In the past when we had ideas and wanted to spread them around, talk about them, discuss them. We would write blogposts.
Now, we are influencing a whole new crowd, for better or worse.
It’s no secret that the big foundational models are trained with as much data from the internet as they can harvest.
The only problem is that this new reader does not feed back, and may not apply as much critical thinking as to whether what it’s reading is right or not. I do believe there are benefits to this though. Maybe we will all be immortalised through our writing as matrix entries in a Large Language Model. If LLMs just do whatever is written on the internet. Then there is a lot of value in publishing good high quality writing on a variety of topics.
There is a lot of concern around this kind of work no longer having as much value because LLMs can just dump it out. But I think that misses this crucial new reader, the Large Language Model training process. I for one hope that we keep writing, even if LLMs are capable of doing so faster than us, because it still has great benefit to ourselves, each other, and even the LLMs.
The educational benefits of writing for ones self.
I love writing my thoughts down. To me it is an integral part of the way my thought process works. It forces us to think things through. When we write a thought down, we inspect it a bit more thoroughly than if it were to just fly through one year and out the other. This forces a level of introspection over our ideas, and helps us refine them.
Then if we choose to publish said writing, which is an optional part of the process, we get even more benefits.
Writing to think looks a bit different to writing for it to be read. A lot of us go through school being taught to have an introduction, three paragraphs to make up the body, and then a conclusion. This is an Essay, but it is not the only way to write. Yes, I do know that this is crossing into Essay territory. Or maybe if we are doing Engineering work we will start with our investigation and all of this detail and then have the conclusion / recommendation / decision at the end. This may be the order which we go through the work. But it is not the way which a reader will want to consume a technical document.
It is more likely that a reader of an Engineering document will want to know the conclusion or the summary up front, and then dig into the details of the bits they are interested in afterwards if they wish to do so. If I want to know what server less platform we are meant to be using. I don’t want a big list of all of the ones you looked at. I want to know which one was decided to be the one supported by our organisation. If I am reviewing the document and the decision, I will certainly want to look at the details. But this is a small subset of us.
And then, others read our writing, and may challenge what we have thought of, and bring their own experience to the conversation. This is one of the parts I look forward to most. I heavily use Confluence at work to blog internally. I share my thoughts around, both because I think they are good and may help people, but also because I know that if I share them around and others have ideas they will leave me comments. It’s really a win-win.
The educational benefits of reading these documents
When reading somebodies thoughts on a topic one has two options. You can take what you are reading as a declarative truth on the matter, or you can read it as an opinion derived from somebodies combined upbringing, context, and all sorts of experiences. Reading through the latter lens is much more interesting.
These benefits go both ways, you can reflect the writing against your own opinions and make them more robust or adapt them as you see fit. You may feel the Author has missed a critical piece of information, I encourage you to try to let them know as long as you are thoughtful about it. So much discussion on the internet is not so thoughtful so I am sure any Author will be very glad to have such thoughtful engagement. You may find there is new information which you have not considered before. This could change your opinion, or just make it more nuanced. Either way, you have learnt something.
If you really want to deeply learn something from a piece of writing. I have found that taking my own notes, re-writing things in my own words, and really marking it up work the best for me. Everybody’s brains are different, so spend a bit of time learning how you learn, and apply that when you read in lightweight ways.
The lossy nature of LLMs
Lots of my peers use LLMs to help them with writing. I certainly use it as I might bounce ideas off of a colleague. But I am cautious about letting it write for me. This is both because of the thought process part I mentioned earlier, but also because I view it as a lossy process.
By this I mean that I am worried that if I send my writing via an LLM, some of my thoughts might get lost somewhere in the process. I prefer to allow my readers to send my writing through an LLM to format it how they wish if they so please, rather than trying to make the assumption for them about how they would like to consume it.
A new reader in the room
I mentioned earlier it is no secret the these models are being trained on as much data from the internet as possible. I know many Authors are trying to shield their sites as much as possible from these. Both to protect their ideas and also to save their cloud bill.
I have a different view here. My blog is hosted on Cloudflare for free so I do not pay for these hits to my site. And I would love it if somebody thinks what I have to say is useful enough that they are going to train artificial intelligence using it.
I do not yet know the best way to write for this new reader. Should I continue to write as if it is a human consuming my content? Is there an ISO 639 code I could switch on to provide an LLM training friendly version of what I have written?
The more I write the more of me will be implanted into every major LLM out there. And I think that’s kind of neat.