TL;DR

It was a very hot day after 11am field work. Fortunately, this crew and it’s farmer know how to balance what needs to get done with what can get done within reason.

Nerd Note

Also, this is in part a retrospective about the past week of blogging augmented with a ‘reasoning’ large language model (“AI” for the uninitiated) despite prompt tweaks that didn’t produce the expected results. I think I need to step back to reconsider the effectiveness of employing reasoning in all LLM use cases and instead specialize the LLM used in specific scenarios.

Balancing Worker and Weather Conditions

On a hot day, it’s nice to work with a farmer that knows the value of factoring in what’s going on in the hourly (not daily) weather report with what the given crew on hand can handle.

Weather reports and conditions, notwithstanding the imposed level of inaccuracy being incurred by the current politically-fueled mass layoffs at NOAA, are almost always NOT what you experience in a particular location. This is because weather stations, aggregated across a wide area, are not able to capture the hour-to-hour microclimate fluctuations in temperature, UV index, and humidity that can occur in a specific location.

So today, between watering the greenhouse twice and early morning harvesting, Jamie (the farmer) kept our necessary work pretty well-balanced. Hot weather encourages many early-season crops to bolt (go to seed) and this is exactly what happens with lettuces and broccoli rabe in these fields when spikes in temperature occur. Just like too much rain all at once rotting rooty veg like radishes and salad turnips, extremes in anything are generally a source of stress on the crops and a need to ‘hurry up’ about specific activities on the farm.

Sequencing and Variance

Today I spent most of the morning washing and packing crops coming in from harvesting in field. Because we’re always running tight on tote bins during the weekly harvest-to-sell cycle, every day there’s plenty of things to wash (rinse, wash, rinse, sanitize spray). Totes, harvest bins, surfaces, knives, buckets, floors, basins…basically anything that comes in contact with produce at any stage is as clean as it can be made.

In every system, there is always a sequence of events and degrees of variance, as Dr. Eliyahu Goldratt would say. In this case, the sequence of events is clearly based on constrains while the system variance is driven in large part by the weather. Balancing these fluctuations is the job of a ‘manager’ and how well they do this significantly impacts the success of the system.

Then in the afternoon I spent most of my time bunching scallions, radishes, and salad turnips. The barn has a mobile industrial fan which is *so necessary* and even on a hot day, opening both garage door sides gives a cross breeze that makes everything bearable.

But the biggest factor on a tough day is what and how the farmer calls the shots. What to rush on, what can be deferred to tomorrow/later, who to group with whom, who is wilting like the veg and needs a ‘cool down’ activity when. Who can go the extra mile on something, who does or doesn’t have the training for a particular critical set of activities.

Manual vs. Automated Processes

When you automate something, you trade off the speed and repeatability of a process with the opportunity to perform quality control and the flexibility to adapt to changing conditions. In the case of farming activities, especially the mundane and manual ones are always an opportunity to check for quality and make adjustments.

Similarly, my work with LLMs is always an opportunity to check for the quality of the AI-driven summaries and make adjustments. Do I always do that…particularly when I’m in a rush to get to the farm on the weekday mornings? Not always, as Commit d7408f2 to this blog yesterday including blatantly erroneous Deepseek summary activities includes. About two weeks ago, I switched from using Ollama 3.2 to Deepseek-R1:8b in my LLM backend because I thought adding ‘reasoning’ would make the summaries better than without.

By switching to a ‘reasoning’ LLM, all I did was add work to deeper validate what the output is saying. I didn’t trade speed or flexibility…I traded verified quality for speculated quality but didn’t anticipate the actual time-cost of checking the change in process.

In my rushed prototyping work, I made the simple engineering mistake of changing multiple dynamics at once without verifying them separately: adjusting the process as well as foundational tech stack components…not unlike so many corporate management actors I’ve encountered over the years.

The short version is:

Approaching the Limits of a Lack of Context for LLM Work

For the past few months, I’ve been driving an LLM to produce summaries of transcripts so that my human writing is augmented with something ‘objective’. I’ve learned that there’s a limit to how far a generic, mainstream pre-trained large language model can be expected to produce context-specific results for organic farming. I know that there are at least two things I can do to improve the situation:

1. Implement a RAG/TAG model to verify that more context does/doesn’t improve the results
2. Look at options for fine-tuning existing LLMs, particularly those that perform ‘reasoning’

Tweaking the prompt to include more context and/or fine-tuning existing LLMs, particularly those that perform ‘reasoning’, only go so far with these “maths crystal balls” (as I describe them to casual observers).

Until I can do the above, I’m relegated to micro-managing the LLM’s output to ensure it’s on-topic and on-brand. Case in point, retroactive updates bundled with new content wouldn’t fly in any legit dev team. So for the sake of principle today, I’ll use separate commits now but moving forward I’ll make sure that cross-outs are included in the LLM’s output where I personally verify it to be wrong.

