Artificial Intelligence - Poll

[Jay Hemdal Verified Community Expert]

I'm working on an article about informational resources for aquarists. I want to include a section on "Artificial Intelligence" but I need to get more information about how this resource is currently being used.

Thanks!

 

[Timfish]

I have breiflyused chatgp to write a couple things and it has helped my write clrearer stuff. However I am hesitant to use it extensively and won't use it for researching things, largely because of the issue of "halucination" AIs have. It seems to me additional work would be required to verify any research quoted.

 

[The_Paradox]

Spent the last two years trying to get HJF to understand the difference between generative AI, predictive AI, and machine learning so I sore subject with me. At the current state, what is commonly called AI in consumer products is at best only good as a data aggregator. At worst it allows ignorant people to get themselves in trouble. There are much more powerful tools out there that are not available to public such as the UC HPC or IHMCs center.

That out of the way, in the current state the biggest drawback is the data silos that consumer “AI” pulls from. It’s classic garbage in, garbage out. In the aquarium landscape I have only seen predictive AI being used which for obvious reasons is rife with issues. Personally I do not have enough datasets to do anything meaningful, and I don’t trust, or rather do not want to validate, where these “apps” are pulling from.

 

[Julbra]

I have used Chat-GPT to ask questions about reef chemistry and dealing with nuisance algae. It "regurgitated" some good and some not-so-good information, all in a very concise manner.

What I find worrying is that it appears extremely confident about its answers, even when there is as of yet no scientific consensus on the topic. It is certainly impressive though.

 

[Jay Hemdal Verified Community Expert]

I have breiflyused chatgp to write a couple things and it has helped my write clrearer stuff. However I am hesitant to use it extensively and won't use it for researching things, largely because of the issue of "halucination" AIs have. It seems to me additional work would be required to verify any research quoted.

“Error with confidence” - that isn’t limited to “AI” (grin).
I don’t understand why their test came up with five fake book titles…

Jay

 

[Jay Hemdal Verified Community Expert]

Spent the last two years trying to get HJF to understand the difference between generative AI, predictive AI, and machine learning so I sore subject with me. At the current state, what is commonly called AI in consumer products is at best only good as a data aggregator. At worst it allows ignorant people to get themselves in trouble. There are much more powerful tools out there that are not available to public such as the UC HPC or IHMCs center.

That out of the way, in the current state the biggest drawback is the data silos that consumer “AI” pulls from. It’s classic garbage in, garbage out. In the aquarium landscape I have only seen predictive AI being used which for obvious reasons is rife with issues. Personally I do not have enough datasets to do anything meaningful, and I don’t trust, or rather do not want to validate, where these “apps” are pulling from.

I agree, that’s why I usually put “AI” in quotes - IMO, the current iteration is a fancy data aggregator. Garbage in, garbage out is still an issue then.

Jay

 

[DIFish]

I agree with the others. I have used chat gpt for work and some basic aquarium questions. It is very good at bulk processing, writing commonly known things, doing conversions and other not so complex or “it depends” questions. It is still not ready for the prime time as it fails to understand situational nuances, does not have and safety/group checking feature that is a common with forums. If somebody gives you wrong info here there are 10 other opinions, if AI gives you wrong info it will present it as almost facts and there will be no differing opinion. In its current state I think it is a fine starting point but I would not add or change anything in my tank without verifying first. It’s like the new intern at work that is eager to help, will do the grunt work, but sometimes does more harm than good. I am exited to see where it goes though and think companies like Neptune could benefit from machine learning with parameter monitoring in the future.

 

[blecki]

When chatGPT4 first released I checked it out to see how well it did a number of things. I'm a software engineer by trade and - well, it can write basic code. It's pretty good at the sort of things that are easy to google - like the differences in library functions between different languages - but anything complicated and it falls apart.

Trying to ask it about reef chemistry is a no go. It's trained on the internet. And the internet is full of BS. Ask it about fish diseases and you'll get a mishmash of misinformation, like recommending freshwater medications for marine ich.

 

[Court_Appointed_Hypeman]

AI is great for writing kids bedtime stories.

After youe previous demonstration, I sort of deliberately had it give me incorrect answers, and haven't tried it since.

 

[elysics]

It combines the worst aspects of blogs and forums (throwing random or wrong ideas without substance in all directions) with not being able to gauge who is talking and whether they know their stuff. On top of that it can combine incompatible ideas or leave out important details.

 

[Miami Reef]

It sucks at chemistry.

If I add 2g of potassium chloride into my 260 gallon tank, how many ppm will potassium increase?

I asked ChatGPT, and it said this.

“
Thank you for clarifying. Let's calculate the increase in potassium concentration when adding 2 grams of potassium chloride (KCl) to 984 liters:

The molar mass of potassium chloride (KCl) is approximately 74.55 g/mol.

Moles of potassium chloride:2g / 74.55 g/mol = approximately 0.0268 moles of KCl.

Volume of solution: 984 liters.

Calculating the ppm of potassium in the solution with potassium chloride: (0.0268 moles / 984 liters) * 10^6 = approximately 27,317.89 ppm.

Therefore, adding 2 grams of potassium chloride to 984 liters of solution will increase the potassium concentration by approximately 27,317.89 ppm.”


That can’t possibly be correct. Can someone tell me the correct answer?

