It may further give rise for quantum technologies like IonQ to play a part in AI compute.
An intriguing aspect of AI's future is the economy created by selling access to data within and among these tools. Similar to the economy speculated with 5G and autonomous vehicles, AI platforms and peripheral systems may want to buy and sell their data. AI systems are already capable of automatically connecting to data sources through APIs, manifests, and JSON.
I believe its only a matter of time before an AI-only marketplace emerges, where platforms like Fetch.ai or Ocean Protocol facilitate data access and problem-solving. AI systems could pay for access using these services to share data and address issues that we may not even be aware of yet.
In the long term, AI's impact will likely extend beyond language models like GPT. GPT is likely the initial touch point. AI’s ability to interact in a metaverse or such is also highly probable.
This development could have enormous societal consequences, with humans potentially serving as support for the AI ecosystem. However, this shift could also present opportunities to improve communities and foster positive change.
Is there a trade in all of this? Perhaps, but this is more of a long term vision and would be more of an investment. Many of these trends quickly lose focus and melt into the economy as a whole. Arguable, enterprise software such as MSFT can leverage this the best. The crypto plays mentioned have seen 75% moves with ~$500m market caps in anticipation. NVIDIA would be another however with every new chip with massive gains does that require fewer sales? This hasn't been the case, but the economics may not be clear cut.
That is my perspective on AI.
Brent @ Mometic