Talk:Prompt engineering

Comparing a previous revision of this article with the current version, I noticed several issues:

• WP:DEADREF - At least 11 other pages link to the section "Chain-of-thought" but that section has been removed, breaking the links (https://en.wikipedia.org/w/index.php?title=Chain-of-thought_prompting)

• WP:FRANKENSTEIN where the cutting and reordering of sentences has changed the meaning. For example, one sentence that originally meant "in-context learning is also known as mesa optimization" now reads something like "the GSM8K benchmark is also known as mesa optimization". One sentence that defined adding an explicit instruction such as "Let's think step-by-step" as Chain-of-thought prompting now incorrectly identifies that same technique as zero-shot learning

• WP:INTEGRITY, text-source integrity has been lost in several places: citations have become disconnected from the text that used to refer to them. For example, this^[1] and several other citations about Chain-of-Thought prompting have became attached to an unrelated sentence about fine-tuning

• Related to the above, WP:OVERKILL in several places where as many as 8 citations have piled up at the end of a sentence where they don't belong (because the content they used to refer to has been deleted or re-phrased)

• Definitions of the key terms prompt, few-shot learning and chain-of-thought have gone missing. Abbreviations like "CoT" are used, but never defined (WP:MOSABBR)

I'm attempting to clean up the article by making the following changes:

• Restoring some of the recently-deleted content from the History and Chain-of-thought sections

• Moving inline citations as close to the content as possible, and adding |quote= wherever applicable, to help maintain WP:INTEGRITY in the future

• Filling out the Text-to-image section which needs expanding

• Fixing a few broken citation issues

I've attempted to fix the issues while preserving all content that has been added since July, but if there's anything missing please add it back in with proper citations. Lwneal (talk) 21:08, 14 August 2023 (UTC)

@Lwneal, I want to express my gratitude for addressing the issues while maintaining the recently added content. As a reader interested in comprehending how LLMs facilitate prompt engineering through in-context learning, I find the current version satisfactory. However, individuals focused solely on utilizing "prompt engineering" to harness these capabilities might prefer a perspective centered exclusively on p.e. I personally favor the present approach, and your skillful adjustments to resolve problems stemming from attempts to rephrase the article solely from the p.e. viewpoint are greatly appreciated. P. S. The topic of dual perspectives requires focused consideration in the article's introduction. While it was previously addressed in the opening, I am contemplating its potential reintegration: "Prompt engineering and in-context learning are terms used to describe a teaching approach in the context of interactions with a large language model. These terms are employed from different perspectives. Firstly, prompt engineering is used from a user's standpoint, emphasizing crafting series of prompts to achieve the best outcomes or guide the model's behavior within the scope of a single conversation. On the other hand, in-context learning is used from a machine learning perspective. It refers to the model's ability to learn and adapt from the very same series of prompts provided by the user during the conversation. This adaptability allows the model to improve its responses based on the current context. The temporary nature prevents carrying the unwanted old contexts or biases, except the ones already present in the (pre)training dataset, from one conversation to the other." DancingPhilosopher (talk) 07:48, 16 August 2023 (UTC)

I agree with this dual-perspectives viewpoint. This article is about two sides of the same concept:

• Prompt engineering is the practice of crafting text prompts to guide or control an AI. It's a skill that anyone can learn to do, like drawing or juggling.
• In-context learning is a special ability some AI systems have, to learn and adapt from text prompts. It's a technical term used in academic machine learning, like domain adaptation or inductive bias.

A human performs prompt engineering when they write a prompt. An AI performs in-context learning when it reads the prompt, understands it, and correctly responds.

If you are an office worker and you want to know how to prompt GPT-4 to do your work for you, then you want an article about prompt engineering with practical information and lots of examples.

If you are a student and you want to know why GPT-4 understands what you say, then you want an article about in-context learning that talks about mesa-optimization, meta-learning, and the temporary/permanent distinction between fine-tuning weights and in-context learning activations.

If, in the future, AI becomes more widespread and this article grows, perhaps in-context learning will become its own article focused on theory, leaving prompt engineering focused on practical matters. An analogy might be driving vs internal combustion engine, or aviation vs aeronautics.

