Fruit growing knowledge, this forum, and AI

About This Forum
1

Hi All,

I am curious if there has been any thoughts from the owners/admins of this forum to feed the whole site archives into an AI engine that can organize all the amazing information into a query friendly interface. I have been playing with chatgpt, gemini etc, and the fruit knowledge of these engines is useful and concise, but not as deep and wide as the collective wisdom of this community. It would be amazing to merge the two.

Some ideas of what AI could help do with all the info contained in this forum:

  • Pull info from disparate threads about a particular cultivar/problem/topic and present condensed information, summarize points of agreement and disagreements, expose knowledge gaps etc
  • answer ‘How to’ questions
  • Process all the image files on site and allow AI to diagnose pests/diseases etc by uploading photos
  • cross reference your zip code to historic and current weather data, pest activity, and cultivar info to provide ideal candidates to plant and spray timing/schedule

What do you all think? Is this feasible?

4 Likes
2

The misinformation on this site would also be included.

16 Likes
3

I’ll stop using this forum as a source of information. I’m not a super active poster but I read a lot.

11 Likes
4

The best way to leverage a LLM with our growingfruit context would be to have a locally ran open source model. That way it is owned and maintained by the owner/users of the site. By feeding the entire website to open ai, anthropic, google,etc…we would loose control of our data and how it is used. Doing that would also introduce outside data into the model and its functionality. If anything I think using the search function and asking questions to members is the whole point of a forum. Even if a llm has access to the entire growingfruit forum, it could still fail to understand nuance of what a person wants, needs, or might mean to say.

10 Likes
5

It’s a safe bet that the AI summaries displayed by search engines are already using the publicly available content on this forum as a reference. Also, what Richard said.

11 Likes
6

It does, I have been fed my own post before by google AI. It doesn’t even qualify that its an opinion or that its not from an expert, it just spews it at you like its the truth. Sometimes it’ll say its from the website, but it doesn’t always.

14 Likes
7

Honestly the AI models pull stuff from here to spit out their info already so what’s the point?

As pointed out it would include mis information and various hypotheses and questions posed in unusual ways as if it were factual information (same problems the ai searches have when pulling from the eider web) and degrade the actual useful information here.

And honestly while I know a lot of folks have bought into the AI hype, as a cranky middle aged skeptic that sees a flaw in thr negative feedback loop of inaccuracy built into the AI designs, I find I am already having a harder time finding good information here with all the AI slop that people post in the past year or two (in good faith I might add -well meaning folks repeating information they assume the AI that cobbled it together was accurate with).

I can only speak for me, but my main reason for coming here as often as I do (pretty much a daily reader, only an occasional poster) is to get information from real people’s real (albeit anectdotal) experiences. If I wanted to trust AI to give me good enough of information, I would just use one of the big name ones to tell me how smart I am for asking the questions I have to start with and cut out the growing fruit “middle man".

As it is the main reason I use this site is to be able to get information I can use my own experience and analytical skills to decipher and evaluate instead.

But that is only my 0.02cents. Kind of true when I said I’m just a cranky middle-aged AI skeptic I guess, so take that as you will.

(For the record while I oppose the idea of more AI on the forum, the notion of discussing helpful features and improving this community in general are laudable)

21 Likes
8

Just recall that all AI is not created equally. The best AI is definitely subscription service. The open AI is many generations less capable.

2 Likes
9

I am not sure I fully understand the value add - I am pretty sure this forum is already fed into all the usual suspects and you could “search” the forum through your AI of choice. I’d keep the scope of the forum true to its original intent - a place for humans to talk. If I anything I would love to get some "note: this post was generated by AI” for posts that were copy-pasted from AI.

9 Likes
10

not necessarily, it will do what you tell it to do when processing the data. I think its a great idea, as there are many threads that are essentially the same and could be combined. Of course we would want to keep the original stuff intact, but the ability to see all knowledge on a topic in one place could be really helpful. It could save a lot of time. as stated there is likely some bad info also that would be incorporated into the output results, but as long as the information is properly cited I wouldnt worry too much about it.

3 Likes
11

This is a feature, not a bug! Something like this would be helpful.

“14 members had success treating the problem by adding more potassium to soil, while one user reported improvement by surrounding the tree with magnets and rubbing coconut oil on the trunk in a counterclockwise fashion”

6 Likes
12

[quote=“TheDerek, post:10, topic:78502”]

the ability to see all knowledge on a topic in one place could be really helpful

[/quote]

Yeah I would agree that is useful, but there is already a search feature which accomplishes this pretty well from my experience.

I do not see where ai would be of benefit in that regard.

5 Likes
13

No. It’s a bug. Plenty of very low quality data presented as if it’s expert data on these forums that has to be sifted through. The number of people who say whatever is irrelevant. The people who have grown extremely large amounts of varieties over many years are much more reliable than a chorus of people who have grown five varieties

7 Likes
14

I’ve seen AI unintentionally slightly alter the meaning of something even when based on a single source. The fact that it rewords things without understanding nuance is a major factor.

