Phenethylamines Here is the comparison of 2C-X drugs with IA (chatGPT,Gemini,DeepSeek)

[Dr.Methyl]

At first here are the result from chatGPT with the prompt : "can you give me a table of comparison for the 2C-X phenethylamine"

Comparison Table: 2C-X Series​

Compound	Typical Oral Dose	Duration	Visuals	Body Load	Mental Effects	Notable Features
2C-B	15–30 mg	4–6 h	Moderate, colorful, cartoon-like	Low–moderate	Euphoric, empathogenic, playful	One of the most popular, sometimes compared to a cross between MDMA & LSD.
2C-I	14–22 mg	6–10 h	Strong, bright, geometric	Moderate (can feel stimulating)	Analytical, energetic, less emotional	Often stimulating, can feel “speedy.”
2C-E	10–20 mg	8–12 h	Very intense, fractal-like	High (nausea, body tension)	Deep, philosophical, serious	Considered one of the “heavier” 2Cs; challenging but profound.
2C-C	20–40 mg	4–6 h	Gentle, pastel-like	Low	Calm, introspective, light	Softer than 2C-B; sometimes called “beginner-friendly.”
2C-D	20–50 mg	4–6 h	Mild–moderate	Low	Cognitive enhancer, subtle psychedelia	Nicknamed the “philosopher’s stone”; can be light but very insightful.
2C-T-2	12–25 mg	6–8 h	Rich, immersive	Moderate–high (possible nausea)	Emotional, introspective	Can be sensual and deep, but sometimes heavy.
2C-T-7	10–30 mg	8–12 h	Very strong, colorful, immersive	High (risk of nausea, body stress)	Euphoric but can be overwhelming	Powerful but associated with safety concerns (overdose risk).
2C-P	6–12 mg	10–16+ h	Very strong, long-lasting	High	Intense, sometimes difficult	Known for extremely long duration; small dose differences matter a lot.

Here is google gemini with the prompt "give me a table of comparison for 2c-x phenethylamines"

Comparison Table of 2C-X Phenethylamines​

Compound	4-Position Substituent (X)	Oral Dose Range (Typical)	Duration (Oral)	Primary Noted Effects & Characteristics
2C-H	Hydrogen (H)	30–100 mg	4–6 hours	The parent compound and minimally active. Used as a precursor to other 2C-X compounds.
2C-D	Methyl (CH3)	20–60 mg	4–6 hours	Low potency, short-acting. Noted for a clear headspace and introspection with mild visuals.
2C-C	Chlorine (Cl)	20–40 mg	6–8 hours	Generally described as gentle, forgiving, and light. Produces moderate visual and cognitive effects.
2C-B	Bromine (Br)	10–25 mg	4–8 hours	The most well-known. Balanced visuals and body high; often noted for tactile and mild empathogenic effects.
2C-I	Iodine (I)	15–25 mg	6–10 hours	Stronger visual distortions than 2C-B. A classic, well-rounded, and distinctly psychedelic experience.
2C-E	Ethyl (CH2CH3)	10–20 mg	8–12 hours	One of the most intense and deepest compounds in the series. Known for powerful visuals and high cognitive load.
2C-P	Propyl (CH2CH2CH3)	6–12 mg	10–16+ hours	Highest potency and longest duration. Can be overwhelming and physically demanding.
2C-T-2	Ethylthio (S-CH2CH3)	12–25 mg	6–8 hours	A "T" (thioalkyl) compound. Highly visual with a distinct psychedelic character and a noticeable "body load."
2C-T-7	Propylthio (S-CH2CH2CH3)	10–30 mg	8–15 hours	Long-acting with a slow onset. Notorious for a high toxicity risk when insufflated and strong, unpredictable effects.

