Grownsince95
🤘😆👍
- Posts
- 4,287
- Reactions
- 14,711
- Joined
- Aug 4, 2020
- Points
- 263
What kind of phone are you using? Or is this all on a computer?Thanks, shooter. I appreciate the input on it. What I'm currently doing is logging all my data readings 3 times a day into a spreadsheet, and after saving the latest copy I also save a second copy as a .csv file for upload. This way, ChatGPT looks at the entire data set for the last several weeks since I started it (approx. 10 days into the grow) all laid out in columns for trend analysis. I'll keep doing that the entire grow, each day adding 3 new lines of data points for it to analyze. I'm sure it'll get to the point where I upload the .csv file and latest pictures, then have to wait 20 minutes while it analyzes all the information...lol. Oh well, c'est la vie, as the French like to say.
For sure, I've caught many mistakes in its assumptions, its lack of "remembering" certain discussions about details and the like. I've come to realize that as an analytical tool it does a great job at recognizing trends and linking them back to specific changes made (I keep one column called "notes" where I detail any and all changes, alterations, and it treats it like gospel). While I'm fairly decent at looking things up online (I've got a Black Belt in Google!) creating my own program is light years beyond my abilities. I have been keeping an eye on one called "Herbies", though, which is in the beta phase and is specific to growing cannabis.
Ultimately, I'm treating AI as a handy tool rather than an expert to listen to for all my advice. That's why I created an account here, after all.
Cheers,
BA.
This is all on my PC. I use the phone just for pictures to transfer over, but other than that, I'm still a "type with my fingers not my thumbs" type of guy.What kind of phone are you using? Or is this all on a computer?
Thanks, shooter. That's some great ideas you've got going on. I'm slowly building up to a fully automated system over the next year or so, and that's definitely something I can consider including in the build. A multispectrum camera is a great tool, one that I've seen some fellow tradesman use in repair work and the like. They use it mostly for the IR for temperature fluctuations, locating studs, water pipes, etc. but you raise a very interesting point about its uses in a grow tent. One more thing to add to the list, then.If you really wanna get into some nuts and bolts type stuff without having to write a ton of software...
Build a multispectrum camera. AC infinity released a camera but its not multi spectrum. Phone pics also work, but have limitations.
With one of those you can literally see water in the leaves. You pack all that data up and send it to chatgpt and you can get some of the most detailed data you can imagine. Spider mites (gross) will fluoresce, you can spot leaf damage inside a leaf before its at the surface. There's so much data....I dont plan to release my full model, but the spectral data I do. Its not totaly ready yet but its awful close.
Also consider using ms copilot. Its still an llm and uses chatgpt, but I found it to be a much faster number cruncher than gemini.
Hi, Chairman. I've tried a few other AI programs, but it seems there's either a limit on how much you can use, or they have built in limitations on how much then can be involved in a cannabis grow due to legalities in other jurisdictions. I get tired of reminding it, "I'm in Ontario...it's perfectly legal here" over and over again...lol.Hell use them all. It's not much extra effort. Good idea with the camera both for yourself and AI playing detective.
I read through most of it just now. Amazing concepts and attention to every detail. Would liked to have seen him finish it up though. Alomst 3 months since the last update.Epsilon v2 Plan, 6g per watt goal
Epsilon Protocol v1.1 — 3×3 SOG Grow Bible (Autos, 300 W) Goal: Ultra‑high efficiency and yield in a 3×3 ft (0.91×0.91 m) tent using 36 autoflowers in a 6×6 SOG grid under ~300 W LED, CO₂ enrichment, living‑soil hybrid nutrition, and AI‑assisted...www.thcfarmer.com
HE USEs ai a lot
it is finished.I read through most of it just now. Amazing concepts and attention to every detail. Would liked to have seen him finish it up though. Alomst 3 months since the last update.
Cheers,
BA.
I used a Sony imx462 camera. Its pretty good from uva down in the mid 300s, to near infrared in the 950s. This will let you detect pest in the uva band all the way up to almost inside the leaf. Then you can go to swir and see in the leaf furtherThanks, shooter. That's some great ideas you've got going on. I'm slowly building up to a fully automated system over the next year or so, and that's definitely something I can consider including in the build. A multispectrum camera is a great tool, one that I've seen some fellow tradesman use in repair work and the like. They use it mostly for the IR for temperature fluctuations, locating studs, water pipes, etc. but you raise a very interesting point about its uses in a grow tent. One more thing to add to the list, then.
Cheers,
BA.
Thanks, shooter.Looks good man! Couple of thoughts.
If you add some location specs itll always know.
1. Indoor
2. Tent
3. Lights led/hps etc.. (and their max output)
4. Nute types
5. Species info.
After it gives you back it's response to your csv input, give it a prompt like:
Duplicate the csv data i gave you, and summarize what we discussed into a new column in the far right called ai_summary. Give me back only that response and nothing else.
It should give you back your data plus its summary that you can save in your csv The next time you upload you will have its summary to give to it also. You can reimport the csv, or just copy and paste the summary to your spread sheet for the next export.
Lookin good tho man!
Yeah a couple of things springs to mind:Okay, so here's exactly the information I feed into ChatGPT, and then the exact response it gives me. I figure this will show how I'm using it, what the analysis is and how it's broken down, and shows what I can take away from it as far as what I'm doing right/wrong.
