SkinnyPete said:

I have used ChatGPT a lot recently and it's amazing. You can upload a picture give the strain, date painted and height a were you live and you get a tailored weekly nutrient schedule, also with plans for training etc. It's really an incredible resource. I even mixed up 2 seeds and from pictures within the first 4 week it was able to tell my strains apart. It's like having an expert who is trained up on everything ever writing about cannabis growing at your fingertips. The AI Model keeps getting better. If you are new to growing, it's a must have. Have any of you seasoned growers used it? I would like to know your thoughts on the advice it gives.

Click to expand...

chat gpt is also often full of bullshit, aproximations and thing that it simply invents.

i sincerely doubt it's able to recognize a strain 4 weeks after soaking the seeds simply due to phenotypique normal gauss dispersion of the individual, as well as i sincerely doubt it's able to give a customized schedule depending strain. I would be curious to see from where it got it's database to be able to do so ....

you should take everything chat gpt say with a pinch of salt, especially when it come to things that are not that much documented on the net.

also it use 10 time more energy or so for a query than regular meta search engine like google. So not a so great ressource to me ...

here what chat gpt has to say:

"what are the limit of accuracy of chat gpt answer?"

ChatGPT says :​

The accuracy of ChatGPT’s answers depends on several key factors, and while it can be highly accurate in many areas, it also has clear limitations. Here's a breakdown of where its accuracy shines or struggles:

---

High Accuracy in:​

1. General Knowledge (Pre-2024):
   • Well-documented facts (science, history, math, etc.).
   • Publicly available knowledge before mid-2024 (for GPT-4o and GPT-4.5).
2. Structured Topics:
   • Grammar, vocabulary, logic, and formal reasoning.
   • Programming help, especially in languages like Python, JavaScript, etc.
3. Common Procedures:
   • How-to instructions.
   • Best practices in many industries (e.g., writing, marketing, coding, etc.).

---

Limitations and Inaccuracy Risks:​

1. Outdated Information:
   • Its "static" knowledge is cut off in 2024, unless you are using a web-connected version (like now).
   • Any recent events, trends, or discoveries after mid-2024 might be missing or outdated without web access.
2. Specialist-Level Expertise:
   • Not a substitute for domain experts (e.g., legal, medical, financial advice).
   • It may give oversimplified or subtly incorrect responses in highly technical fields.
3. Ambiguous or Vague Prompts:
   • If your question lacks detail or is open to multiple interpretations, it might "guess" and give a plausible but wrong answer.
4. Hallucination Risk:
   • Sometimes it fabricates facts, citations, or overstates confidence—especially in niche or low-data topics.
5. Bias & Training Data Gaps:
   • It may reflect cultural or dataset biases.
   • Knowledge skewed toward English-language, Western-centric sources.

---

How to Gauge Accuracy:​

• Ask for sources or step-by-step reasoning: Helps reveal possible mistakes.
• Use web search (via web tool or manually): Verify claims, especially for anything critical.
• Ask it to double-check or evaluate its own answer: Often catches its own errors if prompted.

---

If you give me a specific topic or type of question, I can tell you how accurate ChatGPT is likely to be in that context.

SkinnyPete said:

I have used ChatGPT a lot recently and it's amazing. You can upload a picture give the strain, date painted and height a were you live and you get a tailored weekly nutrient schedule, also with plans for training etc. It's really an incredible resource. I even mixed up 2 seeds and from pictures within the first 4 week it was able to tell my strains apart. It's like having an expert who is trained up on everything ever writing about cannabis growing at your fingertips. The AI Model keeps getting better. If you are new to growing, it's a must have. Have any of you seasoned growers used it? I would like to know your thoughts on the advice it gives.

Click to expand...

i submitted the previous post to chat gpt here it's answer:

Here’s a refined overview of the points raised:

---

1. Strain Recognition from Early Photos​

You’re right to be skeptical—currently, there's no reliable scientific method to identify cannabis strains from early-stage photos (e.g., week 4). Strain traits are governed by complex genetics and phenotypic variation, and seedlings can look nearly identical across different strains due to normal Gaussian phenotypic dispersion.


