How do you rate ChatGPT’s advice from the transcript below? These are some legitimate questions I had as I don’t have a lot of abyss experience.
Wondering if anybody else has been using it and whether it could supplement the wiki, especially for new players.
“In Dungeon Crawl Stone Soup, if I’m going into the Abyss, should I prioritize HP regen or AC?”
In the Abyss, prioritizing HP regen can be more beneficial. The environment is unpredictable, and you might find yourself in prolonged engagements or constantly on the move. HP regen helps you recover from chip damage over time, which can be crucial when you can’t find a moment to rest. AC is still important, but without a steady way to heal, you might find yourself in trouble. Make sure to have a balance, but lean towards regen if you have to choose.
“Will a weapon with the Vampirism brand be helpful there?”
Absolutely, a weapon with the vampirism brand can be a game-changer in the Abyss! It lets you regain health as you deal damage, which is incredibly useful when you’re constantly taking hits from various sources. This way, you’re not solely relying on natural regeneration or consumables to stay alive. Just be cautious with enemies that are resistant or immune to vampirism, as it won’t be effective against them.
“Are there some enemies that aren’t immune to vampirism in the Abyss?”
Yes, in the Abyss, many enemies aren’t immune to vampirism. Common monsters like wretched stars, starcursed masses, and various types of demons can be affected by it. However, you should watch out for undead creatures, nonliving beings, and some demon types, as they can be immune to vampiric effects. It’s a versatile brand, but always keep an eye on the specific resistances of your enemies.
“What level of the Abyss is recommended to go down to to find the rune?”
To find the rune in the Abyss, it’s generally recommended to go down to Abyss:3 or deeper. The rune is more likely to generate on these lower levels. Keep in mind, the deeper you go, the more dangerous it gets, with tougher enemies and more chaotic terrain. Make sure you’re well-prepared before venturing too deep!
"Is an amulet of regen or amulet of reflection better for the abyss?"
For the Abyss, an Amulet of Regen is often the better choice. The constant health regeneration can make a big difference when you’re dealing with relentless encounters and need to recover between fights. The Amulet of Reflection can be useful for bouncing back damage to ranged attackers, but the consistent healing from regen is typically more valuable in the long run. Of course, it depends on your character build and playstyle, but generally, regen gives you more survivability in such a hostile environment.
FWIW, I generally agree with those sections. But I’d add that Stealth and mobility deserve priority (way) over Regen.
Abyss is designed to destroy you if you stand and fight. Sooner or later, the mobs will simply be too much. You can have 15 billion AC and you’ll still die (because, hey, there are things that just laugh at AC and will flat out stop you from regenerating).
Best not to be noticed.
Second best is, if you are noticed, you can get away. Flight, Passwall, Vhi, Swift, Haste Pots, Warp Wands, Roots, BVC, Paralysis, pets, any god ability that allows specific moves, etc etc etc (and ofc metric bucket loads of Blink and Tel scrolls).
When you get to the rune vaults, odds are, you’ll have to fight rather than be able to ninja. At that point, sure, you want HP/MP regen (and AC too). But even then, don’t hesitate to pull out. I’ve seen folks die by stubbornly bulling forward even though the mob is growing outta hand. Bail. It’s the Abyss. There’ll be plenty of other rune vaults.
In general, don’t use chatgpt for anything you don’t already know. It’s not capable of thought or logic. It’s good at generating answers that have the sort of structure of correct answers so they look correct but what it generates is often not just bad advice but advice that is impossible to follow anyway. Asking chatgpt a question to guide you is foolishness because you need to already know the answer for it to help you. Get your answers from the people with the experience to answer rather than the machine that plagiarizes other people’s answers without the ability to comprehend them!
Addressing the questions and answers: (click to expand)
Most of these answers are mostly right. (It helps that you seem to already have an idea of what’s effective so you can ask specifically about that; avoiding impossible tragically misleading advice like “Make sure you prepare well before entering the dungeon” or “Make use of mummy’s natural resistance to fire”)
WRT to “should I prioritize HP regen or AC” it’s rarely a choice of one or the other. Regen amulets are great because it helps you recover quicker while enemies are constantly spawning. If you’re thinking about trading your heavy armor for light troll leather just for the abyss, it’s probably a bad idea. Keep in mind that one of the enemies in the abyss can give you the “sick” status which stops your regen completely. (you can drink cancel to cure this status effect once the zyme is gone)
Vampirism brand is nice in general but a lot of things native to the abyss resist it. If you have a great weapon with a vampirism brand use it. If your vamp weapon does significantly less damage than your main weapon then probably don’t switch.