The math GTP uses is nonsense.

If you add 2 g = 2,000 mg of KCl to 984 L, you have added 2,000 mg/984 L = 2 ppm total KCl.

Since KCl is 52% K, the addition boosts K by .52 x 2 = 1.04 ppm

Potassium chloride to raise potassium

If I add 2g of potassium chloride into my 260 gallon tank, how many ppm will potassium increase? I asked ChatGPT, and it said this. “ Thank you for clarifying. Let's calculate the increase in potassium concentration when adding 2 grams of potassium chloride (KCl) to 984 liters: The molar mass...

www.reef2reef.com

 

[Randy Holmes-Farley Verified Community Expert]

I used it to test its ability on reef chemistry. It got answers ranging from perfect to totally incorrect. Most answers got partial credit.

 

[Jay Hemdal Verified Community Expert]

When chatGPT4 first released I checked it out to see how well it did a number of things. I'm a software engineer by trade and - well, it can write basic code. It's pretty good at the sort of things that are easy to google - like the differences in library functions between different languages - but anything complicated and it falls apart.

Trying to ask it about reef chemistry is a no go. It's trained on the internet. And the internet is full of BS. Ask it about fish diseases and you'll get a mishmash of misinformation, like recommending freshwater medications for marine ich.

The biggest issue I’ve found with it and fish diseases is that it requires an accurate diagnosis to “prime” the question. We know that in many cases here, people misidentify disease - so they ask, “how do I cure velvet?” But their fish actually have Brooklynella. The interactive method used here catches most of these errors.

Jay

 

[Jay Hemdal Verified Community Expert]

I used it to test its ability on reef chemistry. It got answers ranging from perfect to totally incorrect. Most answers got partial credit.

Yes, thanks, I saw your thread, that’s what prompted me to run this poll and update my article.

Jay

 

[ErikVR]

As a software engineer, I never use AI language models like ChatGPT.

This is a huge simplification, but current models are glorified search engines. Scaping content and feeding you that data as a lovely story. The problem is that you don't get any sources and no context whatsoever!

If I ask google how to solve a code challenge in a particular coding language, I can:
- Filter out the obsolete/old results.
- Check the sources of the solutions (are they credible? or is it from a kid without any professional coding experience?)
- Check how many people have upvoted the solution.
- Read the comments to see if the original posts missed something.
- And so on...

ChatGPT does not give you any context and sources. It just writes a nice story for you that you cannot verify unless you do your own research.

AI is in its infancy. It's a toddler that learned a trick. BUT it's growing much faster than a toddler.
Give it a few years and my answer will probably be different.

 

[Randy Holmes-Farley Verified Community Expert]

I would add that there are a growing number of reef2reef users and abusers who are using AI to write posts.

As I have noted elsewhere, there's nothing wrong with a real reef2reef member using AI to help write for them, as long as they agree with what was written,

But In some cases, they are clearly spam being done for reasons I cannot fathom.

One of the hallmarks, IMO, of an AI driven post at reef2reef has sections in it that read like this real one we had recently, where the AI does not realize that the post is actually already doing what was being suggested to be done: ask experienced reefers to help. No real reef2reef member says things like this:

" If you're unsure about the suitability of cinder blocks, it may be advisable to explore alternative reef-safe materials or consult with experienced reef keepers or aquarium professionals who have specific knowledge in this area. They can provide guidance on safe and suitable materials for creating the overhang you desire while ensuring the health and stability of your aquarium ecosystem."

 

[Miami Reef]

I would add that there are a growing number of reef2reef users and abusers who are using AI to write posts.

As I have noted elsewhere, there's nothing wrong with a real reef2reef member using AI to help write for them, as long as they agree with what was written,

But In some cases, they are clearly spam being done for reasons I cannot fathom.

One of the hallmarks, IMO, of an AI driven post at reef2reef has sections in it that read like this real one we had recently, where the AI does not realize that the post is actually already doing what was being suggested to be done: ask experienced reefers to help. No real reef2reef member says things like this:

" If you're unsure about the suitability of cinder blocks, it may be advisable to explore alternative reef-safe materials or consult with experienced reef keepers or aquarium professionals who have specific knowledge in this area. They can provide guidance on safe and suitable materials for creating the overhang you desire while ensuring the health and stability of your aquarium ecosystem."

@diwit is definitely using ChatGPT to answer posts.

 

[The_Paradox]

I would add that there are a growing number of reef2reef users and abusers who are using AI to write posts.

As I have noted elsewhere, there's nothing wrong with a real reef2reef member using AI to help write for them, as long as they agree with what was written,

A growing number of members here who are bots so…

It’s common on all forums now. Compounding these issues is the fact that in this social media driven age whomever has the loudest bullhorn is right. There is a reason why historically professors, instructors, mentors were vetted. Those days are gone. We are seeing the beginning of the movie Idiocracy (watch the trailer if you don’t get the reference).

 

[SPS2020]

The only artificial intelligence used on my tank is my own!

 

[907_Reefer]

Another Software Engineer here, and I have no use for the current "AI" in work or hobbies. As others have noted, what we see today are only fancy algorithms with larger data sets to aggregate from. As Artificial Intelligence does not exist - at best we could label these mechanisms Artificial General Intelligence (AGI), or algorithms targeted towards AGI.