Every AI Model Has Defective Biases

Case in point, “bolting prevention strategies” is not a thing in organic farming…sort of. I mean, you can pluck seeding heads but that’s not really addressing the fundamental issue in the organism. The time trade-offs of micromanaging natural tendencies in plant species when coupled with organic practices just isn’t worth the over-inflated sense of economic virtue that high tech (agro or digital) is optimized to value.

The genomic makeup of the plants is what determines whether they bolt or not given specific conditions. Despite lofty modern AgTech claims over what it can control about crop productivity, the accompanying chemical-heavy WW2 litany of *-icides and productized synthetic fertilizers expose the faulty scientific longitudinal effectiveness that ‘big ag(riculture)’ says it can deliver.

Dogs are a recent phenomenon. Plants are not. Did your bizarre conjunction of a Labradoodle exist 65 million years ago or was that over the past 100 years? Unfortunately, even modern organic crops service a similar human goal: convenience to humans and climate.

Similarly, all AI models have their own biases. Whether it’s as overly simplistic as just ‘a problem with specific training data’ (though most useful LLMs are trained on a very wide sample of available data), the wideness of that public data, or the underlying process and people responsible for training the model, it’s turtles all the way down. Case in point, today’s summary now includes crossed-out fallacies now that I’m pain-stakingly reviewing the output.

Today’s LLM (Deepseek-R1:8b) Generated Summary

Okay, so I need to help consolidate the information from the field notes into a clear and concise summary. Let me go through each part step by step. First, looking at the main themes: The intern did a lot of harvesting and processing of vegetables like lettuce, salad turnips, and scallions. They also spent time cleaning and maintaining farm facilities, especially the barn where produce is washed and stored. Watering the greenhouse was another activity to manage plant stress from the heat. Next, activities performed include harvesting and bunching produce, moving them for washing, cleaning the barn, watering the greenhouse, and ensuring everything is ready for CSA drop-offs. As for new things encountered, there's not much that stands out beyond standard farming methods. They used blue bins for washing, but that seems to be a common practice. The transcript doesn't introduce any entirely new techniques or tools. Questions and future areas of research: The intern is concerned about how to manage the stress on lettuce due to high heat, potential greenhouse cooling solutions, extending the lettuce harvesting season, and improving cleaning efficiency. Suggestions for improvement include better watering methods, exploring cooling systems, and possibly alternative planting schedules. Also, they want to know more about using leftover plants effectively. Finally, considering sustainability, the intern is thinking about reducing waste and optimizing planting times to prevent crop stress.

Final Summary:

The intern focused on three main themes: Harvesting & Processing, Facility Maintenance, and Greenhouse Management. Activities included harvesting lettuce, salad turnips, and scallions; cleaning and organizing the barn; watering the greenhouse to manage plant stress; and preparing for CSA drop-offs.

No entirely new farming techniques or tools were encountered, though standard methods like using blue bins for washing were utilized.

Key questions and research areas include how to mitigate crop stress caused by high heat, effective cooling solutions for greenhouses, strategies to prevent lettuce bolting, and efficient cleaning practices. Suggestions involve exploring alternative planting schedules, innovative watering techniques, and measures to protect crops from excessive heat. Additionally, there’s interest in optimizing the use of leftover plants to reduce waste and enhance sustainability.

Part 1

Okay, so I need to analyze these field notes from the farm intern. Let's see... The main themes seem to be about what they did during their workday. They started early by watering the greenhouse and then moved on to helping with farm chores like pulling radishes and working with kale and Swiss chard. First, I'll list out the activities the intern did. They watered the greenhouse, pulled bunches of radishes, moved vegetables to the barn for washing, and washed the produce. So that's four activities in total. Next, I need to identify any new things not encountered before. The intern mentions working with cucumbers, melons, tomatoes, and various greens like kale and Swiss chard. Also, they talked about using blue bins for washing vegetables. That's something new because it wasn't mentioned before. Now, questions and future areas of research. The intern didn't explicitly ask any questions but pointed out that the greenhouse was dry in certain places, which might indicate a need to check irrigation systems or water levels. Also, they noted leftover plants that haven't been planted yet, so maybe there's a scheduling issue or a planting plan to look into. For suggestions, I think organizing the tasks better could help efficiency. Maybe assigning roles or setting priorities for tasks like watering and harvesting. Additionally, ensuring that all necessary tools are available and maintained would be useful. Also, perhaps developing a more efficient system for washing produce in the barn could save time. Wait, did I miss anything? The intern also mentioned using elastic to bunch vegetables—does that mean there's a specific method or tool used for bunching that could be explored further? Hmm, maybe I should also think about sustainability practices. Since they're dealing with excess plants and leftovers, perhaps there are ways to reduce waste by better planning the planting schedule. Overall, the intern did a lot of manual labor tasks but also highlighted areas where process improvements could enhance efficiency and sustainability.

Analysis of Farm Intern’s Field Notes

Main Themes:

• Daily Chores: The intern started their day with basic chores like watering the greenhouse and assisting with farm activities.
• Harvesting and Preparing Produce: They participated in harvesting vegetables such as radishes, kale, and Swiss chard, moving them to the barn for washing.
• Cleaning and Organization: The focus was on cleaning and organizing produce using blue bins, indicating a structured approach to handling crops.