It’s not recommended to go deeper than abyss:3 on purpose. The rune can only be found by at least abyss:3 so delving their is important but after that, increasing the difficulty isn’t a good idea.
On most characters without a source of regen I would take amulet of regen over amulet of reflection for abyss trips. (note that the reflection amulet doesn’t just reflect though, it also provides SH which is nice against non-ranged attacks too!).
I agree with blink_frog’s stealth recommendation as long as a few more pips of stealth is something available to your character (don’t ditch your heavy armor just to get stealth).
Another thing I’d highlight is teleport scrolls. They take significantly longer to work in the abyss so you want to use them early, when you sense the danger building up. Teleport scrolls are great in the abyss because the level is completely regenerated. On most floors the monsters would still hang around when you teleported away but in the abyss they are erased from existence.
You likely have some ambrosia potions lying around by this point so you can quaff those while to speed up regeneration when you’re not fighting. (Don’t worry about the confusion. The duration is short and a cure potion can get rid of it if something dangerous comes into view)
I mean you wouldn’t want to actually…use this advice, even though it’s picking up on some generally accurate themes.
None of those have living/plant holiness, so vampiric weapons don’t work on any of them. Vampiric really doesn’t have many targets in abyss, so the overall response is bad.
Similarly
There really isn’t much in the abyss that’s hitting you with reflectable attacks, so while this advice is directionally correct the line quoted above is misleading.
Basically this model seems to have enough crawl-relevant text in its training data to write plausible-sounding advice that incorporates some valid ideas, but mixes them with confident-sounding inaccuracy enough that I very much doubt it would help to follow its advice.
Agree with all of the comments above, seems like at best ChatGPT will give pretty shallow info at best and hallucinations at worst. Not a great resource overall.
Also thanks for all the tips, I have now picked up my first abyssal rune (and also got my first 5 rune win)!
I think you’re tarnishing ChatGPT’s reputation for no good reason. AFAIK, it just aggregates info from the web for you in a nice answer. If you’re trash-talking ChatGPT’s answers, then assuming that ChatGPT is good at aggregating info, you’re really just trash-talking the people in forums/wiki article authors/etc. that give the advice. If you use DuckDuckGo, it has an AI-powered feature called DuckAssist and it always tells you the sources it got the info from, so that’s useful.
(EDIT: Reading this back, I realize this may come across as a rude comment. I didn’t intend to be rude to anyone. Feel free to trash-talk bad or outdated advice )
This is basically wrong, although, like the answers above from ChatGPT, sort of truth-adjacent. LLMs aren’t looking up information and then presenting it you you, they’re synthesizing an answer that looks correct after the prompt based on the data and feedback they were trained with - which can sometimes be confident-sounding incorrect statements combining relevant words and concepts.
as crawl players, we must always be careful to xv our assumptions.
Combating myths about chatgpt. Common expectations for gpt don’t match its actual abilities. The examples I mentioned in the first paragraph of the (click to expand) section of my post were paraphrased but were real suggestions from chatgpt.
It’s true that the training data for the underlying model is sourced from the web. GPT-4 (which chatgpt relies on to generate text) predicts the next “token”(word or part of a word) to use based on what’s already there. It’s predictive text, you see? The model is generating something that’s statistically similar to what it was trained on. This is why it’s capable of generating false answers that no one has quite expressed before. Like this gem:
This guide details a strategy for completing Tomb: Serpent King in DCSS, specifically tailored for a Mummy of Fedhas utilizing the Cleansing Flame strategy. This approach relies on the Mummy’s natural fire resistance and regeneration, paired with the potent damage and self-healing properties of Cleansing Flame.
(there’s no serpent king in dcss, Cleansing Flame is TSO specific, mummys have fire vulnerablilty not resistance and don’t get regeneration bonuses)
You’re better off just searching the web and reading the actual source.
Scanning the DuckAssist announcement blog post, it looks like it whitelists some information sources, searches those sources and feeds that info into an LLM to generate a summary. This does hopefully reduce the odds that it will just “hallucinate” information but I’d still caution against it. We’ve seen in this thread how these generative text machines can produce wrong or misleading info.
It can link you to the source that it’s summarizing but that’s not the same as all the sources that it trained its model on.
The key useful part of this: the sources, isn’t the predictive text’s work at all. The search engine finds a suitable source first! You can ignore the machine generated summary of the source and look at the source itself, removing that extra layer of potential misinformation.
