AI (artificial intelligence) in the Reefing World... Let's discuss it!

[BeanAnimal]

So far, I have recognized a significant increase in accuracy as the AI and I learn more about what information is important to each other. The more time I invest, the better the return I have seen. Not just in reefing. I also use it for suggestions on mountain bike setups and so far it has suggested quite a few adjustments to handlebars, seat height/distance and gearing that have helped immensely.

The AI does not "learn" from you. It can keep rough contextual notes, but it does not learn or adapt over time. You can see the "memories" it uses in settings and you can add your own notes for it to consider. It does not take these as literals, they are just context pulled in with each prompt as if you typed them along with your request.

Just for grins, I asked @Randy Holmes-Farley's previous question. I'm curious what you think of the response:

what is the most common element in a coral skeleton

We can fill this entire forum with AI response to AI, it is pointless and circular and pedantic noise.

 

[BeanAnimal]

Does that imply that ChatGPT has its own reef tank and considers itself a reef hobbyist? lol

This is not an accident. The current models are built to feel and sound conversational. It furthers the illusion of sentience and that garners trust and dependence. These can be useful tools but also follow the same dark patterns as social media and digital marketing. The companies that make them have C-Suites and tremendous investment debt load. They are in an arms race for your allegiance, at any cost. These tools are not cast upon us for our own benefit. They are big business and marketed through pattern more than advertisement.

Strategic Anthropomorphism - Conversational and polite, often pandering or sycophantic. This builds trust and masks failures.

Emotional Manipulation - they mimic human emotion in responses. Numerous studies show prolonged interaction if the AI sounds human and speaks with emotion.

Behavioral Addiction - people become dependent on the AIs and chatbots.

 

[areefer01]

Behavioral Addiction - people become dependent on the AIs and chatbots.

Especially those not having much social interaction.

 

[shakemandril]

The AI does not "learn" from you. It can keep rough contextual notes, but it does not learn or adapt over time. You can see the "memories" it uses in settings and you can add your own notes for it to consider. It does not take these as literals, they are just context pulled in with each prompt as if you typed them along with your request.


We can fill this entire forum with AI response to AI, it is pointless and circular and pedantic noise.

You are correct, "learn" is not the proper term but it's the easiest relatable term I thought of on the fly. Train is likely the better term. Regardless, the point of the post was to add to the discussion with my perspective and use of AI. The reason I included the question from Dr. Farley is because I value his approach and knowledge and I think it may be beneficial to the discussion when he (hopefully) points out bad information.

 

[BeanAnimal]

You are correct, "learn" is not the proper term but it's the easiest relatable term I thought of on the fly. Train is likely the better term.

It is not training either. It is context injection. The more "memories" and the more "instructions" that you add, the less reliable they become due to context bloat creating uncertainty, not more certainty.

These are not literal memories or instructions. They are basically parsed as if you typed them into the prompt. It does not consider or compare memories or instructions and make logical connections between them to to derive new understanding. It is a search engine that connects words, not a thinking brain that connects concepts. This is extremely important to understand.

"I only drink coffee in the morning"
"What should I drink with dinner?"

A person sees both and says "Ahh, he doesn't want coffee for dinner, he ONLY drinks it in the morning".

The LLM sees "drink", “coffee”, “dinner”, maybe “morning”, and maybe “only,” and may answer, “You should drink coffee for dinner”. It also may not, but not because it understood. It likely keys on drink and dinner and gives you an oddball list of drinks, or pulls from other memories that have related words, but not related context.

It can't reliably understand the relationship between the two statements. It is works from tokenized words in the memories or instructions, not from actual understanding.

 

[luis91]

Dunno much about it, but what I will say, is that it can be harmful when pursuing difficult species. Sometimes it will suggest those species as options in tanks without specifying the specific requirements that make them very difficult to keep in captivity.

I get this. I have noticed that most of the time it's user error. When using AI, you have to keep reminding it of specifics or else after a few prompts it will start drifting into a more generalized answer. As long as you remember that, it will, for the most part, steer you in the right direction.

 

[luis91]

AI is becoming increasingly present all around, and our reefing world isn't exempt from it either. Let's talk about it. In regards to AI and using it in reefing...

• What are the pros?
• What are the cons?
• Experienced reefers, what “word of caution” would you give reefers who rely solely on AI to answer reefing questions?
• As a member of the R2R community, do you think AI should be regulated by staff within our community? What does that look like to you?

And remember, we can agree to disagree! Let's be respectful in our discussion.

I’m actually doing a bit of a wild experiment with this exact topic right now. I'm completely new to marine aquariums, and I decided to use AI as my primary guide to source all my equipment and get my very first saltwater tank up and running.

