The Fork in the Road

Photo by Luke Jones on Unsplash

The whole Mythos and Fable situation got me thinking, and not necessarily about Anthropic.

What caught my attention was how quickly access disappeared. One day, people were talking about Mythos/Fable as the next major step forward in AI. The next day, Fable was gone. Poof. And Mythos was determined to be too powerful to release. Whether the government’s concerns were justified is a debate I’ll leave to cybersecurity experts. What interested me was something much simpler. It reminded me how little control most of us actually have over the tools we use.

I’ve been around technology for a long time. I’ve seen software platforms come and go. I’ve seen companies abandon products that people depended on. I’ve watched services shut down, licensing terms change, and entire ecosystems disappear. Usually, there is plenty of warning. Sometimes there isn’t. The Mythos/Fable story felt like another reminder that access to technology should never be taken for granted.

At the same time, I’ve spent the last year moving in the opposite direction.

While working with local AI hardware, I found myself running experiments on just about everything I own. My main desktop uses a Ryzen processor and an Intel Arc GPU. I have a Mac Mini M4. I have an inexpensive Chromebook that I converted to run Linux. I even picked up an 8GB MacBook Neo because I wanted to see what could actually be done with modest hardware rather than what people claimed was possible.

The results challenged many assumptions I had going into the project.

If you spend enough time watching AI videos or reading social media, you could easily come away believing that anything short of the latest frontier model is barely worth using. The discussion is almost always about who has the biggest model, the highest benchmark scores, or the most advanced reasoning system. There is certainly value in those things, but they are only part of the story.

What I discovered is that a surprising amount of writing assistance does not require the smartest AI in the world.

I write books, articles, technical content, and research notes. When I’m working on a manuscript, I need help organizing information, testing arguments, spotting weak sections, and finding better ways to structure ideas. Those tasks matter far more to my workflow than having access to the absolute best benchmark score on the planet.

Over time, I found myself using smaller local models more often than I expected. A 4B or 8B model isn’t going to replace the strongest cloud systems, and anyone claiming otherwise is fooling themselves. The larger frontier models remain better at many tasks. They reason more deeply, handle complexity more gracefully, and generally produce stronger results.

What surprised me was how often that difference didn’t matter.

When I’m outlining a chapter, brainstorming article ideas, reviewing research notes, or looking for weaknesses in an argument, a smaller model often gets me most of the way there. It may not produce the perfect answer, but it frequently produces a useful answer. For many writing tasks, useful is enough.

That realization slowly changed my thinking about local AI.

A year ago, I viewed it primarily as a compromise. If you couldn’t afford the best cloud services or didn’t have reliable internet access, local models gave you another option. Today I see something else. I see resilience.

When a model is on my machine, I can use it whenever I want. I don’t have to worry about a subscription tier changing, a company discontinuing a service, or a feature disappearing because of a policy decision. That doesn’t make local AI superior. It simply means I control access to it.

The Mythos and Fable story reinforced that point.

For years, the AI conversation has focused almost entirely on capability. The question has always been, “What can these systems do?” After watching the events surrounding Mythos and Fable unfold, I started asking a different question.

Who gets to use them?

That may end up being just as important.

The industry will continue building larger and more capable systems. Governments will continue worrying about security implications. Companies will continue balancing innovation against regulation. None of that is likely to change.

Meanwhile, the open-source community keeps pushing in a different direction. Models become more efficient. Quantization improves. Hardware that seemed inadequate a year ago suddenly becomes useful. Every few months, another limitation disappears, and something that once required expensive infrastructure becomes practical on consumer hardware.

That trend has been just as interesting to me as the race toward bigger models.

Maybe the Mythos and Fable episode turns out to be a temporary event that people forget about six months from now. Maybe access returns and the controversy fades away. What I suspect I will remember is the question it raised.

The best AI in the world doesn’t help me much if I can’t use it.

As a writer, I care far more about tools that help me get work done than I do about tools that exist behind closed doors. That’s why I’ve become increasingly interested in local AI, smaller models, and open systems. They may not represent the cutting edge, but they are available, and availability is paramount.

As I sit here today, it seems to me that the AI industry is approaching a fork in the road. One direction leads toward increasingly powerful systems that few people will ever touch. The other leads toward tools that ordinary users can run, experiment with, and build their own workflows around.

After spending the last year working with both approaches, I find myself paying much closer attention to the second path.