I propose adding a new section ==In-context learning== containing an expanded discussion of the machine learning perspective, while the other sections continue to focus on the user perspective. Lwneal (talk) 19:31, 16 August 2023 (UTC)

How is RAG to be pronounced? "rag" or "r-a-g"? WiseWoman (talk) 14:49, 22 June 2024 (UTC)

It's pronounced like the English word. People will also say "retrieval augmented generation" when reading something aloud even if the text itself says "RAG"

Stellaathena (talk) 20:45, 13 February 2025 (UTC)

The section is confusing and I believe it is wrong. This is a better description: https://aws.amazon.com/what-is/retrieval-augmented-generation/ — Preceding unsigned comment added by CaliViking (talk • contribs) 14:49, 9 July 2024 (UTC)

As written it seems accurate to me. What did you find that was inaccurate? Michaelmalak (talk) 19:15, 9 July 2024 (UTC)

The article seems to dive into the technical vector implementation before positioning the overall concept:

"This relevance is typically determined by first encoding both the query and the documents into vectors, then identifying documents whose vectors are closest in Euclidean distance to the query vector."

The AWS description seems to be more logical and easier to follow. The technical descriptions are in the context of the broader explanation. Saying that it is "wrong" may be a too strong statement, I would rather say that it is "confusing". Do you agree?

CaliViking (talk) CaliViking (talk) 18:11, 10 July 2024 (UTC)

As you are not disputing the accuracy of the section, I've updated the tag template to {{Confusing section}} to reflect this, and included your comment. I agree there is a tendency for technical topics on Wikipedia lack a decent definition before launching into the details of implementation. When you spot this, I encourage you to be bold and add a short definition of the subject! Turtlecrown (talk) 07:48, 16 August 2024 (UTC)

I see no definition of what "dense embedding" is, There is no definition on the page and I could not find a relevant article in the Wikpedia. Articles on the embeddings never mention "dense" ones. See Word embedding, Embedding, Vector embedding Petrgl

@User:Michaelmalak: Could you please give a bit more of a comment why you removed the review? The article is often based on links on help pages of OpenAI; I think a review would contribute to a better quality. --Minihaa (talk) 18:49, 12 August 2024 (UTC)

It was unencyclopedic to say "here's an excellent resource". Any opinion of flattery has to: 1) be in quotation marks, and 2) cited specifically, preferably from multiple sources. Additionally, the prose before the citation needs to substantially contribute to and flow with the rest of the article. Saying "here" is like 1990s GeoCities. Michaelmalak (talk) 20:47, 12 August 2024 (UTC)

This article seems, at least in its current form, to be highly biased towards the idea that "prompt engineering" is scientifically valid; that's far from the consensus understanding in computer science, however. Indeed, most of the citations on this page are either unpublished preprints, press releases from companies with a strong conflict of interest, or popular summaries of preprints and press releases. Perhaps even more seriously, some of the citations (such as the one to CNET) may themselves be the output of LLMs, given CNET's use of LLMs.

The stochastic parrots article may be an interesting example for an article on the subject more in line with WP:NPOV. cgranade (talk) 21:56, 23 October 2024 (UTC)

I don't really see where the bias is, so my prior is that the "POV" template is not warranted. But it's the first time that I see someone question the validity of prompt engineering as a concept, so I would be interested in seeing your rationale. Alenoach (talk) 00:59, 1 November 2024 (UTC)

Do you mean that you don't think that prompt engineering improves the results? Or that it can work but isn't scientific? Alenoach (talk) 01:17, 1 November 2024 (UTC)

As mentioned, the sourced cited mostly seem to have a strong conflict of interest — OpenAI, Microsoft, and Google documentation and press releases are very biased towards sales of their respective products, and as a result, towards the non-consensus view that LLMs work and are programmable. Similarly, a lot of the prose in this article takes its sources as literal fact (e.g.: "the ability for in-context learning is an emergent ability of large language models"), again biasing away from consensus views in computer science. I'd definitely suggest a rewrite along NPOV that makes it clear that the article is documenting claims by AI vendors, rather than the consensus understanding of fact within computer science. cgranade (talk) 18:17, 1 November 2024 (UTC)