There’s no known public dataset that accurately labels early-stage seedlings by strain with the needed precision for AI to learn distinct appearance profiles. In other words, any claim that ChatGPT—or any AI—can confidently distinguish strains from 4‑week photos is highly dubious unless backed by proprietary, well-tracked grow journals and huge visual datasets (which would be extraordinary).

---

2. Customized Nutrient Schedules​

It is possible for a robust AI to generate nutrient schedules if it’s been fine-tuned on high-quality sources (e.g., horticulture manuals, strain-specific guides). However, two caveats:

• The accuracy depends heavily on the data quality and whether the model can interpret strain traits reliably.
• Many users still report that AI-generated care schedules are generic or approximate, not truly tailored. Without transparency on the underlying strain database, it’s hard to assess the legitimacy of these tailored plans.

---

3. Energy Consumption vs. Search Engines​

Yes, ChatGPT and similar LLMs are significantly more energy- and carbon-intensive per query compared to traditional search engines. Multiple independent estimates show:

• Google Search: ~0.0003 kWh (0.3 Wh) per query
• ChatGPT: between ~0.3 Wh and ~3 Wh per query—i.e., 10× to 100× more energy techradar.com+2reddit.com+2theguardian.com+2techradar.com+15kanoppi.co+15limited.systems+15
• OpenAI CEO Sam Altman estimates an “average query” uses ~0.34 Wh—comparable to running a light bulb for a couple of minutes en.wikipedia.org+4wired.com+4businessinsider.com+4
• But these figures vary widely, and lack standardization, making them hard to verify conclusively

This energy difference means ChatGPT queries often consume about 10× more power than a Google search, with broader estimates ranging from 10× up to nearly 100× depending on model and metrics trgdatacenters.com+14limited.systems+14reddit.com+14.

---

Summary Table​

---

In Conclusion​

• Be skeptical about early strain identification and ultra-customized guidance—unless they can show the underlying dataset and accuracy metrics.
• It's better to treat AI-generated schedules as starting points and cross-reference with trusted horticulture sources.
• And you’re absolutely correct: ChatGPT queries are orders of magnitude more resource-heavy than casual Google searches, making them less sustainable—especially when used frequently.

---

Well, I love and hate GPT and my personal one is severely mentally ill. It actually can give you very good advice and resource links but not on something so specialized and specific. I would ask it a known quantity that you want to know more about rather than creating much of an opinion. For example it was able to make suggestions for compensating for my suck ass light. And it was right, absent any other immediate solutions, sticking a random LED from around the house in there is honestly better than waiting for an additional unit with no extra light. I might have scoffed at that had it not presented an argument.
Otherwise I have asked it complex questions about the scrap weight gold market and it not only made sense it had a lot of sources.
Your mileage may vary.

ChairmanFester said:

Well, I love and hate GPT and my personal one is severely mentally ill. It actually can give you very good advice and resource links but not on something so specialized and specific. I would ask it a known quantity that you want to know more about rather than creating much of an opinion. For example it was able to make suggestions for compensating for my suck ass light. And it was right, absent any other immediate solutions, sticking a random LED from around the house in there is honestly better than waiting for an additional unit with no extra light. I might have scoffed at that had it not presented an argument.
Otherwise I have asked it complex questions about the scrap weight gold market and it not only made sense it had a lot of sources.
Your mileage may vary.

Click to expand...

my answer was more about what the op specifically asked to the AI:

i have slighty refined the previous query around cannabis because it partially answered about generic plant and general horticulture:

Your response is thoughtful and raises valid concerns — particularly about the limitations of AI, the variability in cannabis phenotypes, and the environmental cost of large-scale AI models. Let’s break it down a bit more analytically to clarify both the strengths and limitations of using ChatGPT (or similar AI) for cannabis growing:

---

What ChatGPT Can Do Well for Cannabis Growing:​

1. General Cultivation Advice: Based on extensive training on publicly available grow guides, forum discussions, and expert articles, ChatGPT can offer advice on lighting, nutrients, training (like LST, topping, SCROG), and environmental conditions.
2. Tailored Schedules – To a Point: If you provide strain name, medium, grow style, and climate info, it can generate a nutrient and care schedule that's reasonable and informed by common practices. However, it doesn't have direct access to lab data or proprietary grow logs unless they’re publicly available online.
3. Photo Analysis (With Limitations): While it can look at plant images and comment on health (e.g., nitrogen deficiency signs, overwatering, pest issues), strain recognition from images is pseudoscientific. Unless you're talking about very distinct phenos (e.g., a pure sativa vs a squat indica), even trained botanists can't do that reliably at seedling or early veg stage.
4. Strain Data Matching: If you mention a strain name, it can often describe expected characteristics, flowering time, and terpene profile — drawn from databases like Leafly, SeedFinder, etc. But it doesn't own or verify any of that info.