When I added "give me more 2C-X"

Less Common and Substituted 2C-X Phenethylamines​

Compound	4-Position Substituent (X)	Oral Dose Range (Typical)	Duration (Oral)	Primary Noted Characteristics & Structure
2C-G	3,4-Dimethyl	20–35 mg	18–30 hours	Longest duration among the non-thioalkyl 2C-X compounds. Considered by Shulgin to be "The King" of the simple 2C series.
2C-T-4	Isopropylthio	8–20 mg	12–18 hours	An isomer of 2C-T-7, known for its extended duration and strong visual/mental effects.
2C-T-21	2-Fluoroethylthio	8–12 mg	7–10 hours	High potency. Known for a distinctive and often intense mental space and strong visuals.
2C-N	Nitro (NO2)	100–150 mg	4–6 hours	A relatively high-dose compound. Described as primarily stimulant and psychedelic-light with a low visual component.
2C-B-FLY	Rigidified 2C-B	5–15 mg	6–10 hours	Part of the "FLY" series where the methoxy groups (OCH3) at the 2- and 5-positions are chemically 'locked' into benzodifuran rings, increasing potency.
2C-G-3	3,4-Trimethylene	16–25 mg	12–24 hours	A homolog of 2C-G, known for an extremely long duration and heavy body load.
2C-G-5	3,4-Norbornyl	10–16 mg	32–48+ hours	One of the longest-lasting compounds in the entire series, with a very slow onset and incredibly extended duration.

When I ask to add 2C-YN :

• Effects: It is reported to have relatively mild psychedelic effects.
• Dosage (Limited Data): Chemical literature suggests a dosage of around 50 mg.
• Duration (Limited Data): The reported duration of effects is approximately 2 hours.
• Data Availability: Very little scientific data exists regarding its pharmacology, metabolism, and toxicity in humans.

As a substance lacking extensive research, its use carries significant unknown risks.

Here is DeepSeek with the same prompt then is added "give me more 2C-X"

Compound	Typical Oral Dose (mg)	Duration (Hours)	Key Characteristics & Subjective Effects	Potency & Onset	Unique Notes & Risks
2C-B	15 - 25	4 - 8	- The "Goldilocks" Psychedelic. - Strong visual/tactile enhancement, clear headspace, euphoric, sensual.	Onset: 45-90 mins. Potency: Moderate.	- Well-regarded for its manageability. Favorable safety profile relative to others.
2C-I	15 - 25	6 - 8	- Stimulating, "electric." Patterned, neon visuals. Stronger body load.	Onset: 60-90 mins. Potency: Similar to 2C-B.	- Associated with vasoconstriction. Linked to fatalities in combinations.
2C-E	10 - 20	6 - 10	- The "Deep Thinker". Intense introspection, intricate visuals, heavy body load, challenging.	Onset: 90-150 mins. Potency: High.	- Not for beginners. Steep dose-response curve.
2C-C	20 - 40	4 - 6	- The "Gentle" Psychedelic. Mild, warm, forgiving. Soft visuals, minimal body load.	Onset: 60-90 mins. Potency: Low.	- Often used as "landing gear" from longer trips.
2C-D	50 - 100	4 - 6	- The "Psychedelic Catalyst". Very subtle alone. Potentiates and modifies other psychedelics.	Onset: 60-90 mins. Potency: Very Low.	- Primary interest is in combination use.
2C-T-7	10 - 20	8 - 12	- Complex & long-lasting. Initial MDMA-like euphoria, then deep psychedelic phase.	Onset: 90-180 mins. Potency: High.	- High Risk: Fatal interaction with MAOIs. Linked to fatalities.
2C-T-2	15 - 25	8 - 12	- Similar to 2C-T-7 but generally considered slightly less intense and more visual. - Empathogenic and euphoric come-up, transitioning to a classic psychedelic state.	Onset: 90-150 mins. Potency: High.	- Shares the same MAOI warning as 2C-T-7. Much less common and very little human data.
2C-T-21	15 - 25	6 - 10	- Reported to be gentler than 2C-T-2 or 2C-T-7.<br- > Milder headspace with distinct visual patterns. Lighter body load.	Onset: 60-120 mins. Potency: Moderate.	- Extremely scarce. Anecdotal reports are limited and old. Safety profile is largely unknown.
2C-P	6 - 12	12 - 20+	- The "Marathon" Psychedelic. - Extremely long duration. - Very heavy body load and strong vasoconstriction. - Profoundly powerful and immersive visuals and headspace.	Onset: 2 - 4 hours. Potency: Very High.	- EXTREMELY DANGEROUS. - Has a very narrow safety margin. Overdoses can be fatal due to intense vasoconstriction and seizures. - Absolutely not for recreational use.
2C-G	5 - 15 (Very limited data)	16 - 24+	- One of the most obscure and longest-lasting. - Effects are described as subtle, complex, and "shamanic." - Extremely slow onset and offset.	Onset: 3 - 5 hours. Potency: High.	- Almost no human data exists. - The incredibly long duration poses significant risks (dehydration, psychological fatigue).
2C-N	50 - 150 (Limited data)	6 - 10	- Known for its rapid onset and short duration relative to its dose. - Effects are described as "loud," stimulating, and visually complex.	Onset: 20 - 45 mins. Potency: Low.	- Unusual due to its high dosage range and fast onset. Requires large amounts of material, which can be impractical.
2C-H	(Not Active)	N/A	- This is the base phenethylamine structure for the 2C-X family. - It is not psychoactive on its own.	N/A	- Important as a chemical precursor. It is what you start with before adding the substituents (B, I, E, etc.).