8 Likes
15

Agree! But this presupposes that readers will research each poster to determine how much of an expert they are, and unless this is your life, is not practical. Why not let they AI determine which users are more credible based on posting history, badges, consensus etc and provide its assessment in a transparent manner?

The value add is not to create or extrapolate knowledge, but to present it in a more interactive, useful manner.

The goal is not to eliminate the forums and threads, that’s the lifeblood of this community. It is just to organize and collate. Things won’t be perfect, but more usable than current search requiring reading dozens of threads that veer of track..

2 Likes
16

Not sure how a few lines of computer code can be relied on to sift out what information is expert vs what expert information is relevant to the reader’s own growing situation (a disease resistent stone fruit in the PNW may not be as grower friendly in the humid mid atlantic as an example) vs what information is just posted in the voice of expertise vs what information simply is presented in a way that many people agree with without expert first hand experience/knowlwdge of their own to validate thier beliefs…

But to address your point about having to slog through some posts that veer off topic within threads in order to do the research one’s self; would a different solution, such as adding a button that users could use to flag off-topic posts combined with an option in the search that could be toggled on to collapse such often flagged posts for the searcher be useful to reduce the time needed to parse the information (at the slight risk of missing some potentially useful but erroneously flagged posts)?

That (or similar) feature might provide a way to reduce some extraneous reading for those that want without risking adding additional AI slop into the the mix here?

2 Likes
17

we could probably have the contributors rated based upon experience and helpfulness as judged by the responses to their posts in threads also.

18

Good feedback, and I strongly endorse your proposal to allow toggling to filter of off-topic replies. These replies add to the warmth and community feel of this forum, but dilute the nuggets of knowledge that we search out, so both would be useful.

In regards to AI capability… These “few lines of code” LLMs are precisely very good at correlating and teasing out patterns.. Each poster would have a profie… location, granular details of what they grow and what they know and reputation. It will not weigh in advice from Alaska based on quieries about Florida.

I don’t know what you mean by adding slop to the mix? The AI tool wouldn’t be posting in forum, it would be an individualized search and query tool. People can already cut and paste all the slop they want.

1 Like
19

If it can do that accurately it’s far beyond current AI. Once we have AI that powerful it can do a heck of a lot more than that

1 Like
20

[quote=“bigiggye, post:18, topic:78502”]

In regards to AI capability… These “few lines of code” LLMs are precisely very good at correlating and teasing out patterns.. Each poster would have a profie… location, granular details of what they grow and what they know and reputation. It will not weigh in advice from Alaska based on quieries about Florida.

I don’t know what you mean by adding slop to the mix? The AI tool wouldn’t be posting in forum, it would be an individualized search and query tool. People can already cut and paste all the slop they want.

[/quote]

What I meant by that last post is that the information that AI puts together doesn’t exist in a vacuum, if I search for something using AI and the result is either misunderstood or just wrong, but I (Assuming that the AI answer is correct, and an AI based on a forum that I consider to include valuable information, I would be naturally more inclined to accept as accurate) then I post a conclusion/answer/etc based eitger in part or full on the faulty information, then it degrades the quality of information available here by diluting the actual good information.

That error then gets compounded by the inherent flaw in AI where it adds to a feedback loop where there is additional bad data being taken in by the AI, and further increasing the likelihood of faulty information being presented by the AI in future searches, and thus continue the cycle. (Basically the issue is the old adage “junk in, junk out" for computing, but in the case of AI when it spits junk out, inevitably the junk gets re-ingested during the next “in" cycle).

I take your point that it could be theoretically possible for the ai to be tailored to take into account various details like that, but the challenge I see with that is getting the necessary (and correct) information in the first place: some users may not be comfortable or feel safe sharing all of the needed information. Some may inadvertently put in incorrect information, and some may even intentionally put in false information (not saying I understand the use of the latter on a forum like this, but it is the internet and some fraction of people will do that (even if not trolling, then to simply make themselves seem more important or to simply avoid giving correct answers if they fall in the camp of folks uncomfortable with doing that).

In addition, users would need to input a lot of data for that to function like I (think?) I understand you suggesting. Some growers here grew things for decades in one environment, then moved and grew things for another decade in a totally different one. There would need to be historical information on each profile with what was grown, what years it was grown, where it was grown, how it was grown (was it a backyard 2 tree setup. A commercial orchard. Did they spray regularly? never? Somewhere in between?)

Without that I would expect ai would struggle to accurate sift through the data.

If I posted information I thought I knew (but was wrong about) regarding a fruit tree cultivar “example apple” in 2020. But then I began to grow it in 2022 and by 2027 I had substantial experience to answer some questions about it accurately, the ai search would have to be able to detect and filter out the less accurate info while including the more accurate info and thus would need a lot more than a crowdsourced general trustworthiness rating to go by to ensure accuracy.

Anyway, I don’t mean to sound argumentative if I do. There are valid points on both sides, and the thing I think everyone in this thread has in common with eachother is wanting the forum to be as good as it can be, we just have differing opinions on how to approach that shared goal. I hope that is the main takeaway from this discussion by everyone. :slight_smile:

2 Likes