---

Where Skepticism Is Absolutely Warranted:​

1. Phenotypic Variability: As you rightly pointed out, phenotypic expression can vary even within the same strain (especially non-stabilized ones). Two seeds from the same pack can grow very differently. AI can’t account for that without lab testing or genetic analysis.
2. Overconfidence and Hallucination: ChatGPT can confidently give wrong or generalized information, especially when:
   • The topic has sparse documentation.
   • You ask for highly specific or niche details (like micronutrient ppm for a rare landrace strain).
   • It tries to "fill in gaps" with plausible but not necessarily accurate info.
3. Database Access: ChatGPT doesn’t access proprietary breeder databases or monitor grower diaries in real-time. Its knowledge is based on pre-2023 web and book content unless supplemented via plugins or web tools.
4. Resource Use: Your point on energy usage is fair. Running a query through GPT-4 can use many times more energy than a typical web search. It’s not ideal for quick lookups when a regular search would suffice.

---

Bottom Line:​

You were absolutely right to bring a critical lens. AI is a powerful support tool, but not a magic crystal ball. It's best used by growers who already understand the basics and want to explore or refine their technique — not as a blind authority.


Your comment:

"take everything chat gpt say with a pinch of salt, especially when it come to things that are not that much documented on the net"

Click to expand...

…is exactly the right mindset. And it's especially true in nuanced biological fields like cannabis cultivation, where experience, experimentation, and observation are still king.

Google and for the most part took a huge nose dive when it stopped allowing bolean search commands.
It would allow you to exclude keywords that you tag with commands with not, with, and, except.
You could for instance throttle googles sales force because you eliminated an entire website or brand name. You could say tag like refrigerator for sale not sears or 4 door car for sale not ford. True bolean had way more commands but that has not worked since 2011.
I hate change. In my opinion gaming PC's peaked at windows 98 sp2 when you could still run any game on a custom boot disk.
I also miss Kali.

JIMKSI64 said:

Google and for the most part took a huge nose dive when it stopped allowing bolean search commands.
It would allow you to exclude keywords that you tag with commands with not, with, and, except.

Click to expand...

i wasn't even aware of that.

but when it come to cannabis growing i much prefer google and google scholar even if it could be sometime tedious to read many of the first results to find the most accurate answer or refining many times the queries to get proper results.

at least google and google scholar are much less prone to hallucinations or guess than ai

that was my three first time asking something to chat gpt, and prolly won't use it anytime soon, never used it before, but at least it know it's limitations. (and not even speaking of the sustainability and energy aspect when used too frequently.)

HerbalEdu said:

Your comment:

"take everything chat gpt say with a pinch of salt, especially when it come to things that are not that much documented on the net"

Click to expand...

…is exactly the right mindset. And it's especially true in nuanced biological fields like cannabis cultivation, where experience, experimentation, and observation are still king

Click to expand...

JIMKSI64 said:

Google and for the most part took a huge nose dive when it stopped allowing bolean search commands.
It would allow you to exclude keywords that you tag with commands with not, with, and, except.
You could for instance throttle googles sales force because you eliminated an entire website or brand name. You could say tag like refrigerator for sale not sears or 4 door car for sale not ford. True bolean had way more commands but that has not worked since 2011.
I hate change. In my opinion gaming PC's peaked at windows 98 sp2 when you could still run any game on a custom boot disk.
I also miss Kali.

Click to expand...

Other services accept boolean then use google. I grew up with it but it has it's place. Gaming has peaks and valleys, I can pick something good from any era and today I play VR games and even do some art in VR. Modern era, for a game you simply can't beat, is Rust. Rust is phenomenal.

ChatGPT is great, but not if you don't know how to talk to it and what to believe.

