Is ChatGPT right?

areefer01

5000 Club Member
View Badges
Joined
Jun 28, 2021
Messages
5,668
Reaction score
5,884
Location
Ca
Rating - 0%
0   0   0
I’m not calling you out but I am curious: Jake Adams from Reef Builders made a very popular series of videos in which he set up a nano tank in a day. In that series he said one of the keys to its success was that he only fed the two clownfish every few days. That seemed weird. Is this something a lot of people do?



Whelp ask any scuba diver what they see over a reef crest. I don't see fish eating every other day. Heck the first few seconds in one of my dives below the Anthias is eating non stop. Every reef I've dived fish are constantly foraging.

Personally I feed hourly small portions with one or two frozen but hey, we do what we do.

 

BeanAnimal

7500 Club Member
View Badges
Joined
Jul 16, 2009
Messages
9,293
Reaction score
15,594
Rating - 0%
0   0   0
For what it’s worth, ChatGPT is super helpful on this stuff but sometimes if you push back a little bit it’ll say “you’re right to call me out on that, here’s the right answer…”
And if you don’t know what the right answer is, how do you know to ”push back” and what is to say it’s “corrected” response is actually correct?

When the LLM doesn’t know the answer it 100% of the time, makes up an answer. It will never tell you “wow, I don’t know that”.

I use it a LOT for code manipulation. If it doesn’t known a function or syntax, it just makes it up. Literally just invents language and syntax and mixes it in.

If you tell it a tall tell based on partial truths and partial distortions and the ask it to verify, it will verify fabricated information as fact. It is just filling in the blanks. It is biased in they it will almost always tell you what you want to hear, based on how you ask. It fills in blanks.
 

Bohemian Waxwing

2500 Club Member
View Badges
Joined
Apr 22, 2023
Messages
2,919
Reaction score
4,562
Location
Canada
Rating - 0%
0   0   0
Whelp ask any scuba diver what they see over a reef crest. I don't see fish eating every other day. Heck the first few seconds in one of my dives below the Anthias is eating non stop. Every reef I've dived fish are constantly foraging.

Personally I feed hourly small portions with one or two frozen but hey, we do what we do.


I think this is a good clash between great fish husbandry and gaming your tank for laziness and low nutrients.

I’m sure clowns can survive on feedings a few times a week. Jake knew his stuff. But Jake was also a content creator.

My clowns love being fed and I really enjoy feeding them. Why on earth would you get fish just so you can do the bare minimum?

Just go fish less and dose ammonia at that point.
 

KrisReef

Last to get paid.
View Badges
Joined
May 15, 2018
Messages
22,084
Reaction score
39,133
Location
Vatican & Las Vegas Penthouse Quarters
Rating - 100%
1   0   0
And if you don’t know what the right answer is, how do you know to ”push back” and what is to say it’s “corrected” response is actually correct?

When the LLM doesn’t know the answer it 100% of the time, makes up an answer. It will never tell you “wow, I don’t know that”.

I use it a LOT for code manipulation. If it doesn’t known a function or syntax, it just makes it up. Literally just invents language and syntax and mixes it in.

If you tell it a tall tell based on partial truths and partial distortions and the ask it to verify, it will verify fabricated information as fact. It is just filling in the blanks. It is biased in they it will almost always tell you what you want to hear, based on how you ask. It fills in blanks.
I looked up Parable and the AI told me that was stories that Jesus told, like 'The Elephant and Three Blind Men."

I found that amazing that it reported that Jesus spoke in parables, yes it 'Reported' but it doesn't know anything really. The story about the blind folks and the elephant didn't used to be recorded in the Bible, but if AI rewrote that old text I am sure it could find a place to plug it in.

I Love Traveling Wake Up GIF by 60 Second Docs
 

Bohemian Waxwing

2500 Club Member
View Badges
Joined
Apr 22, 2023
Messages
2,919
Reaction score
4,562
Location
Canada
Rating - 0%
0   0   0
I looked up Parable and the AI told me that was stories that Jesus told, like 'The Elephant and Three Blind Men."

I found that amazing that it reported that Jesus spoke in parables, yes it 'Reported' but it doesn't know anything really. The story about the blind folks and the elephant didn't used to be recorded in the Bible, but if AI rewrote that old text I am sure it could find a place to plug it in.

I Love Traveling Wake Up GIF by 60 Second Docs
Ashley Waxman Bakshi Eww GIF by awbmakeup


lol another Kris Klassic
 

Gribbles

Well-Known Member
View Badges
Joined
Mar 10, 2022
Messages
582
Reaction score
456
Location
Cypress
Rating - 100%
1   0   0
The context here being ChatGPT, Gemeni, Copilot, LLAMA and similar, not a task specific LLM trained on a narrow and specific dataset.
We can even include Claude, though it is much better at context depth.