A famous example of people being misled into trusting what is misleading called “AI”:
“We made a good faith mistake in failing to believe that a piece of technology could be making up cases out of whole cloth,” the firm’s statement said.
Maybe. But that’s my point — it’s those people, not the AI. The AI just reflects other people.
Oh with that I definitely agree. There was even that Google employee that thought their AI became self-aware because of an answer it gave him, it’s that silly.
I didn’t read your examples above, I think.
Your statement is a claim about the ability of AI to aggregate sources. Tbh I’ve never used ChatGPT, I only use the DuckAssist feature in DuckDuckGo. I’ve experimented with it a bit and I think it occurred to me once that it evaluated the text in the source incorrectly and gave an amusing reply. But otherwise it seems to work pretty well so far.
The same could be said about Wikipedia, for example. Wikipedia aims to be a user-edited tertiary source that aggregates secondary sources (to a larger extent) and primary sources (to a smaller extent). Is Wikipedia reliable? Should readers of Wikipedia check the quoted sources or talk pages? These are legitimate questions, but I’m saying it here because in this aspect, the DuckAssist feature is not much different than Wikipedia, perhaps they even have a similar reliability rate. EDIT: Nah, Wikipedia is probably more reliable.
I don’t remember ever reading that announcement (although I didn’t open the link to check). You’d be better trying it for yourself. It always takes two of the top results from the SERP, generates an answer based on them, and gives you the links to them so you know exactly which sources it used. Sources that were pre-sourced in the development/training of the AI are obviously not listed, but perhaps they document it in their help pages? Although I think they’re less important so I haven’t checked.
The AI cannot tell truth from lies, and therefore the AI is also ‘reflecting’ the liars. You can’t just skirt around a very relevant counter argument because it isn’t your point.
Why does the AI need to tell them apart? Consider this question: Why do Wikipedia editors need to tell lies apart from truths? (They use sources which are generally considered to be good, and link to them. In comparison, the DuckAssist feature takes a few of the top results of your search and links to them.)
Btw, I assume that ChatGPT can be asked for the sources to its replies?
If the AI cannot tell them apart, it will just give you false information based on said lies. Also, Wikipedia has strict guidelines and is moderated by people who possess critical thinking.
Part of that is that I misread something, I thought you said that something isn’t your point when you has, in fact, said that it was your point. However, your response to the point of bad advice existing on the internet seemed to be (I think) that the AI doesn’t reflect bad advice. Which isn’t really correct, remember Google’s AI search assistant giving hilariously wrong results?
That’s the opposite of what I said. I said that assuming the AI does a generally good job at aggregating its sources, it’s the sources’ fault and not the AI’s fault.
You would get the same lies if you accessed the sources directly.
AIs also have developers that improve them, and Wikipedia’s policies don’t replace critical thinking from the reader, rather they allow and encourage it.
Yeah, I struggled to figure out exactly what you were saying, I’m not surprised that I misinterpreted it. But you are working under an assumption. And since AI will pull from Reddit, it can spit out bad advice, or be fooled by people making jokes or sarcastic statements in authoritative voice.
Accessing the source directly might make the issue more obvious.
@zikada you say over and over that you’re making assumptions then when people with more specific knowledge point things out you just double down with more assumptions.
The original question is specifically about ChatGPT which functions differently from the Duck sauce thing you brought up. The duck is a straw man and you’re arguing from a place of ignorance.
I use ChatGPT a lot for work and I’ve tried it for DCSS. It absolutely sucks balls for the things you think it is good at.
I thought the discussion was about AI in general with ChatGPT as an example. The OP wanted to get info about the game and asked about ChatGPT because that’s what they know of and it’s AI so they doubt its reliability. So I felt DuckAssist is relevant to the discussion and the OP is free to try it and see how well it works for them for finding info on the web, including tips for DCSS. Furthermore, as far as I know, DuckAssist and ChatGPT use the same models (or different versions of the same models?), so their quality should be similar. (Also, stating that I’m thinking things about ChatGPT because of my experience with DuckAssist helps make my position more transparent regarding the bias I might have.)
The assumptions I’m “making” are inherent to the core of the argument and I’m mentioning them for transparency and correctness. You trying to say I’m wrong for “making” assumptions means you didn’t understand what I posted. Sometimes, stating the obvious is necessary when building an argument, and you don’t want to assume what others would see as obvious as what you see is obvious.
)