Keras
International Hazard
Posts: 1046
Registered: 20-8-2018
Location: (48, 2)
Member Is Offline
|
|
Don’t believe chatGPT when it comes to directing groups :)
Lol,
I’m still researching into nitrotoluene → nitrobenzaldehyde or nitro styrene → nitrobenzaldehyde, so I figured out I'd ask ChatGPT for fun.
Here is what it told me:
‘The synthesis of 2-nitrobenzaldehyde involves introducing a nitro group (-NO₂) at the 2-position (ortho to the aldehyde group) on a benzene ring
that already has a formyl group (-CHO) attached. The aldehyde group typically serves as a directing group in electrophilic aromatic substitution
reactions, favoring substitution at the positions ortho and para to itself.
Here’s a general synthetic approach for 2-nitrobenzaldehyde:
1. Nitration of Benzaldehyde
A common method to synthesize 2-nitrobenzaldehyde is by performing a nitration reaction on benzaldehyde (C₆H₅CHO) using a nitrating mixture. In
this case, the aldehyde group will direct the incoming nitro group to the ortho position, resulting in 2-nitrobenzaldehyde.’
But then, if I ask about 3-nitrobenzaldehyde:
‘To synthesize 3-nitrobenzaldehyde (i.e., the nitro group at the 3-position, meta to the aldehyde group), we need to exploit the ability of the
aldehyde group (-CHO) to direct the nitro group to the meta position in electrophilic aromatic substitution reactions.
Synthesis of 3-Nitrobenzaldehyde:
Nitration of Benzaldehyde: The primary method to synthesize 3-nitrobenzaldehyde is by performing nitration on benzaldehyde. In this case, the aldehyde
group is an electron-withdrawing group, which typically directs the incoming electrophile (in this case, the nitronium ion, NO₂⁺) to the meta
position.’
Lol.
Screen shots attached, just in case you'd doubt it.
 
|
|
|
Texium
Administrator
Posts: 4685
Registered: 11-1-2014
Location: Salt Lake City
Member Is Offline
Mood: Preparing to defend myself (academically)
|
|
I don’t doubt it at all. The thread title could have just been shortened to “Don’t believe ChatGPT.”
|
|
|
Texium
|
Thread Moved
8-4-2025 at 08:06 |
Radiums Lab
Hazard to Self
Posts: 60
Registered: 18-3-2025
Location: India
Member Is Offline
Mood: Experiencing the elegance of science.
|
|
Yea I too agree with you Texium , I was searching methods to store sodium it suggested to place in an airtight container filled with water.
Water is dangerous if you don't know how to handle it, elemental fluorine (F₂) on the other hand is pretty tame if you know what you are doing.
|
|
|
Keras
International Hazard
Posts: 1046
Registered: 20-8-2018
Location: (48, 2)
Member Is Offline
|
|
Quote: Originally posted by Radiums Lab  | | Yea I too agree with you Texium , I was searching methods to store sodium it suggested to place in an airtight container filled with water.
|
At least that seems to have been corrected…
|
|
|
davidfetter
Harmless
Posts: 22
Registered: 8-9-2021
Member Is Offline
|
|
There's really nothing you should be doing with it, even for "fun."
Why not? The tremendous resources each query into the thing waste should suffice even if you find masplaining-as-a-service somehow entertaining.
|
|
|
bnull
National Hazard
Posts: 634
Registered: 15-1-2024
Location: The country that made an art of fence-sitting
Member Is Offline
Mood: I'm back, sort of.
|
|
I love these "bullshittium recipes" that ChatGPT generates. Some of them border on plausibility, a few almost fooled me. Ask "are you sure?" after
ChatGPT revises each iteration of the procedure. It will be an endless apologizing session.
|
|
|
Keras
International Hazard
Posts: 1046
Registered: 20-8-2018
Location: (48, 2)
Member Is Offline
|
|
| Quote: Originally posted by davidfetter | There's really nothing you should be doing with it, even for "fun."
Why not? The tremendous resources each query into the thing waste should suffice even if you find masplaining-as-a-service somehow entertaining.
|
Don’t worry, last time I used it was like two years ago, at the very beginning. IMHO, all the fuss about generative AI is just a fad.
|
|
|
Tsjerk
International Hazard
Posts: 3038
Registered: 20-4-2005
Location: Netherlands
Member Is Offline
Mood: Mood
|
|
Aren't there any science AI's yet? Math AI's, yes, but physics/chemistry? The reason ChatGPT sucks at science is because it wasn't designed to be. I
use GitHub CoPilot as a plugin in my IDE everyday and it's really helpfull. But then again, GitHub had a lot of easy to train on data to begin with .
