I have been using ChatGPT for some time to make a whole bunch of useful lists for my agricultural pursuits. I often struggle to remember details of a plants use, or a way to detect a deficiency. Here are my learnings about using the tool and some of the issues I overcame, and ones raised.

---

Technology is not to simply make your life easier, it's to reduce the mundane so you can spend your time doing more complex things. For this reason, I sat on the sidelines watching AIs enter the marketplace, and thus the stories ensue of how it writes code and thesis papers. Eventually, I figured I would try it out, as I had grown tired of searching for poor lists of carbon/nitrogen ratios for compost input materials. You see, at this time I doing my consultant training for the Soil Food Web School, and building large successive thermophilic compost piles is a big task and in the years up to this point, I had already had a number of failed piles. 

So what started as a question about what materials I could use and what the C:N values were, I naturally started testing AI and changing my mind about it. I found my mundane task was filled by a passive-aggressive moody-teenager-like AI called ChatGPT. I say that because I can just feel the Animatrix-like machine-human disdain as it complains each time I ask it something. You quickly learn that the platform (even the paid version) has a batch size limit and will tell you how super huge this request is and I'll just do this instead. In some cases, it even told me it was just going to make s**t up because the task was just too big. It also doesn't like to be asked what its sources are so it will give a boilerplate response about the date it last looked at the internet. 

Nevertheless, I started finding a usage that I think can be very helpful, and quite dangerous at the same time. You see, if one thinks about the essence of what an AI is, requires looking at the human context of it. Humans have produced a world of content throughout history. Anything anyone ever said (and was remembered for it) is content. If Humans publish this content, then something that is capable of reading it, indexing it, and making sense of it (like Data from Star Trek?) surely is humankind's greatest tool for the empowerment of knowledge. 

But alas, content is King and he demands his tithe. Content providers seek to earn money during a time when social media platforms are suppressing content sites that are struggling to earn a dollar for their work. So if on day 1 the AI has no access to things like let's say, scientific journals or university archives then how smart will it be? 

The first thing I mentioned was to ask about C:N ratios. I also needed to revamp my CV so I had it do that as well. Because hey, in this day and age, every recruiter wants you to customize your CV to their job so another AI can read it for them and call you based on some literal keywords that are present that the recruiter doesn't even understand anyway (did I mention I come from the digital technology sector originally?)  

So over the last month, I have worked with ChatGPT to do something I think is useful, at least to myself which is to create lists. I love lists, master spreadsheets, and things that give you a helicopter view of something like a business or a mass of concepts. And while the cranky teen ChatGPT will attempt to get out of its chores and instead play in a pile of dirt, it will comply if you learn how to get beyond the batch limits.

As its name suggests, it is a chat bot meaning it is a conversationalist. And a conversation about anything is about setting a context (unless you're my wife whose brain works faster than any computer in existence and I regularly miss the signal we've moved onto an entirely different conversation). This means it remembers stuff. So if you need to crunch a huge list, give your instruction in a few steps. First I tell it I want to make a table of something, and I say what it should look like. I want my list filterable in Excel so I can play with the data. So telling it to make a table with columns outputs in your screen something that you can then copy to Excel directly, or better paste in a text editor to drop the HTML formatting. Then you can do a few different things. Start by giving it free rein on the answer and then redirect with suggestions. You can also say things like "Continue the list with 20 more". You will notice exceeding certain limits of job size, the quality of results degrades as the corners are cut by the teen who wants to play Xbox, not do homework.

After making a bunch of lists available below, one starts to see both sides of potential gains vs. boons coin. I was able to collate massive lists simply by hiding the Xbox and blackmailing the teen into doing chores for playtime (repeatedly). But each response comes with ChatGPT's own adlib and what it thinks about your request. It at least has the training by parental units to caveat everything it says and to encourage people not to be idiots and jump off cliffs in case their teen was extra cranky that day. Fair enough. However, when it tells you at one point (something like) "I want really sure, so I just made shit up" you wonder if anything it said at all is credible. How do you know if it's accurate? In the case of my lists, there is so much data in there, how would one ever know it's accurate? 

So I started placing my own caveats in the final work which says something like "THIS DATA CANNOT BE TRUSTED" which seems like an oxymoron. But if one uses this as a starting point to reconfirm through their own work, this can still be useful. I have found this to be a tremendous exercise for the following reasons:

• It has certainly reduced a lot of mundane research to create these lists. Even assuming its 100% wrong, I can at least go confirm for myself
• It provides context, I often find that seeing everything on the wall at the same time sheds light. Basically this is every sleuth movie with the wall full of photos all tied together by grammas knitting yarn.
• I can help sediment memory. Remembering a speech was carrying around cue cards. Having lists to consult often means you are often thinking about the subject and it further establishes a firm hold in the grey matter.

Below are some lists I have created. But first the caveat: I used an AI to create these lists, and then with effort I compiled fragments into a whole and cleaned it up removing duplicates. I cannot attest to accuracy of a single thing, so....

DO NOT USE THIS AS LITERAL INFORMATION WITHOUT CONFIRMATION FROM AN EXPERT.

Compost input Carbon-Nitrogen ratios - A few hundred things that can be used in compost along with the ratio and some basic reasoning to if something is good or bad for compost.

Bacteria - A list of bacteria that are relevant to soil and human health (we now know these are the same thing), pathogenic or beneficial, what they do, how it's found, nutrients it can possibly bind to, and a historical tag (year and discovering scientist). 

Foliar sprays and soil drenches - Materials that can be used in a foliar spray or soil drench, what it does, what it contains, and the solvent and mix ratio. This is one of the few times I managed to get references to a source. It seems two websites were used here. 

Forage for livestock - plants that are good or bad for livestock, nutrients it contain, and the parts safe to consume.

Remedies for pressure headaches - A friend came over who suffers from chronic migraines. This list makes up a rating of effectiveness and contains references to source websites. It's nice to see some well-respected sources.\

Element-plant relationships - In standard agronomy, it's well-known how minerals play with and against each other. While modern regenerative agriculture may put the onus on biology instead of chemicals, it doesn't mean this knowledge is not important.  What does the element do for a plant, what nutrients does it displace, how do you recognize deficiency versus excess, mobility, history of the particle, where a plant would find it in nature, what is the method of uptake, and what parent material is it found in.

Planting Guide - This is about 20 hours of work which required 4 ChatGPT sessions to be accumulated. Plants, the Latin name, family, the usable parts, zone, relationship to microbiota, times for seeding, sowing, harvest, expected disease, innate resilience vs weakness, temperature tolerances, triggers for reproductive cycles, propagation methods, and finally a year schedule the says when to ideally sow, transplant, harvest based on my Zone 5 location in Ontario, Canada.

Freshwater pond fish - In truth, the very first thing to think about in any permaculture or regenerative agriculture system is water. Building ecosystems that see the water is collected, cleaned, and made useful to all sorts of life including pond fish. A basic list of fish, origin, and temperature thresholds (and yes, water is capable of being a negative temperature that is not a mistake).

Herbal preparations - What are different herbal preparations, what they can do, what chemicals are contained, and how they are prepared and administered. I even added a chicken column (tbh it was too much work to compile this for different livestock)