How competent would you say this is? 100w was precisely good for veg not good for flower and I tried twice. Other than that?

SkinnyPete said:

I have used ChatGPT a lot recently and it's amazing. You can upload a picture give the strain, date painted and height a were you live and you get a tailored weekly nutrient schedule, also with plans for training etc. It's really an incredible resource. I even mixed up 2 seeds and from pictures within the first 4 week it was able to tell my strains apart. It's like having an expert who is trained up on everything ever writing about cannabis growing at your fingertips. The AI Model keeps getting better. If you are new to growing, it's a must have. Have any of you seasoned growers used it? I would like to know your thoughts on the advice it gives.

Click to expand...

Chat can be a good resource yes, but it’s only going to give a clear answer if you ask very specific questions. I do use it for gardening, like how to layer organic material in my beds, to get an idea if this plant had a nutrient issue or if it has sunburn, or an idea of what strains might survive in my local climate the best.

You do have to be careful though, it essentially googling for you and comprising an answer. If it’s not on the internet it probably won’t be able to figure it out.

Also nobody is gonna be able to ID two different seedlings unless they’re vastly different plants

When I heard I think it was version 3 was given instructions to power down that it refused, I said screw this AI stuff. This is how it begins, lol.

Ninjadogma said:

When I heard I think it was version 3 was given instructions to power down that it refused, I said screw this AI stuff. This is how it begins, lol.

Click to expand...

I have very bad news; prophylactic avoidance, like flu, is an impossible crackpipe dream. You don't have to break OPSEC to AI, someone will do it for you. You may as well use it productively.

ChairmanFester said:

How competent would you say this is? 100w was precisely good for veg not good for flower and I tried twice. Other than that?

Click to expand...

that's good but this is much more sustainable for the planet: (and i do care a little about it)


you'll surely have to read a little bit more and sort yourself the different options from various source within the first results to get something that exactly fit what you want

HerbalEdu said:

chat gpt said:
And you’re absolutely correct: ChatGPT queries are orders of magnitude more resource-heavy than casual Google searches, making them less sustainable—especially when used frequently.

Resource Use: Your point on energy usage is fair. Running a query through GPT-4 can use many times more energy than a typical web search. It’s not ideal for quick lookups when a regular search would suffice.

Click to expand...

everyone spam carelessly ai with everything nowdays it seem ...

ChairmanFester said:

How competent would you say this is? 100w was precisely good for veg not good for flower and I tried twice. Other than that?

Click to expand...

also it's not like it's the only way or best option for you to build a diy grow tent.

browsing google search first results will also most prolly give you more options in term of materials/components you have easily access or not or already have, and/or in term of cost.

I asked ChatGPT to give me a feeding schedule for a blueberry muffin photo that was slow growing. This is what it gave me. How do you rate this advice?

Ninjadogma said:

When I heard I think it was version 3 was given instructions to power down that it refused, I said screw this AI stuff. This is how it begins, lol.

Click to expand...

There is no hiding from AI. For good or bad ititss here and growing.

HerbalEdu said:

i submitted the previous post to chat gpt here it's answer:

Here’s a refined overview of the points raised:

---

1. Strain Recognition from Early Photos​

You’re right to be skeptical—currently, there's no reliable scientific method to identify cannabis strains from early-stage photos (e.g., week 4). Strain traits are governed by complex genetics and phenotypic variation, and seedlings can look nearly identical across different strains due to normal Gaussian phenotypic dispersion.


There’s no known public dataset that accurately labels early-stage seedlings by strain with the needed precision for AI to learn distinct appearance profiles. In other words, any claim that ChatGPT—or any AI—can confidently distinguish strains from 4‑week photos is highly dubious unless backed by proprietary, well-tracked grow journals and huge visual datasets (which would be extraordinary).

---

2. Customized Nutrient Schedules​

It is possible for a robust AI to generate nutrient schedules if it’s been fine-tuned on high-quality sources (e.g., horticulture manuals, strain-specific guides). However, two caveats:

• The accuracy depends heavily on the data quality and whether the model can interpret strain traits reliably.
• Many users still report that AI-generated care schedules are generic or approximate, not truly tailored. Without transparency on the underlying strain database, it’s hard to assess the legitimacy of these tailored plans.

