Chat GPT designed this NFT system, what's wrong with it?

[jamestee98]

Hi All,

I am new to hydroponics but I am really keen to create a compact NFT system that fits on my balcony.
I know that other techniques may be better in this situation but I think NFT looks cool and want to try it out so please assume that is an essential part of the system. I wanted to design the system all theoretically before I paid for anything. There is so much information online so I turned to my friend ChatGPT for help. Below is a summary of the system we came up with. Does anything in there look crazy or has Mr. GPT done a good job? I think it really struggled with the pump power calculation!

"JTS System Summary 1:
• System Type: Nutrient Film Technique (NFT)
• Number of Channels: 2
• Plants per Channel: 2 (Chillies)
• Plants per Channel: 2 (Strawberries)
• Channel Dimensions: 50cm (L) x 7.5cm (W) x 15cm (D)
• Plant basket dimensions:
• Strawberries: 5cm (W) and 7.5cm (D)
• Chillies: 5cm (W) and 10cm (D)
• Reservoir Capacity: 8 liters
• Reservoir Dimensions: 40cm (L) x 10cm (W) x 20cm (D)
• Desired channel water Depth: 4-5mm
• Pump Power: 30 watts
• Flow Rate: 1-2 liters per minute
• Recommended EC (Electrical Conductivity) and pH ranges:
• Seedling Stage: EC: 0.8-1.2 mS/cm, pH: around 6
• Vegetative Stage: EC: 1.2-2 mS/cm, pH: around 6
• Flowering Stage: EC: 1.8-2.8 mS/cm, pH: around 6
• Nutrient Formulation:
• To achieve these general macronutrient concentrations:
• Nitrogen (N): 2-3%
• Phosphorus (P): 1-2%
• Potassium (K): 3-5%
• Calcium (Ca): 1-2%
• Magnesium (Mg): 0.5-1%
• Sulfur (S): 0.2-0.5%

• Macronutrients:
• 20-20-20 N-P-K Mix
• Nitrogen (N), Phosphorus (P) and Pottassium (K) and Add 64 grams of 20-20-20 mix to the 8L reservoir
• Calcium Nitrate:
• Calcium (Ca): Add 3.2-8 grams of calcium nitrate to the reservoir
• Epsom Salt (Magnesium Sulfate):
• Magnesium (Mg): Add 1.6-4 grams of Epsom salt to the reservoir
• Chelated Micronutrients: Follow the manufacturer's instructions for appropriate dosage and mix with the nutrient solution.
• Silicates: Add potassium silicate to the reservoir to achieve the desired concentration of 50-200 ppm.
• pH Adjustment:
• If the pH is too high, gradually add a pH down solution (e.g., phosphoric acid or citric acid) while monitoring the pH until it reaches the desired range.
• If the pH is too low, gradually add a pH up solution (e.g., potassium hydroxide or sodium bicarbonate) while monitoring the pH until it reaches the desired range.
• EC Adjustment:
• If the EC is too low, increase it by adding more nutrient solution or specific nutrient salts according to the required elements.
• If the EC is too high, dilute the solution with water or adjust the nutrient concentrations by adding water and reducing the amount of nutrient solution added.
• Temperature maintained at 18-22 degrees Celsius.
Please note that the specific dosages and adjustments may vary depending on water quality, plant response, and other factors. Regular monitoring, observation of plant health, and adjustments are crucial to maintaining optimal pH and EC levels throughout the growth cycle in your JTS system."

[webmaster]

ChatGPT is generative AI, not really an AI the way many people think of it.

The difference with generative AI is that all it is doing is outputting words in the order that one word would statistically follow a previous word. Same at the sentence level.

There is no "knowledge" in the output. It's just the algorithm outputting "the weasel" if you input "pop goes," because "the weasel" is what statistically follows the phrase "pop goes."

I haven't had ChatGPT blatantly hallucinate and lie to me yet. But I have had it fake that it knows the answer when it does not. But that's something that generative AI tends to do.

The technique used to train InstructGPT (the underlying technology of ChatGPT) was meant to teach it to provide correct answers but it backfired and turned into teaching it to please human evaluators, not to output the correct response. So it became proficient at sounding authoritative enough to fool the human evaluators and pass the evaluation tests.

It's a technique called, Reinforcement Learning from Human Feedback (RHLF).

AI training is a "reward" positive-reinforcement type scenario, where the AI is trying to receive a positive evaluation, like a dog that's trying to earn a treat for following commands. So, rather than learn to provide the correct answer, it learned to provide a correct-sounding answer.

[jamestee98]

That's cool thanks for the explanation of ChatGPT. So are you saying the system will work?