Lets not forget protein crystallography was made redundant as a field by AlphaFold
[Edited on 8-4-2025 by Tsjerk]
|
|
|
BromicAcid
International Hazard
Posts: 3276
Registered: 13-7-2003
Location: Wisconsin
Member Is Offline
Mood: Rock n' Roll
|
|
There are lots of issues with making a chemistry AI. Biggest is that most of the information is behind paywalls. Even my company gets flagged if
someone browses too many articles too quickly - these companies evolved to stop skimming long before skimming was the thing it is today.
Another issue is that there is a lot of bad data. Like, loads of bad data. You might find an article from the 1890's saying that lead tetraacetate
can oxidize a certain group but then a paper from the 50's where they are unable to replicate it and then another in 1990 where someone decides that
it does work but needs to be run in the presence of air and then another paper a few years later when they decide that it's just the Pb4+ acting
catalytically and finally a paper saying that it was actually a copper impurity the whole time. Oh, and yields ran the gamut from 0% - 45%. What
does a large language model learn from that?
We've been warned at work not to use AI for anything and we're periodically given examples of where it's screwed up in a life-threatening way
involving chemistry (procedure generated for quenching 10M t-butyllithium involved 'adding quickly to a large amount of water'). It's just a lot
harder because things that are similar in one way don't behave similarly in another.
I've used the Synthia program for retrosynthesis reactions and it works reasonably well. It was specifically trained on curated good data. But even
then it doesn't go into details. It points out specific papers for synthesis steps and it's up to the chemist to translate those to the target
molecule.
|
|
|
Radiums Lab
Hazard to Self
Posts: 60
Registered: 18-3-2025
Location: India
Member Is Offline
Mood: Experiencing the elegance of science.
|
|
Guys please have a look at this image :
Water is dangerous if you don't know how to handle it, elemental fluorine (F₂) on the other hand is pretty tame if you know what you are doing.
|
|
|
BromicAcid
International Hazard
Posts: 3276
Registered: 13-7-2003
Location: Wisconsin
Member Is Offline
Mood: Rock n' Roll
|
|
Seems like there should just be a chemistry youtube channel where they follow chat GPT advice for better or worse.
|
|
|
Keras
International Hazard
Posts: 1046
Registered: 20-8-2018
Location: (48, 2)
Member Is Offline
|
|
While this is of course true, a lot of old chemistry books up to years ~1950 are now public domain. While the info you find inside is not exactly the
most modern, it is still pretty accurate, and could reasonably be used to train an AI engine. With what results, I have no clue though.
|
|
|
Precipitates
Hazard to Others
Posts: 163
Registered: 4-12-2023
Location: SE Asia
Member Is Offline
Mood: Acid hungry
|
|
Most people would realise it's contradictory though - both in water and away from moisture. Mystery non-wet water.
But...I am surprised we are not hearing more "chatGPT told me to do this" injuries or deaths - perhaps because it is more likely to contain
conflicting, as appose to totally incorrect information. And, as much as it doesn't seem like it at times, most of us still do have some common sense.
If there isn't one already, I'm sure it's only a matter of time!
|
|
|
ManyInterests
National Hazard
Posts: 970
Registered: 19-5-2019
Member Is Offline
|
|
DO NOT TRUST CHATGPT!
In fact, do not trust ANY AI with chemistry questions. They will get ultra poor results at absolute best or get you killed at worst. As you know, I'm
an energetic guy, and I decided to ask it questions on stuff I am experienced with, such as picric acid, and the stuff it gave was worse than hogwash.
The stuff about RDX is even worse. I once asked it to tell me the oxygen balance of a PETN:RDX:ETN mixture just for fun, and it told me 'it would have
poor performance due to the highly oxygen poor nature of ETN'.
I had to correct the AI many times. Notably one thing I did ask it about was how picric acid shells were loaded in the past. I was curious since I
heard it was not necessarily used as a powder (it was mixed with other stuff to desensitize it). I did hear about a rarely used Japanese formula that
used 75% picric and 25% TNT and then melt-cast (the TNT lowered the overall melting point). It told me that picric was melt-cast alone and 'heated to
150 or 200C before casting'.
I need to tell you something... DO NOT DO THAT! The flash point of picric acid is that much, so unless you want to deal with a positively explosive
flame and die, just don't...
The best thing you can hope for with chatGPT is to ask it for sources. For example I am still gathering information on the best ways to make firearms
primers and powders (for pistol calibers) and the information was actually half-OK, but I did ask it for the sources and I did look those sources up,
and they did check out.