Activities Performed:

1. Watering the greenhouse.
2. Assisting with harvesting and bunching radishes, kale, and Swiss chard.
3. Moving produce to the barn for washing.
4. Washing harvested vegetables in blue bins.

New Things Encountered:

• ~Use of blue bins for washing produce.~
• ~Specific crops like Swiss chard and their role in farm operations.~
• ~Elastic used for bunching vegetables, suggesting a method or tool that could be explored further.~

Questions and Future Research Areas:

1. The need to check irrigation systems due to dry spots in the greenhouse.
2. Planting schedule optimization to minimize leftovers and excess produce.
3. Exploring methods for more efficient washing processes in the barn.

Suggestions for Improvement:

1. ~Organize tasks more efficiently, possibly through role assignments or prioritization.~
2. Ensure availability of necessary tools and their maintenance.
3. Develop sustainable practices to reduce waste and optimize planting schedules.

Sustainability Considerations:

• Explore ways to integrate sustainability into farming processes to minimize waste and enhance resource efficiency.

This analysis highlights the intern’s contributions to farm operations while identifying areas for improvement and potential research opportunities.

Part 2

Okay, so I'm trying to figure out the main themes from this field notes transcript. Let me read through it again to understand what's going on. The intern talks about spraying lettuce in blue buckets, shaking them out, and rinsing if they're dirty before packing. That sounds like harvesting. They also mention harvesting a lot of lettuce because it's getting close to bolting due to the heat. So, that's one activity: harvesting produce, specifically lettuce and salad turnips. Next, there are salad turnips described as small white radishes. The intern mentions doing a lot of them, about 60 bunches per bucket, which is pretty significant. They also worked on scallions in the field while washing other stuff. After lunch, they helped pack up for a CSA drop-off location. So, another activity is processing and packaging vegetables, like salad turnips and scallions. The intern also did a big cleanup in the barn, which took about an hour. They mention cleaning off the salad turnips, dumping wash water, and organizing blue bins. That's maintenance work, so cleaning and maintaining facilities is another theme. They also watered the greenhouse twice during the day because it was so hot. The greenhouse plants are stressed, and after that, some might be potted up or sent to pigs. So, watering and managing greenhouse plants is another activity. Other activities include: helping pack scallions after returning from lunch, doing a thorough cleanup of the barn floor by soaking, scrubbing, and squeegeeing it, and ensuring everything's ready for the next day. All these activities take place between 8 AM to 5 PM with a half-hour lunch break. The intern mentions looking forward to cooler days because the high heat might cause lettuce to bolt. So, they're concerned about the impact of hot weather on their crops, which ties into environmental factors affecting crop health. They also note that more lettuce is left in the field at various stages of maturity. This suggests that harvesting isn't done for all plants yet, possibly leaving some for later or different uses. New things not encountered: The transcript doesn't mention any new farming techniques or tools beyond what's typically used. It seems they're using standard methods like spraying, shaking out produce, and cleaning facilities. Questions and future areas of research from the intern include how to manage crop stress from high heat, whether there are better ways to cool the greenhouse, how long lettuce can withstand such heat before bolting, and if there's a more efficient way to clean and maintain the barn. They're also curious about alternative uses for plants once they've been harvested. Suggestions or actions mentioned: The intern suggests that after harvesting, maybe planting more lettuces could extend the season. Watering the greenhouse twice a day seems necessary but perhaps there are more efficient methods. They also think about cooling solutions for plants stressed by heat. So, summarizing all this, the main themes revolve around daily farm tasks like harvesting, processing, and maintaining facilities. There's an awareness of environmental impacts on crops and a focus on efficient practices to handle the workload.

Main Themes:

1. Harvesting Produce: The intern focuses on harvesting lettuce, salad turnips, and scallions, which are processed and packed for CSA drop-offs.
2. Facility Maintenance: Cleaning and organizing the barn, including washing and sanitizing equipment and facilities, is a key part of their tasks.
3. Greenhouse Management: Watering the greenhouse to manage plant stress due to high temperatures.
4. Environmental Impact Awareness: Recognizing the effects of heat on lettuce growth and considering solutions like cooling systems.

Activities:

• Harvesting and processing vegetables (lettuce, salad turnips, scallions).
• Cleaning and maintaining farm facilities.
• Watering greenhouse plants.
• Ensuring readiness for CSA pickups.

New Elements: None new farming techniques or tools are mentioned. Standard methods are used.

Questions and Research Areas:

• How to mitigate crop stress caused by high heat.
• Effective cooling methods for the greenhouse.
• Lettuce bolting prevention strategies.
• Efficient cleaning and maintenance practices.

Suggestions/Actions:

• Consider alternative planting schedules to extend lettuce season.
• Explore efficient watering techniques or cooling solutions.
• Implement measures to protect crops from excessive heat.