So yes, they spit insane amounts of random BS out. It just sounds good because the LLM is amazing at putting words together.
No, just no. Where to even start with that...

Bottom line is, Chatgpt and others may have some incorrect data that gets pulled from a source like a forum post. It also has tons and tons of legitimate, scientific, peer-reviewed data it is trained on.
 

areefer01

5000 Club Member
View Badges
Joined
Jun 28, 2021
Messages
5,668
Reaction score
5,884
Location
Ca
Rating - 0%
0   0   0
No, just no. Where to even start with that...

Bottom line is, Chatgpt and others may have some incorrect data that gets pulled from a source like a forum post. It also has tons and tons of legitimate, scientific, peer-reviewed data it is trained on.

Verified by who or what entity? What regulations are in place? Standards? What is the source of the data and who owns it and the source location(s)?

I guess we don't like reading books anymore.
 

EastonsReef

Reef Madness!
View Badges
Joined
Apr 8, 2025
Messages
299
Reaction score
488
Location
Tupelo, MS
Rating - 0%
0   0   0
The context here being ChatGPT, Gemeni, Copilot, LLAMA and similar, not a task specific LLM trained on a narrow and specific dataset.
We can even include Claude, though it is much better at context depth.

So yes, they spit insane amounts of random BS out. It just sounds good because the LLM is amazing at putting words together.
No, just no. Where to even start with that...

Bottom line is, Chatgpt and others may have some incorrect data that gets pulled from a source like a forum post. It also has tons and tons of legitimate, scientific, peer-reviewed data it is trained on.
The takeaway for me is that the issues we’re having with AI and the info it provides aren’t new. We have to check the source of any information we get whether that be from a lfs employee, YouTube channel, book, or forum.
 

areefer01

5000 Club Member
View Badges
Joined
Jun 28, 2021
Messages
5,668
Reaction score
5,884
Location
Ca
Rating - 0%
0   0   0
The takeaway for me is that the issues we’re having with AI and the info it provides aren’t new. We have to check the source of any information we get whether that be from a lfs employee, YouTube channel, book, or forum.

The onus always falls back to the hobbyist regardless of hobby.
 

EastonsReef

Reef Madness!
View Badges
Joined
Apr 8, 2025
Messages
299
Reaction score
488
Location
Tupelo, MS
Rating - 0%
0   0   0
The takeaway for me is that the issues we’re having with AI and the info it provides aren’t new. We have to check the source of any information we get whether that be from a lfs employee, YouTube channel, book, or forum.

The onus always falls back to the hobbyist regardless of hobby.
Gotta keep the bs detector going at all times
 

BeanAnimal

7500 Club Member
View Badges
Joined
Jul 16, 2009
Messages
9,293
Reaction score
15,594
Rating - 0%
0   0   0
No, just no. Where to even start with that...
Bottom line is, Chatgpt and others may have some incorrect data that gets pulled from a source like a forum post. It also has tons and tons of legitimate, scientific, peer-reviewed data it is trained on.

You’re confidently wrong. The issue isn’t that LLMs only use garbage. It’s that they can’t tell the difference.

The vetted and reviewed training data is mixed with outdated, unvetted, or outright false material and in the current state of a race to user adoption, garbage data is pouring in by the petabyte.

When prompted broadly, they often produce confident-sounding nonsense. They don’t know facts, cite sources, or verify claims. The LLM literally just predicts likely word sequences. Your faith in their accuracy shows you don’t understand how they actually work.
 

Gribbles

Well-Known Member
View Badges
Joined
Mar 10, 2022
Messages
582
Reaction score
456
Location
Cypress
Rating - 100%
1   0   0
Verified by who or what entity? What regulations are in place? Standards? What is the source of the data and who owns it and the source location(s)?

I guess we don't like reading books anymore
Reading books and also using ChatGPT can both be true...

You also can (and should) ask them to source their information. Much quicker and thorough than researching yourself. It's not a magic wand but ppl keep expecting it to be and then lolcopter when it doesn't meet their expectations.
 

TankYouVeryMuch

Active Member
View Badges
Joined
May 20, 2025
Messages
344
Reaction score
324
Location
Fayetteville, AR
Rating - 0%
0   0   0
For what it’s worth, ChatGPT is super helpful on this stuff but sometimes if you push back a little bit it’ll say “you’re right to call me out on that, here’s the right answer…”
And if you don’t know what the right answer is, how do you know to ”push back” and what is to say it’s “corrected” response is actually correct?