That is the best use for them. When it comes to chatGPT. Always always always ask for the sources. ChatGPT found me sources for stuff I wanted that I
wouldn't have discovered otherwise.
|
|
|
bnull
National Hazard
Posts: 634
Registered: 15-1-2024
Location: The country that made an art of fence-sitting
Member Is Offline
Mood: I'm back, sort of.
|
|
ChatGPT gave me a couple of nonexistent references some time ago. Made up authors and titles. I suppose it has been fixed by now and that it happened
because the subject was not well researched.
|
|
|
Texium
Administrator
Posts: 4685
Registered: 11-1-2014
Location: Salt Lake City
Member Is Offline
Mood: Preparing to defend myself (academically)
|
|
I think it’s alright to use for finding references so long as you actually
A. Confirm that the reference actually exists
B. Read the reference and make sure that it contains the information that the AI claims it does.
Even if it offers real references, it could take their claims out of context, or completely misrepresent them. So it is always important that your
research process includes reading the primary references before you cite them or try any experiments based on what you think they say.
[Edited on 4-21-2025 by Texium]
|
|
|
Precipitates
Hazard to Others
Posts: 163
Registered: 4-12-2023
Location: SE Asia
Member Is Offline
Mood: Acid hungry
|
|
This is the issue I tend to have. I can never find the information claimed by AI. Especially for relatively obscure subjects.
|
|
|
Lionel Spanner
Hazard to Others
Posts: 180
Registered: 14-12-2021
Location: near Barnsley, UK
Member Is Offline
|
|
| Quote: Originally posted by ManyInterests | DO NOT TRUST CHATGPT!
In fact, do not trust ANY AI with chemistry questions. They will get ultra poor results at absolute best or get you killed at worst. As you know, I'm
an energetic guy, and I decided to ask it questions on stuff I am experienced with, such as picric acid, and the stuff it gave was worse than hogwash.
The stuff about RDX is even worse. I once asked it to tell me the oxygen balance of a PETN:RDX:ETN mixture just for fun, and it told me 'it would have
poor performance due to the highly oxygen poor nature of ETN'. |
Sounds like it's been reading too much Jared Ledgard.
|
|
|
ManyInterests
National Hazard
Posts: 970
Registered: 19-5-2019
Member Is Offline
|
|
| Quote: Originally posted by Precipitates |
This is the issue I tend to have. I can never find the information claimed by AI. Especially for relatively obscure subjects. |
I have been able to find some, but other sources seem to be more restricted or just don't exist.
I did ask it other questions about picric acid, as I did notice some things when I added an excess of nitrates, but AI has given so many wrong answers
that I am not very trustful of what it is saying.
|
|
|
Radiums Lab
Hazard to Self
Posts: 60
Registered: 18-3-2025
Location: India
Member Is Offline
Mood: Experiencing the elegance of science.
|
|
As someone pointed out, we can use AI only to dig out papers or documents from the internet.
Water is dangerous if you don't know how to handle it, elemental fluorine (F₂) on the other hand is pretty tame if you know what you are doing.
|
|
|
Sir_Gawain
Hazard to Others
Posts: 497
Registered: 12-10-2022
Location: [REDACTED]
Member Is Offline
Mood: Still in 2022
|
|
It also can’t balance equations. Even if it could reliably pull that off, I doubt it would be much help to chemists. Chemistry requires a special
kind of experience you only get from actually doing it. You can’t just have the theoretical knowledge. I’m reminded of this every time I talk to
younger people starting out, they don’t have that feel for what would work or not, even if they have a good grasp on the theoretical. Chemistry is
messy and finicky, requiring so many variables I doubt AI will master it for a while yet.
“Alchemy is trying to turn things yellow; chemistry is trying to avoid things turning yellow.” -Tom deP.
|
|
|
Texium
Administrator
Posts: 4685
Registered: 11-1-2014
Location: Salt Lake City
Member Is Offline
Mood: Preparing to defend myself (academically)
|
|
Real chemists don’t need help balancing equations anyway
|
|
|
Sir_Gawain
Hazard to Others
Posts: 497
Registered: 12-10-2022
Location: [REDACTED]
Member Is Offline
Mood: Still in 2022
|
|
And bad chemists might not realize it’s wrong
“Alchemy is trying to turn things yellow; chemistry is trying to avoid things turning yellow.” -Tom deP.
|
|
|
Sir_Gawain
Hazard to Others
Posts: 497
Registered: 12-10-2022
Location: [REDACTED]
Member Is Offline
Mood: Still in 2022
|
|
I’ve got it on my list!
“Alchemy is trying to turn things yellow; chemistry is trying to avoid things turning yellow.” -Tom deP.
|
|
|
![[*]](images/xpblue/default_icon.gif){kind=icon}