Then I tried to add 2C-YN :

2C-YN

1 - 3

8 - 14

- Extreme Potency. - Effects similar to 2C-I or 2C-E: stimulating, visual, with a strong body load and headspace. - Anecdotal reports of anxiety and physical tension being pronounced.

Onset: 60 - 120 mins. Potency: Extremely High.

- CRITICAL WARNING: Active in the sub-milligram to low milligram range. - Extremely narrow therapeutic index. A slight miscalculation can lead to a severe, dangerous overdose. - Represents the new frontier of "ultra-potent" phenethylamines. Handled only with professional lab equipment.

 

[PsychedelicSummer]

Many things right, many things simply wrong like this about 2C-T-21 "Known for a distinctive and often intense mental space and strong visuals."
Without knowledgable human editing these kind of AI summaries can cause harm.
Labeling such as for 2C-P "The "Marathon" Psychedelic." is ridiculous, if you compare to things like DOB or DOC.

 

[xdrc]

I think there was a previous discussion about AI slop on the Bluelight, and I am strongly opinionated that it does not belong here. I think if the errors were corrected, it would perhaps be acceptable, but I'm not sure what your intention is with this content. Btw 2C-YN as per Trachsel is not very potent, so this shows how competent AI is at hallucinating.

 

[perpetualdawn]

Yeah I feel pretty strongly about this too think threads like this should just be deleted. Don't get me wrong, I like the LLMs very much, but this is not the place for posting low effort AI junk.

 

[The Holy Quadruplty]

What do y'all think about Nutmeg metabolizing into psychedelic phenethylamines, and Nutmeg main ingredients themselves?

What about Phenethylamine itself from chocolate, or Coffee combined with Nutmeg given it's MAOI effects, and also coffee's MAOI effects as well?!

I know Elemicin may possibly metabolize into Trimethoxyamphetamine, Trimethoxycinnamic Acid, etc... And not to mention similar metabolites of Mescaline itself.

 

[Vastness]

Yeah I'm also a big fan of LLMs for the most part but this shit is such low effort, I'm not totally against the idea of stuff like this being posted but not just verbatim copy/paste prompt and output, if absolutely anyone could ask the exact same question why in the hell are you posting it here, this is the modern equivalent of if forums existed when search engines like Google were invented and people being like "Look what I searched for and here's my copy paste of the results!" and just some slop direct copy of the first 10 search results with all the formatting messed up... the fact that LLMs can now format neat tables for you super easily doesn't mean the information is actually any more useful, if anything it's less useful and certainly more dangerous because of the illusion of competence... just so tiresome.

Absolutely no one should rely on an LLM for dosage advice with any substance right now, and again I say this as a fan of LLMs generally and with the recognition that in actual fact, with more specific, definite queries and some precautions taken (and some pre-existing knowledge, questions that aren't life or death ideally, etc) some frontier models are probably capable of giving somewhat safe answers to many questions a very large proportion of the time. But not like this, this is pure slop, too much to ask for in one shot, it's not possible to trust without verification that any of this is reliable.

