Obvious Business Opportunities
for applying LLMs and their relatives
For the WSJ, Christopher Mims writes,
Underneath the shiny and the new, lurking in IT systems where it creates security vulnerabilities and barriers to innovation, is an accumulation of quick fixes and outdated systems never intended for their current use, all of which are badly in need of updating.
Technical debt manifests in myriad ways, from system failures and slower innovation, to security breaches. It was behind the cancellation of more than 13,000 Southwest Airlines flights in late December 2022, which stranded passengers and bags all over the U.S. during the height of the holiday travel season. It’s also, according to experts, a primary driver of the many software vulnerabilities which led to dozens of hacks in the past 12 months, including exploits of critical systems operated by Google, Apple and Microsoft.
Mims goes on to write,
Many software professionals are also worried that the rise of AI as an aid to software development will only make the problem worse. The reason is that anything that makes it easier for coders—especially less-experienced ones—to write and ship software tends to lead to more technical debt. For coders, getting features out the door is often prioritized, rather than taking the time to optimize that code to be as efficient as possible.
I would make the opposite bet. I recommend starting a business that promises to use the new pattern-matching models to go through old software and make it more robust, secure, and up-to-date. If a corporation can get rid of the pain of fixing old software without devoting headcount to the task, I predict it will pay a decent amount.
Other Business Ideas
Probably the lowest-hanging fruit for large language models is customer support. Those phone menu trees with foreign workers hanging on the end of them drive consumers crazy. You can start a business to help firms use LLMs to improve customer support and lower their labor costs at the same time.
Next, there are a lot of college graduates, MBAs, and even PhDs employed doing back-office paperwork in finance, law, and medicine. There already is software to help with this, but it is often clunky and requires a ridiculous amount of training. It should be possible to build much better apps for handling this work using LLMs. The overall size of the market is huge.
This opportunity would be even larger if you could crack the market in government and education, but in those sectors there is little pressure to reduce headcount or cut costs. In fact, government labor unions exert pressure in the opposite direction.
Robots
As you know, some famous tech bros are excited about humanoid robots. As you also may know, I am inclined to make a different bet, on robots designed with specific goals in mind.
The humanoid approach has a couple of advantages. One is that manufacturers could standardize on a single form factor that might be useful in many contexts. Another advantage is that our houses and factories have been designed for with the human form factor in mind, so that a humanoid robot would seem to be able to fit in.
But I think differently. For example, one use case for robots is in the field of security. If I want to walk to the local subway station and not get mugged again, I would like the area to be patrolled by very small robots that sense when a citizen might be vulnerable. I want robots that can fly, take high-quality pictures, and summon police (I am not yet ready for robots that autonomously make arrests). The muggers would have to worry about being photographed well enough to be identified. They would have to worry about being followed by a small swarm of robots until the police catch up.
Or consider the use case for robots in surgery or other medical procedures, such as starting an IV. What is the advantage of having two big hands at the end of two arms? Instead, you probably want tiny hands, able to function without arms or cords of any kind. The doctor or nurse should be able to summon as many tiny hands as needed.
As I have said before, with large language models it should be much easier to train robots, whether they are special-purpose or humanoid. I think that the business opportunities are in providing special-purpose robots that are trained to meet specific needs that arise frequently.
Please read Handle's comment to the previous post https://arnoldkling.substack.com/p/three-components-of-social-order/comment/51272808. There is more than enough widely available tech already to make American streets almost as safe as Japanese ones - has been for decades. The hurdles on the road towards that objective are political and cultural, not technological.
"As you know, some famous tech bros are excited about humanoid robots. As you also may know, I am inclined to make a different bet, on robots designed with specific goals in mind."
The physical capability "multipurpose killer app" for humans is probably our hands, which really are amazing in the wide variety of things they can do with a high degree precision (though often fragile in proportion to the level of precision). Instead of human body-resembling machines, I anticipate a large number of specialized platforms and approaches used to protect and transport those hands to the things they are intended to handle, and then deploy them for that work. I hope you can forgive me but I simply cannot resist naming these "HandleNoids".
I suspect we are going to get a lot of things that look like the wide variety of arthropods, but with deployable hands if they need to manipulate anything and also with wheels for ground movements (immensely superior to legs when on flat terrain), or propellers for flying or for movement on or in water. If swimming in water then probably more like fish or cetaceans than arthropods. If on uneven ground, "robot spider-centaur" seems intuitively ideal as a basic design, reminiscent of the tachikoma "think-tanks" from Ghost in the Shell.
Nature and engineering are both full of examples of physical specialization having large advantages in efficiency and performance to do any particular physical task vs any kind of general purpose approach.
Consider what we do with the "sub-robots" we already have in our tools. It's common for many ordinary people to have a number of slightly different hand-held motorized rotary tools rather than one general purpose one, for example, a drill / driver, an impact wrench, a "dremel", a compact router, a motorized ratchet driver, a polisher, and arguably circular saws and angle grinders qualify. That's not counting larger rotary tools like a drill press, lathe, or bench grinder. It often doesn't make sense to try to perform many of those functions by means of swapping out attachments or bits or making modifications to a single motor tool, though Lord knows I learned that the hard way.