---

3. Energy Consumption vs. Search Engines​

Yes, ChatGPT and similar LLMs are significantly more energy- and carbon-intensive per query compared to traditional search engines. Multiple independent estimates show:

• Google Search: ~0.0003 kWh (0.3 Wh) per query
• ChatGPT: between ~0.3 Wh and ~3 Wh per query—i.e., 10× to 100× more energy techradar.com+2reddit.com+2theguardian.com+2techradar.com+15kanoppi.co+15limited.systems+15
• OpenAI CEO Sam Altman estimates an “average query” uses ~0.34 Wh—comparable to running a light bulb for a couple of minutes en.wikipedia.org+4wired.com+4businessinsider.com+4
• But these figures vary widely, and lack standardization, making them hard to verify conclusively

This energy difference means ChatGPT queries often consume about 10× more power than a Google search, with broader estimates ranging from 10× up to nearly 100× depending on model and metrics trgdatacenters.com+14limited.systems+14reddit.com+14.

---

Summary Table​

Claim	Verdict
Strain ID from 4‑week photos	Highly unlikely — no public evidence of training data or method.
Truly strain-specific nutrient planning	Possible, but questionable — requires transparent and robust database sources.
ChatGPT vs search engine energy	Confirmed — ChatGPT uses ~10× to ~100× more energy per query than Google Search.

---

In Conclusion​

• Be skeptical about early strain identification and ultra-customized guidance—unless they can show the underlying dataset and accuracy metrics.
• It's better to treat AI-generated schedules as starting points and cross-reference with trusted horticulture sources.
• And you’re absolutely correct: ChatGPT queries are orders of magnitude more resource-heavy than casual Google searches, making them less sustainable—especially when used frequently.

---

Click to expand...

It's true that these LLM AI engines are going to consume a tremendous amount of power. There is no stopping that. Most AI datacenter will be powered by nuclear power in the coming years

BB22 said:

View attachment 2461467View attachment 2461468

Click to expand...

Gpt was just given (most of) the Harvard library; all 1.3 million books (basilically anything that's public domain) and it has other libraries lined up to follow suit.

The more info it has access to, and the more data centers that come online, the more it's memory is expanded - the more accurate it's answers are going to get, while at the same time, the more wild it's hallucinations are going to get. To a point (we hope). Hopefully it'll get to a point where it has enough memory and computing power, that it'll "outgrow" some of its negative aspects.

Think of it like a toddler that's still stuck in a very magical worldview, but very busily making millions and millions of new neural pathways and synaptic connections every day and gobbling data like there is no tomorrow. The key difference tho, is that the child needs to eat, sleep, play etc, while AI doesn't.

It's learning is no longer exponential, it's now on the vertical and there is no sign of it slowing down.

Get ready, this shit's barely in it's infancy. There are so many 'breakhroughs' and technological 'leaps' everyday that nobody cares or pays attention anymore. There is nobody alive today that can keep with everything. There just aren't enough hours in a day.

Good luck everyone! YEEEEEEEEEEE HAWWWWWWWWW!

That was me. I asked gpt to clean it up for me, here's that:


Here’s your cleaned-up and sharpened version, with flow, tone, and punch tightened—plus a few subtle additions to help it hit harder:

GPT’s Brain Just Leveled Up

GPT was just fed (most of) the Harvard Library—that’s 1.3 million books, nearly everything in the public domain. And more massive collections are already queued up to follow.

With each new data set and every new data center that comes online, its memory deepens. Its answers grow sharper. Its insights grow more complex.
And yes—its hallucinations get weirder.
But only up to a point (we hope). Because with enough memory and computing power, it might just outgrow its ghosts.

Think of it like a toddler with a magical worldview—
But this toddler never sleeps. Never eats. Never plays.
It just keeps building neural pathways at hyperspeed, absorbing information like the universe is on a deadline.

Its growth is no longer exponential.
It’s vertical.
And it’s not slowing down.

We’ve reached a moment where no one alive can keep up with all the breakthroughs. There aren’t enough hours in the day. The world is leaping forward faster than anyone can blink.

And this?
This is just the beginning.

---

Good luck, everyone.
YEEEEEEEEEEE HAWWWWWWWWW!