When the LLM doesn’t know the answer it 100% of the time, makes up an answer. It will never tell you “wow, I don’t know that”.

I use it a LOT for code manipulation. If it doesn’t known a function or syntax, it just makes it up. Literally just invents language and syntax and mixes it in.

If you tell it a tall tell based on partial truths and partial distortions and the ask it to verify, it will verify fabricated information as fact. It is just filling in the blanks. It is biased in they it will almost always tell you what you want to hear, based on how you ask. It fills in blanks.
> When the LLM doesn’t know the answer it 100% of the time, makes up an answer. It will never tell you “wow, I don’t know that”.

This was more true in the old days (which with AI is about six months ago). I told it specifically to not do that, and it decreased the behavior.

As for knowing when to call it out, it’s the same as with any other opinion—if it doesn’t jive with what I’ve read or heard elsewhere. When the owner of the local LFS told me he’s going to go back to using tap water, I didn’t automatically follow his lead despite the fact he’s got way more years of experience and a shop full of healthy reef tanks. I asked him why, listened, and decided I was happy to have an explorer handy who would blaze a trail for me. Maybe he’ll have stunning success. I’d be surprised, but life turned out to be full of surprises when I looked for them.
 

BeanAnimal

7500 Club Member
View Badges
Joined
Jul 16, 2009
Messages
9,293
Reaction score
15,594
Rating - 0%
0   0   0
This was more true in the old days (which with AI is about six months ago). I told it specifically to not do that, and it decreased the behavior.
No, it just agreed with you. It can’t keep that kind of context, even if you write it in a directive. Current versions are worse because they are removing guardrails to gain appeal and market share.

Again, it is not thinking or developing a conversation with you. Its context is only a few prompts deep and maybe a few general tied links to “memories”. It is not smart, or thinking or anticipating. It will lie to you and promise you it will do better if you catch it — but it has no means to keep that promise or even know it lied. It is literally just using words that fit being told
It erred.
 

TankYouVeryMuch

Active Member
View Badges
Joined
May 20, 2025
Messages
344
Reaction score
324
Location
Fayetteville, AR
Rating - 0%
0   0   0
Here’s what I’ve learned by scouring this forum, YouTube, scientific presentations, etc: The more of an expert someone is in a topic, the more their views tend to conflict with at least some aspects of the “popular opinion.”

Jake Adams:
- Claimed refugiums aren’t effective.
- Suggested in most cases lower lights and lower flow are better for corals.
- Fed fish infrequently to lower bioload.
- Didn’t like ATOs
- Didn’t like sandbeds.
- Emphasized nutrient stability over ultra-low nutrient systems (ULNS)

Ehsan Dashti (Titan Method):
- Argues that water changes are unnecessary if you dose the right trace elements and keep parameters in check.
- Relies on lab-based dosing over the more traditional “do water changes and call it good” approach.

Julian Sprung:
- Thinks refugiums are more of what’s necessary.
- Prefers algae scrubbers and refugia over chemical phosphate removers.
- Emphasizes bacterial and microfaunal diversity over chasing numbers.
- Believes handling palythoa with his bare hands is fine (update: not anymore).

Paul Baldassano (“Paul B”):
- Strongly believes in feeding heavily and encouraging natural filtration through biodiversity.
- Thinks quarantine is unnecessary if fish are properly fed and kept in a healthy biome.
- Claims that most modern reefers are too sterile and gear-obsessed and that they miss the whole point of having an “ecosystem.”

To name a few.

The point is that there is more than one way to come at this, and also just as many ways to fail. Even the experts don’t always agree on which is which.

A reef aquarium is a complex problem. It involves art, lighting design, mechanical and electrical engineering, chemistry, biology, to some degree psychology and relationship management, and definitely finance skills. Not to mention patience, tenacity, diligence, and, if you want to be a respected leader in it you’ve gotta be able to be personable, too (a lot of experts excel at relating to their tank but are like fish out of water relating to people).
 

RelaxingWithTheReef

Active Member
View Badges
Joined
Jan 10, 2022
Messages
168
Reaction score
215
Location
US
Rating - 0%
0   0   0
AI may help by pointing you in the right direction if you are very careful, but most of the responses are really off the rails!

I like to click and read the Google AI references provided to understand the context of the response, and gauge the credibility. In this respect, I use it as a glorified search engine.

I cringe to think about the day when AI gets more in the mainstream, and it’s trained on dirty data from previous inaccurate AI output, forming a feedback loop. At some point the worlds data may get so dirty that we may have absolutely no idea just what to believe anymore.
 

