TL;DR

We harvested salad turnips, celeriac, and potatoes. I had a hand at chisel plowing the taters up and helped with other heavy lifting chores later in the day.

Tractor Ops

At the beginning of the growing season, I expressed interest to Jamie the farm operator in getting experience on tractors. During some parts of the summer, the tractor work just needs to be done quickly and there’s very little time for anything slow like hands-on training. But progressively, increasingly, I was able to demonstrate responsibility over other advanced tasks and eventually myself and another crew member got time on the tractor.

Now, Jamie will ask me to help by moving the tractor from place to place based on anticipated jobs around the multiple fields across the property. Sometimes this includes discing, plowing, seeding and hauling with the tractor. The logistics are feeling more natural, even between tractor models with different controls. On other farms this would likely not happen due to time, risk, and operator flexibilities so I’m grateful for the opportunity.

What’s “Taters”, Precious?

Unlike Smeagol from Lord of the Rings, I know all about potatoes. Yes you can boil them, mash them, stick them in a stew. But there’s so much more to the lifecycle of potatoes than just eating them. Potatoes are a staple food in many cultures, and they have a long history of cultivation. They are also a versatile ingredient that can be used in a variety of dishes. One of my favorite uses of them is in Indian Chaat, but as I have shared before, my all time favorite way to prepare them is in a simple boil and steam with herbs and simple salt/oil serving.

If you buy potatoes at a supermarket, you’re likely to get a predetermined amount, like a small 1-2 pound bag or a large 5-10 pound sack. Unless you’ve planted, mulched, hilled up, chisel plowed, and dug them up for harvest, it’s very likely you have no idea how disgustingly cheap they are at the store compared to the effort it takes to bring them to market. At this farm, we weigh out the potatoes by hand and put them in compostable bags made of mostly corn starch. At first I was worried they’d soften or mold, but after being bagged for weeks, they are still as dry and crisp as the day they were harvested. A single one-pound bag of our potatoes is well worth the $5 we sell them for at retail. I use them at home almost exclusively, unless there’s a big sale on relatively local potatoes at a local grocer.

Potatoes are ‘tubers’ which means they are underground storage organs that store food for the plant. They are also a good source of carbohydrates, fiber, and vitamins. Sometimes the resulting spud (the nodules of the roots) can be quite small, or large, or somewhere in between. Also the size and starch content can vary greatly based on the variety of potato being grown. I’ve grown many varieties over my decades of backyard gardening…in the ground, in buckets, in grow bags…and no one method seems to be better than another, save that the more space the better. I think the root exudates signal each other to grow further apart if there’s room. Also, the soil can’t be too compact or too loose either because the roots need some compaction to encourage extended growth while the spuds need space to grow as well.

Anyhow, following the above tractor operations experience, Jamie was kind enough to let me practice chisel plowing a row of potatoes up. Later in the afternoon, I hand-dug this row in order to experience the difference in being ‘on the line’ vs. not does to the time and effort follow-up. Let me just say that the rows that Jamie plowed up were much easier to dig than the ones I did, even with him helping me by riding along on the tractor while I drove it.

AI Summary from Field Notes

When the transcription process gets a word wrong, it can throw off the LLM; case in point ‘cedar’ vs. ‘seeder’. My transcription process is based on Whisper C++ which, while it uses some advanced algorithms, is completely disconnected from the context in which the language and words used apply. The only time more domain-specific context to agriculture comes in to play is through my prompts to LLM and the multi-document parsing used to chunk the transcript up into segments sized for the LLM to process.

Were I to address this issue, I can think of a few options already:

1. Use a custom pronunciation lexicon, similar to speech-to-text apps, before producing the final transcript (high effort, low risk)
2. Run the transcription through a domain-specific language model or LLM augmented with embeddings based on prior published works to get a ‘fixed transcript’ (low effort, medium risk)
3. Prompt an LLM for potential ambiguities, mismatched terminology, or unclear language in the original transcript then use the results to refine the transcription (medium effort, low risk)

There might be need to use different lexicons based on different contexts, but this would hopefully be addressed internally by the LLM maths given enough training on domain-specific data. Thus, I come back to my initial hypothesis that better and more long-term solutions to this and other problems with generic large language models is to better train them on more detailed data sets specific to agriculture and farming.

For today, the 90% accurate transcription is good enough for the purposes of this summary.

Summary