If you absolutely must post shit like this IMO there should be a rule that you need to include commentary that confirms that you verified the output yourself (and from where? include sources) with CLEAR INDICATIONS where any incorrect information has been generated. OR AT THE VERY LEAST some kind of commentary as to why you think your post is worth making, what's interesting about it, how people should interpret it, your own thoughts, just WHATEVER... god this is so stupid...

This post actually does kinda have the makings of having something interesting to say about stuff since you compared different models, but you didn't even do that properly since you couldn't even keep your extremely simple opening prompt straight between the first 2 models you asked. I'll provide some of the commentary you missed, though - reiterating that no LLM, even the frontier models should be relied on for substance-related advice at present - but, that being understood, DeepSeek is most definitely one of the LEAST safe models, yes DeepSeek is technically very impressive but it's hallucination rate generally is sky high compared to the larger closed-source models. In fact unfortunately currently no open source model is beating any closed source multi-trillion parameter model, the hallucination problem is just not sufficiently solved.

After that, free ChatGPT should not be trusted either because you have no clue what model you're getting, OpenAI will randomly shunt you around based on demand, but more importantly the GPT models out of all the closed source frontier models have the highest hallucination rates AND a disturbingly high tendency to sycophantic answers, doubling down on being wrong, etc, etc. This has been a problem since GPT-4o, but remains an unsolved problem in o3 and o4, where the sycophancy gets better but hallucination rates really peak. GPT-o1 and GPT-5 with reasoning might be pretty reliable in most cases but o1 is only accessible via API now and tbh I haven't really kept up with OpenAI's fuckery since the release of GPT-5 so I'm not totally sure where it's at as far as the hallucination problem, I'd imagine it's still a problem though.

Gemini 2.0+ are fairly solidly aligned models with hallucination rates that are still significant but not alarming, out of the models you tested this is the most reliable one. You didn't test the Claude family either but they are also very strongly aligned models, and with the caveats that you're going to still verify any output would probably be the most trustworthy. Both Gemini and Claude will sometimes refuse to answer certain substance related questions outright, but this is the correct behavior from an aligned model.

 

[iom]

LLM hallucination rate is 100% because they are relying on probabilistic relationships between snippets of text. That their hallucinations appear accurate is essentially by design, a consequence of how they are optimized. Ultimately there is no solution to the "hallucination" problem with LLMs that doesn't involve subjugating it to a completely independent information system, a system of "hard facts" if you will. Such a system of hard facts needs to be maintained manually just like any other kind of knowledge base, with all the problems that entails, and in the end, the LLM will then only function as a veneer of personality through which a user interfaces in order to access said hard facts. Think of it as like having to trick the AI into doing the Google search you want it to do because you can't do the damn search yourself.

Meanwhile, the VC and tech industries have pretty much bet everything on this one tech at the expense of work on practically everything else that was or could have been. Now they've committed so many resources that they (rightly) fear no one will take them seriously if they can't pull it off, so they are desperate to keep scaling up up up---hoping that maybe something will finally click---and the models will finally "stop hallucinating" or whatever. Nope!

 

[xdrc]

Think of it as like having to trick the AI into doing the Google search you want it to do because you can't do the damn search yourself.

I would agree if it weren't for the fact that I noticed not being able to "find shit" anymore, while I previously was able to 5 - 10 years ago. Have I gotten more incompetent? Has Google (and other search engines) gotten more incompetent? Has the knowledge base (i.e. the web) become more incompetent? It's certainly a mixture of the latter three*, because how could the first be true? AI so far has been quicker for some requests I found near impossible to find manually - The free model of ChatGPT meanwhile has gotten better and better, but then seemingly got lobotomised for my use cases. Still saving a lot of time (while still absolutely requiring independent checking with the actual sources!).

I'm curious how this will all end up. I believe AGI is possible (it kind of has to be, depending on ones interpretation of what AGI exactly entails), but I doubt scaling up current systems is the solution. This is based on intuition on the peak of Mt. Dunning-Kruger, not on any detailed technical insights.