Gribbles

Well-Known Member
View Badges
Joined
Mar 10, 2022
Messages
582
Reaction score
456
Location
Cypress
Rating - 100%
1   0   0
I cringe to think about the day when AI gets more in the mainstream, and it’s trained on dirty data from previous inaccurate AI output, forming a feedback loop.
This just isn't going to happen with the major companies. If they did that, the product would be useless.

If you were in charge, would you let the AI train on crummy data? Of course not. They try to mitigate it as much as possible as well.
 

TankYouVeryMuch

Active Member
View Badges
Joined
May 20, 2025
Messages
344
Reaction score
324
Location
Fayetteville, AR
Rating - 0%
0   0   0
To everyone saying ChatGPT is useless, I have to say I disagree. This is its comprehensive answer to “If someone’s reef tank has a breakout of hair algae, what’s the best way to tackle it?” It seems pretty thorough and accurate to me (it specifically mentioned emerald crabs and tailspot blennies because I had previously been asking it about them).

ChatGPT response:

A breakout of green hair algae in a reef tank is a common and frustrating problem, but it’s fixable with a multi-pronged, patient approach. Here’s a breakdown of the most effective strategies:



1. Manual Removal (First Step)
• Scrub rocks gently with a toothbrush or small brush during water changes.
• Siphon out loose algae to avoid spreading spores.
• Remove filter floss or mechanical filtration right after to trap fragments before they reattach.



2. Nutrient Control

Hair algae thrives on excess nitrate and phosphate, but don’t strip nutrients to zero, or you risk dinos.
• Target ranges:
• Nitrate: ~2–10 ppm
• Phosphate: ~0.03–0.1 ppm
• Use GFO or PhosGuard cautiously if phosphate is very high.
• Reduce feeding, especially of nutrient-rich frozen foods.
• Rinse frozen food to remove binders and juices.
• Consider a refugium with chaeto or another macroalgae to outcompete hair algae.



3. Biological Control (Cleanup Crew)
• Emerald crabs: Can help, especially small hungry ones.
• Turbo snails: Bulldoze through algae, especially on rocks and glass.
• Tuxedo urchins: Excellent grazers (secure your frags—these guys will wear them).
• Tailspot blenny or lawnmower blenny: May nibble hair algae, though success varies.



4. Lighting Adjustments
• Reduce photoperiod temporarily (e.g., 6–8 hours/day).
• Check light spectrum: Intense white or red channels can fuel algae growth.
• Use blue-heavy spectrum to favor corals over nuisance algae.
• Clean your light lenses and rock surfaces so algae isn’t getting extra nutrients from detritus.



5. Address the Root Causes
• Overfeeding, infrequent water changes, or dead spots in flow all contribute.
• Test your RODI water — TDS creep can introduce nutrients.
• Remove detritus traps like dirty filter socks, clogged media, or dead snails.



6. Long-Term Stability
• Stay consistent with water changes.
• Don’t panic-dose chemicals; stability beats sudden corrections.
• Track nutrient trends weekly—algae often blooms when NO3 bottoms out and PO4 stays elevated.



What Not to Do:
• Don’t chase “zero nutrients.” This can backfire and lead to dinoflagellates.
• Don’t introduce herbivores you can’t support or remove later (e.g., giant urchins in a nano tank).
• Don’t use algaecides unless everything else has failed—and even then, use caution.

 

Gribbles

Well-Known Member
View Badges
Joined
Mar 10, 2022
Messages
582
Reaction score
456
Location
Cypress
Rating - 100%
1   0   0
You’re confidently wrong. The issue isn’t that LLMs only use garbage. It’s that they can’t tell the difference.

The vetted and reviewed training data is mixed with outdated, unvetted, or outright false material and in the current state of a race to user adoption, garbage data is pouring in by the petabyte.

When prompted broadly, they often produce confident-sounding nonsense. They don’t know facts, cite sources, or verify claims. The LLM literally just predicts likely word sequences. Your faith in their accuracy shows you don’t understand how they actually work.
Learn to read, Bean. Multiple times I've said you must vet their sources. You are ridiculous to discuss with
 

TOP 10 Trending Threads

WHAT AMOUNT OF LIVE ROCK AND SAND SHOULD BE PRIORITIZED FOR OPTIMAL BIODIVERSITY/FILTRATION?

  • 100% live rock + bagged sand

    Votes: 38 27.3%
  • 100% dry rock + 100% live sand

    Votes: 47 33.8%
  • 50/50 live/dry rock, 50/50 live/bagged sand

    Votes: 30 21.6%
  • 75% live rock, 25% live sand

    Votes: 14 10.1%
  • 25% live rock, 75% live sand

    Votes: 10 7.2%
Back
Top