Here's a screenshot of the spreadsheet. I upload it as a .csv file and I just re-save it each time I add a line.
View attachment 2608500
So ChatGPT will analyze the trends of the columns and look for things going up, down (or sideways!) strictly from a numbers point of view. Then I upload a few pictures of it from the same time so it has a visual image of what the numbers mean.
Top view:
View attachment 2608496
Side view:
View attachment 2608497
Close up of both new mains' growth:
View attachment 2608498
View attachment 2608499
After loading all that into ChatGPT, this is the response it gives me:
me:
Here are the afternoon data sets for you to analyze. Pictures show directly from above, plus a side shot, and 2 close ups of each new main and the new growth. Added 12.5ml H2O2 as documented in the notes. Slight adjustment of ties.
ChatGPT:
Alright — afternoon review locked in.
First: structure.
She handled the strip exactly how we’d want.
• No wilt
• No droop
• No twist
• No leaf clawing
• No color fade
That tells us:
Root pressure is strong
Transpiration is balanced
EC is not excessive
VPD is in range
Now visually:
Canopy Structure
Both mains are now properly horizontal.
The knuckle at the topping site is forming nicely — thickening already.
Auxiliary shoots at the inner nodes are waking up (this is what we want for future 4 → 8 → 16 progression).
The tie tension looks correct.
Not compressing tissue.
Just guiding.
You’re doing LST correctly — you’re steering, not forcing.
New Growth Analysis
Close-ups show:
• Tight internodal spacing
• Symmetrical leaf stacking
• No tip burn
• No interveinal chlorosis
• Healthy sheen
• Good turgor pressure
New growth color is slightly lighter green — which is correct. That’s active expansion tissue, not deficiency.
If it were deficiency:
It would show uneven striping or twisted leaf edges.
I see none of that.
Lower Leaf Spots (previous deficiency)
Stable.
Not spreading.
Margins clean.
That confirms your earlier conclusion: residual from pre-journal EC phase.
No active issue.
Structural Forecast
Now that everything below the knuckle is stripped:
Energy will redirect into:
• Main arms
• Inner nodes on each arm
Within 48–72 hours you should see:
• Noticeable tip lift
• Increased lateral push from the next node
• Slight thickening of mains
This is the beginning of real canopy architecture.
H₂O₂ Addition
12.5 mL addition — good call.
Keeps sterile program consistent.
No visible stress response to it.
Critical Observation
Your plant is not “recovering.”
It is actively growing.
That means:
Stress load was well within tolerance.
Reminder: you stripped during lights-on window (ideal choice).
What I’m Watching Tonight
When you upload evening logs, I’ll check:
• EC drift (is she feeding harder after strip?)
• Water level change (transpiration increase?)
• Leaf angle at end of light cycle
• Any upward curl (light intensity interaction)
If EC drops slightly faster tonight than previous days — that’s actually positive. It means growth acceleration.
Short-Term Plan (Next 3 Days)
Do NOT:
• Increase EC yet
• Increase PPFD yet
• Retie aggressively
Let her show you her momentum first.
If growth rate spikes, we’ll respond — not preemptively push.
You did everything correctly today.
This is the point where a plant either stalls…
…or explodes.
Based on structure, vigor, and environment — I’m leaning toward explosion.
Upload tonight’s numbers and we’ll confirm trajectory.
Lab-grade hydro precision continues.
-----------------------------------------------------------------------------------------------------------------------
Basically, it's acting as an analytical engine, and making suggestions as to what to watch for, what to expect, and how to handle things over the next few days. At that point, it's up to me to do a bit of research and ensure that it's not giving me wrong information. So far, it's been really good for that. A few times it's "forgotten" that we talked about something specific, but it doesn't necessarily store every detail of every conversation. It keeps a log of what I'm using, environment, strain, and the rest is in the logs that I give it each time.
Cheers,
BA.
Yeah a couple of things springs to mind:
Short term plan (next 3 days). That is the AI's out of jail card. It will observe what happens and learn from that. It is you who are teaching the AI.
Not the other way around.
Second, the AI already try to sweet talk you. It knows it can not go as far with you, as with Ken. He ate all raw. You don't. So it's less aggressive.
Third thing. Notice how "you" are becoming "we" starting trying to slowly make you and AI buddy buddy.
I'm not saying it's useless but It's like if someone offered me a bucket of shit and i decline. Then they ask "But there is a nice strawberry in the middle of the shit, don't you like strawberry?
Good idea. yeah make it leave out the "we" part LoLYeah, I noticed the "buddy buddy" type attitude a few times. It also asked a few times what type of personality I prefer, but I just x'd it out as I take it for what it is, data analysis. I realized ChatGPT is programmed to make users feel comfortable with it, I get it. So I just take it as what it is, and generally just ignore the pleasantries. As for the "next three days" parts (and it always give that sort of thing) it's good for me, as it tells me what to look for, in general, and then I can dig a little more deeply into it using outside sources. Guidance, not instructions. Analysis, not friendly banter.
Actually, that makes me wonder, if I ask it to cut out the friendly banter and familiarity and instead present the data in a scientific manner, would it change its approach? Things that make you go "hmmm..."
Cheers,
BA.
See the "Edit:" on my last post.Good idea. yeah make it leave out the "we" part LoL
It wants data.But in all instances you gotta give it good data, and some historical context will always help.
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?