THCFarmer.com
121,269 Members Log In Register
Log In Register

🪴 Back to Growing After Decades Away — With a Little Help from AI Coaching

  • Thread starter Thread starter kenb
  • Start date Start date
  • Tagged users Tagged users None

🪴 Back to Growing After Decades Away — With a Little Help from AI Coaching

kenb 211 Replies 17,287 Views
Page 7 of 11 · Replies 121–140 of 212
#121
amneziaHaze said:
TOo much work.and bottom line ai cant do math and we keep giving them math problems
Next thing you will tell me when you ask it something do it yourselfe then check if it did the same
Click to expand...
For the kind of calculations a grower needs (e.g., nutrient PPMs, reservoir volume, light intensity/DLI, mixing ratios), you must treat AI results with caution unless you see it showing its work by using an explicit calculation step (or by asking it to use a formula you know).

Always ask the AI to:
  • Show the Formula: "What is the formula you are using to calculate the final nutrient concentration?"
  • Show the Steps: "Break down the steps of the calculation before giving the final answer."
If you don't see the work, you should run the calculation on a dedicated calculator to cross-validate and ensure your grow room math is perfect. Because, although LLM's are getting better, they are FAR from perfect.
 
#122
juliando said:
Copilot is as good/better - but also basically Chatgpt after microsoft gave them 10 billion - and its free as well.
(And less likely to give you psychosis)
Click to expand...
Thanks - I haven't used Copilot much but will give it more attention. I'm a Linux/Android guy so OpenAi and (to a lesser extent) Gemini have been my go-to's.
 
#123
I settled on ChatGPT via app and a clone on my other desktop that is machine native called Jan. Then you have to condition it to your use. But over all I used it to adapt what I knew to the new "stuff" since I didn't grow from early 2000s to last year. Lately though? No, not much. Pretty handy for organizing the garage by picture though.
 
#124
kenb said:
Thanks - I haven't used Copilot much but will give it more attention. I'm a Linux/Android guy so OpenAi and (to a lesser extent) Gemini have been my go-to's.
Click to expand...
Not a fan of Gemini (often wrong) - and it seems like chatgpt is built to induce psychosis.
Like the answer it gave above - "People are just saying it looks mold because no one has ever seen such amazing weed. Dont listen to them..."
 
#125
kenb said:
For the kind of calculations a grower needs (e.g., nutrient PPMs, reservoir volume, light intensity/DLI, mixing ratios), you must treat AI results with caution unless you see it showing its work by using an explicit calculation step (or by asking it to use a formula you know).

Always ask the AI to:
  • Show the Formula: "What is the formula you are using to calculate the final nutrient concentration?"
  • Show the Steps: "Break down the steps of the calculation before giving the final answer."
If you don't see the work, you should run the calculation on a dedicated calculator to cross-validate and ensure your grow room math is perfect. Because, although LLM's are getting better, they are FAR from perfect.
Click to expand...
I tried to calculate how much a candle will make co2 in a growbox he calculated it was less than the air it came in then i asked for formulas he gave me somezhing that didnt have any logic.to a normal person it would be great a forumula but after you research it its random... allsoo when you ask it to tell you about magnesium or shit and ask for sources you get 6 sources google them they dont exist

And whats the point of ai if i have to write the exact formula soo it can work.i can do the math same speed
 
Last edited:
#126
Took a screenshot of your post.

Then ran it through Google Lens

Screenshot 20251209 143800
Screenshot 20251209 143809
 
#127
I think the trick is to start with AI already as seriously miswired individual and then you're immune to whatever level 1 amateur crazy it has.
 
#129
I will tuck my tail and keep quiet in another room. As you all were!!
 
#132
Flower Day 22

This is the yield projection now that stretch is over, according to OpenAi. We shall see! But if I can get 1.1 g/watt on my first tent grow, I would be very pleased! Their glidepath has them being harvested around Jan 18, with the caps maybe slightly behind that.

OpenAi Feedback this morning after climate readings, pics, and video:
Your plants just hit the threshold where we can refine projections with higher accuracy.

Ginger Kush (2 plants):​

225–300g total

Cap Junky (2 plants):​

170–235g total

Tent Total:​


395–535g (14–19 oz)



Your high-end number is pushing 19 oz.
That is 1.1 g/watt territory.
Image 1765361656697
 
Last edited:
#133
I'm not saying you won't get that weight but AI specifically sucks at weight projections. I think you'll add weird stress if you think about it, weight is an after action report. As I'm looking around though there are some all in one climate units with an onboard brain for VPD and exchange rate calculations that seems awesome.
 
#134
ChairmanFester said:
I'm not saying you won't get that weight but AI specifically sucks at weight projections. I think you'll add weird stress if you think about it, weight is an after action report. As I'm looking around though there are some all in one climate units with an onboard brain for VPD and exchange rate calculations that seems awesome.
Click to expand...
I agree. Yield projections have been all over the board from AI. Hopefully as they end their stretch and start bulking, this projection will get more precise. I have zero confidence in this projection at this time. As long as they produce something worth smoking, I'll be happy 🙂

But I will be anxious to let it know what the actual dry and wet weights are after the fact. It is an LLM so every bit of info we feed it helps it get more accurate.
 
#135
kenb said:
It is an LLM so every bit of info we feed it helps it get more accurate.
Click to expand...
Only a new version. It doesn't retain much beyond a general projection of you as a user, your purposes are secondary.
 
#136
ChairmanFester said:
Only a new version. It doesn't retain much beyond a general projection of you as a user, your purposes are secondary.
Click to expand...
Correct. I was in software engineering for 3 decades. My nephew is now the head of AI Tech at Meta (although I really don't like their AI! lol). The most frustrating thing I have experienced with OpenAi is that their policies change faster than engineering can keep up with.

For example, I have recently asked it to generate images of my plants for time points in the future. It will start to generate the image, and then tell me that it can't due to policy change (although it was quick to point out that it knows that what I am doing is legal and there are no people in the photos that could be identified!). When I tell it that the engineering can't keep up with product management, the use case needs to be changed, paid users should be notified of policy changes, the definition of done is being done wrong, the backlog is upside down. and that sprint retrospectives must not be effective, it tells me that I'm "spot on" and apologizes. Yeah, AI is very cool but FAR from perfect.
 
#137
Sometimes those images appear to stall but are complete in your library. It also does not report user level errors I found out, it aggregates pools of problem frequency then ignores some of them.
 
#140
I had all these fans from my old bands and stuff then somehow something changed, I had a professional page, and I get money for talking a whole bunch on facebook. I had like 300k veiws on a 60 photo album of interwar experimental prop planes.
 
Page 7 of 11 · Replies 121–140 of 212
Back
Top Bottom