I added this to my watch list because the article needs work, but I don't think I agree with this complaint. "Prompt engineering" is basically just "trying to write good prompts." If there are sources that say "trying to write good prompts doesn't exist" or "trying to write good prompts is useless," then provide those sources. Otherwise, I don't see what point of view would need to be represented here to make this more neutral. Asparagusstar (talk) 02:20, 1 November 2024 (UTC)

Wikipedia is not required to prove a negative on pseudoscience topics. The predominant point of view represented here by peer-reviewed science is the stochastic parrots one. The sources that are self-published by large AI companies are not scientific, they are marketing, so they can only be appropriately used to contextualize "prompt engineering" as a marketing term. rspεεr (talk) 21:22, 1 November 2024 (UTC)

Even supposing that the stochastic parrot hypothesis were true, it would not necessarily invalidate the concept of prompt engineering.

For the sources, it would be better if they were from reliable secondary sources. But I can understand that the field is very young, has grown very fast and that peer-reviewed journals are overflowed. So perhaps one alternative way to estimate their quality is to check the number of citations, many have more than 100 citations, which is a sign of quality. Alenoach (talk) 22:03, 1 November 2024 (UTC)

The stochastic parrots paper does in fact preclude prompt engineering as a concept (as does the whole idea of an LLM in the first place, but I digress). To be sure, the concept of prompt engineering is at odds with the consensus view in computer science. As you note, however, the hype around AI has grown quickly enough that high-quality evidence in AI is extremely difficult to find. As a function of that hype, citations are an unfortunately very poor measure of quality — even a cursory look through some of the preprints cited here shows severe methodological errors, such that there's very little left to support the claims made in this article.

I think the cleanest solution to the NPOV problem in this article would be to rewrite the prose to be exceptionally clear about which institution is making which claims; virtually none of those claims are in keeping with consensus, and should be cited not as fact, but as statements from highly biased sources. cgranade (talk) 06:52, 2 November 2024 (UTC)

Yes, if you see a claim (or suggestion, or theory) presented as a fact, you can make clear that it is a claim and who is making the claim. Also, if it is a dubious claim made by a single source, you can just remove it. Asparagusstar (talk) 16:08, 2 November 2024 (UTC)

Sounds good. I've started editing to that effect; most of the claims here are pretty dubious and stem back to non–peer-reviewed publications by Google and/or OpenAI, so I've tried to indicate that where possible. I'll make another pass and remove the least substantiated claims in another pass, then, as per your suggestion. cgranade (talk) 19:56, 2 November 2024 (UTC)

Please be moderate in your modifications. Chain-of-thought, for example, is pretty widely acknowledged, it's not just a Google thing. The article already has various sources for this. Saying once that it's according to Google can be relevant, but repeating it many times hinders readability. Thanks for your understanding. Alenoach (talk) 20:10, 2 November 2024 (UTC)

I think I was quite moderate? I disagree entirely that chain-of-thought is widely acknowledged outside of AI vendors that have a specific financial interest in their claims. The vast majority of sources I've found supporting that chain-of-thought is a real thing are from Google, with a small number from Amazon. What third-party sources I have found seem to each cite back to Wei et al by Google as evidence that chain-of-thought is real.

In the interest of being moderate, I left the section in rather than deleting dubious claims as you suggest.

For the point about repeating the attribution, I appreciate the feedback; do you have any suggestions as to how to edit accordingly, given that there's whole paragraphs that make strong non-consensus claims tracing back to a single institution? cgranade (talk) 20:27, 2 November 2024 (UTC)

To elaborate: I looked at the current sources for this article. It's possible that some of the preprints here have since been peer-reviewed, but I see only _one_ paper that is attributed to a peer-reviewed source -- the AutoPrompt paper by Taylor Shin et al. That paper makes no claims about LLMs except that it "may be useful for ... GPT-3", a claim that is presented speculatively, because the "AutoPrompt" system being described was not designed for GPT-3. The evidence in the paper is about _masked language models_ -- that is, the previous generation of language models, such as BERT. Separately, there are many sources that are news articles about prompt injection, which has its own article. rspεεr (talk) 21:41, 1 November 2024 (UTC)

I'm reading through all nine paragraphs here I'm still not sure what is being suggested here ... if there are additional neutral reliable sources being suggested, or unreliable non-neutral sources currently in the article that someone is suggesting be removed ... but honestly I don't think you have to explain that here. Just add whatever reliable source you want to add, remove whatever unreliable source you want. Go for it! Improve the article! Asparagusstar (talk) 00:34, 2 November 2024 (UTC)

The problem is that there aren't really any reliable sources backing up the idea that "prompt engineering" exists. There's a lot of tech company marketing to that effect, and a small community of non-reviewed papers that primarily cite each other, but beyond that, there's not really any empirical or theoretical support for the idea of "prompt engineering."