*EDIT: meant to write two, but keeping it is awesomely ironic.

 

[Pfafffed]

Dr. Methyl is new to Bluelight. It's nice to see them engaging. And I did find this interesting, mostly from the perspective of seeing what the LLMs got right, what they got wrong, what they thought was salient, and what they missed. But going forward, I think most of us would prefer to see posts including stuff like this framed. What is it that you think is important? Why are you sharing it? I think that would be better received.

@The Holy Quadruplty You might want to repost this in its own thread or just one of the social threads. There's a lot to unpack there, and none of it is on topic.

 

[Vastness]

LLM hallucination rate is 100%

lol, I guess technically but it's kind stretching the definition of a "hallucination" here. Arguably humans also have a 100% hallucination rate given that consensus reality is a very tenuous and fortuitous alignment of the ever present hallucination of consciousness which is generated by our brains every moment we are sentient and is all we can ever really know. At some point you gotta just take the "pretty good" solution - from humans or machines - rather than hoping for a "perfect" one.

Obviously calling erroneously generated outputs "hallucinations" rather than just "wrong data" or something is kinda just an extension of the anthropomorphizing of LLM type AIs, since it implies a subjective belief from the LLM that they are in fact stating a correct fact rather than just making shit up. Without yielding entirely to the loopiness of ascribing too many human tendencies to something entirely different and entirely ahuman, even allowing for a fairly loose definition of what it is to have a belief and whether or not software can "believe" anything to be true - it's not only possible that they might hallucinate, it's also possible that they may be lying.

Obviously yeah for all intents and purposes an entirely deterministic system is the only truly "hallucination proof" one, but such systems would still be vulnerable to "hallucinations" (mistakes) from humans or machines that assisted in putting together in the first place - or perhaps in some cases even a bit-flip from a cosmic ray causing an unreproducible garbled output. On that note, although the hallucination problem will never be totally "solved"...

I believe AGI is possible (it kind of has to be, depending on ones interpretation of what AGI exactly entails)

IMO AGI - or more definitively perhaps, "ASI" (Artificial Superintelligence) could be considered to have been essentially attained when AI systems have effective hallucination rates in any given task that, over a large sample size, are consistently lower than the best that humans can achieve (ie, not just individuals, but even large corporations and such with multiple checks and balances in place). I'm fairly sure this has already happened in highly specific areas, but I guess AGSI, generally superintelligent entities, would be expected to be the point where we'd start to genuinely need to question what it is to be alive, so to speak, since such entities would necessarily be as competent as the leading frontier LLM in communicating with humans, and additionally retain all the more fine-grained superhuman pattern matching "intuitions" which we currently mainly see in other non-chatbot machine learning systems (and as such because they don't talk, we don't call it "intuition", currently)...



Yeah I agree @Pfaffed, sorry @Dr.Methyl, gods sometimes I accidentally read my own fucking posts back and realise all I do is fucking complain about shit, fuck. Welcome to Bluelight.

 

[xdrc]

But going forward, I think most of us would prefer to see posts including stuff like this framed. What is it that you think is important? Why are you sharing it? I think that would be better received.

I liked seeing 2C-D being dubbed "the philosopher's stone", because I kind of agree and am very fond of this material. Wonder where it got that from, because I never saw it before.

 

[Leprechaun]

I’ve tried using ChatGPT for chemistry. It’s bad digging into chemical formulas and chemical names it falls apart quickly.

You can see it already saying weird nonsensical things with 2c-b-fly being “rigified” which isn’t even a word

Even more interesting is that the phrase "rigified" is sourced from bluelight itself. There is no other record of that phrase. You can do this yourself by searching for 2c-b-fly and rigified.

Now all of a sudden, a laypersons off hand comment has fallen into the training data…

 

[moonyham]

The crimson color of everything, kind of like looking at a phone with polarized sunglasses that I experienced on 2c-b was something I will never forget.

I got some about 2 years ago but lost it amongst my possessions when I moved. I know it's somewhere.. I'm going to look really hard when I'm back home. Anyone know if it should be sweet? It will be somewhere dark and dry.

 

[Pfafffed]

It will be fine