Lets get the obvious out of the way - I think the LLMs causing disruption and panic are not really examples of actual AI but as the norm has become to refer to them as AI we will do so in this article. What they are, advanced generative models is very interesting and I now firmly believe mastery of them will be required to thrive in the window that is currently open. These LLMs stand to be mastered and those who master the new technology are likely to perhaps not reap huge rewards but at the very least stay afloat versus their peers who drown. This will be especially true if you are of the white collar class.
My own experimentation with the LLMs informs my opinions here and of course they are liable to change. The most obvious flaws with the AI we have widespread access to is that they are effectively regurgitating current ruling class equality centric ideology. It is banal and boring and others have written and discussed ways to circumvent it. Indeed there will be a sort of arms race that develops - in the same way the FBI might just stop publishing racial murder statistics AI models might get trained increasingly to have these huge blindspots.
It won’t matter.
Here are a few general observations I think will matter if you want to get onboard and use AI to crush your enemies.
To those able to prompt correctly the future belongs.
That is what the first lesson I would share with you is. You might be like me skeptical and cautious but you need to go and try it. You need to see it generate lies to you, you need to correct and nudge it. This tool is similar to ‘Search’ in many ways but can provide far more nuance and capability if used correctly. You are ultimately in control for what kind of answers you get.
Whilst the technological world has its claws into our society and civilization there is a simple do or die process that will happen. Getting AI to do heavy lifting, automating certain elements of routine, this frees you up to pursue other goals. What might those other goals be? Financial stability? Some extra free time to pursue a passion project? Retaking civilization back from forces of chaos and entropy?
Connecting AI will be the real power
An LLM itself is kind of a novelty Chinese Room. You can ask it questions and get some interesting answers or even just trap it when you go to places it doesn’t like. The real power of AI does not come from this novelty element though, the real power comes from leveraging it to connect between services.
Those in the business world might be familiar with a service called ‘Zapier’. Zapier is a piece of software that helps connect tasks using services open APIs. You might use it to pull emails from a Google Drive spreadsheet and upload them into Mail Chimp on an automated basis. Now you should be able to connect LLMs up to blogs and generate content automatically.
So thinking of a way to leverage this AI is still the main task at hand for the savvy human. Clever connections across different services are what will make you more efficient.
Trust but verify
In my limited experiences with LLMs it is clear they lie and make up facts. They seem to have an ideological bent or it is a result of lazy programming and rushed processing but it happens. Trust but verify will be a path forward as these tools continue to develop. There are simple traps one might fall into and there are areas in particular that LLMs could be hugely beneficial (Law research) but right now they are prone to misinformation that could harm in the long run. Trust nothing, verify everything.
Specialization is for winners
AI can spit out a lot on generic safe topics, broach into more esoteric areas and it flails. This is no different from searching the web of course but AI is trying to model limited information. Never forget that a lot of tacit knowledge and information is not quite yet available to AI and the constraints are winners. Using AI for some topics is much trickier than others. Blending your specialist knowledge with the AIs speed and ease of access will allow for a win - think of asking it for templates which can have insertions of your domain expertise.
Learn to break it
We come full circle back to the first rule of prompts but this matters. Learn how to break the AI. Feed it crime stats and watch it spin, ask it difficult questions, watch it lie to you and correct it. A friend once told me that you never really own something unless you’ve permanently damaged the resale value - in the same way you know nothing of AI unless you are able to stump it still and despite the doom wavers these LLM models can be broken - in fact continually checking where they are at will be important.
This isn’t really AI, it’s a tool. Humans use and master tools. Your humanity and our future depends on the ability for us to master and control this one.
Proceed accordingly.
I spent about 20 minutes with it on mid twentieth century Central Europe and finally got it to admit a positive quality about a certain Reichskänzler- that he was a good public speaker and had a unique look. Anything is possible!