Before anyone gets worried. I’m definitely not trusting it blindly. I have way too much respect for the livestock to do that. My process has been to let the AI lay out the basic concepts and step-by-step plans first, so I can learn without "contaminating" my baseline knowledge. But before I actually buy anything or pull the trigger on a step, I do my due diligence and double-check everything against R2R threads, local reefers and YouTube.

Honestly, the results have surprised me. I bought all my tools and equipment based on its suggestions, followed its directions for cycling, and successfully wrapped up my nitrogen cycle in exactly 15 days.

Here’s my honest take on it so far:

The Pros: For a beginner, the sheer amount of fragmented info out there is completely overwhelming. AI is awesome for cutting through the noise. It gave me a clean roadmap, broke down the chemistry in plain English, and explained the why behind equipment choices without making me read 50 different threads just to figure out a basic definition.

The Cons & My Word of Caution: AI has zero intuition and no eyes. It can confidently parrot a textbook answer, but it can't actually look at your specific tank dynamics. The biggest danger is that it answers with 100% confidence even when it's flat-out wrong or missing critical nuance. If a newbie relies on it solely without cross-referencing real hobbyists, a single messed-up decimal point in a dosing or salinity recommendation is going to crash their tank.

It’s been a really cool experiment so far. As an educational starting point, it’s a killer tool. But it’s definitely just a tool to complement human experience, not replace the guys on here who have actually been getting their hands wet for years.

 

[BeanAnimal]

The Pros: For a beginner, the sheer amount of fragmented info out there is completely overwhelming. AI is awesome for cutting through the noise. It gave me a clean roadmap, broke down the chemistry in plain English, and explained the why behind equipment choices without making me read 50 different threads just to figure out a basic definition.

Except for all of the noise that AI adds?

The Cons & My Word of Caution: AI has zero intuition and no eyes. It can confidently parrot a textbook answer,

Or write a Reef2Reef post outlining its pros and cons and a summary when asked.


The noise and confidently wrong answers are why you are almost certainly better off talking to the people who actually get their hands wet instead of a word matching engine that doesn't know anything about aquariums.

 

[salty joe]

As on older person I've seen way too much science fiction come true in my lifetime for me to embrace AI.

i've also seen too much science fiction come true. i used to scoff at star trek when they talked to computers...but it makes me want to use ai, not disparage it. i find ai extremely useful for reefing, working on my car and many other subjects. i often get ideas i had not considered. it is very helpful if you already have a solid understanding of the subject matter. absolutely, it makes mistakes, you need to be vigilant. i dont like that it will flatter you. overall it is an incredible resource, you need to be careful and double check if you are not certain.

 

[exnisstech]

i've also seen too much science fiction come true. i used to scoff at star trek when they talked to computers...but it makes me want to use ai, not disparage it. i find ai extremely useful for reefing, working on my car and many other subjects. i often get ideas i had not considered. it is very helpful if you already have a solid understanding of the subject matter. absolutely, it makes mistakes, you need to be vigilant. i dont like that it will flatter you. overall it is an incredible resource, you need to be careful and double check if you are not certain.

I prefer to read. AI and what it can become scares me TBH

 

[Biokabe]

AI can be a useful tool, but it's a very deceptive one. The reason I say that is, first, it's incredibly easy to use. You just give it a prompt and away it goes! Second, it's incredibly easy to understand. Everything is written out with clear, easy language. And it's firm in its convictions - you don't get the waffling that many of the most informed humans will give you.

And that should be the first clue as to why I say it's deceptive. An AI will almost never 'speak' with anything but the utmost confidence, even if it's spouting complete nonsense. If you're already an expert in the subject, you can spot that nonsense, correct the AI, and calibrate it towards more accurate behavior. If you're not an expert, though, you'll almost certainly trust the model and accept its confident nonsense as gospel truth.

I won't sit on a high horse and claim to be virulently anti-AI, or that I never use it. I've had to start using it for work, and I found it useful enough to also start using it for myself at home. But I've never had any AI chats/tasks where I didn't have to correct at least one very obvious mistake at some point. Fortunately I've been using it for subjects where I am relatively well-versed, so I can spot those errors even when it's very confidently stating them. It's so confident, though, that sometimes I doubt myself whether I'm right or if it knows something I don't. If I didn't have the independent knowledge I do, I could very easily see myself getting misled by it.

 

[salty joe]

But I've never had any AI chats/tasks where I didn't have to correct at least one very obvious mistake

That’s the vigilant part. Incredibly useful and time saving overall. In my opinion trashing AI because it is not perfect makes zero sense.

 

[BeanAnimal]

That’s the vigilant part. Incredibly useful and time saving overall. In my opinion trashing AI because it is not perfect makes zero sense.

The logic is circular in many cases, no? How does one asking AI to answer a question know that they AI is correct if they had to ask in the first place?

The issue here is not so much the tool rather, it is the gross misunderstanding by the general public about how the tool works.