From that standpoint, it's difficult to think of improvements to the article. As it stands, the article is almost entirely comprised of unsubstantiated marketing claims. From the perspective of notability and factual accuracy, the page is an ideal candidate for deletion, but that doesn't feel like the right answer given the amount of money and investment predicated on "prompt engineering." Perhaps one solution would be to replace the content of the page, which effectively endorses the unsubstantiated claim that there's a factual basis to "prompt engineering," with a description of industry marketing, job roles and practices, and so forth. cgranade (talk) 22:14, 15 January 2025 (UTC)

It is sometimes necessary to make assumptions to write an article (see WP:MNA). And if you consider a paper like the one introducing chain-of-thought, it already has around 10,000 citations despite the fact that it was published only 3 years ago, which shows that it had a large academic impact. I concede though that some techniques have probably lost part of their relevance over time. For example, some LLMs are now trained to "think step by step" when addressing a mathematical problem, even when you don't explicitly indicate it in the prompt. Alenoach (talk) 20:26, 16 January 2025 (UTC)

WP:MNA seems to point strongly against this article? Using the example there about creationism, it wouldn't be relevant to an article if a paper about creationism has had 10,000 citations over three years — creationism still flies in the face of the broad understanding of how biology works.

In this case in particular, I've read the chain-of-thought paper and there's very little there to substantiate its claims. The experimental results cited consist of cherry picking a few models and running some benchmarks; there's no evidence presented at *all* that the improvements that they see on those benchmarks is consistent with their proposed explanation for how LLMs "work." It's notable that in the appendix, they hook up their benchmarks to Python's eval() function to address the fact that LLMs do not, as a general property, perform arithmetic computations. That the paper lacks an empirical and theoretical basis might go a long way towards explaining why a preprint remains without peer review three years after its initial announcement on arXiv; that's definitely an unusually long time outside of formal mathematics. Similarly, it's notable that every individual author is on a team with significant financial interest in positive results — while that alone doesn't invalidate anything, it does make it harder to use as a primary source of evidence for an extraordinary claim that flies in the face of established computer science knowledge.

Pulling back for a moment, I understand and acknowledge that Wikipedia is not itself peer-review, nor the place to get into the epistemological failings of individual papers. The more general problem remains, though, that the prompt engineering article as-is depends only on extremely biased and unreviewed sources for extraordinary claims. From a NPOV perspective, that seems like a problem that would warrant significant rewrites, if not outright deletion. cgranade (talk) 22:04, 16 January 2025 (UTC)

WP:MNA is not a reason to regard unreliable sources as reliable. Most of the citations currently in this article do not meet our standards for reliability. XOR'easter (talk) 17:36, 18 January 2025 (UTC)

I wrote this regarding the debate on whether prompt engineering is a valid concept, not to justify using unreliable sources. What I mean by this is that it's hard to write a good article without assuming that prompt engineering is a valid concept and can potentially improve the performance of LLMs. Contrary to what cgranade says, the concept of prompt engineering is not even particularly controversial in the scientific community, and the paper introducing the chain-of-thought technique was published in NeurIPS 2022 and thus peer-reviewed. Alenoach (talk) 17:50, 18 January 2025 (UTC)

It may not be "particularly controversial" in the community of people who post unreviewed preprints to the arXiv about it, but that doesn't mean that we can use those unreviewed preprints to provide further information about it. If all the details come from such sources, then we can say that prompt engineering exists as a concept, but not more. In other words, if the sources aren't there, then we can't write the article, no matter how much we want to.

The paper that originally introduced the chain-of-thought technique is a primary source for that technique and should not be heavily relied upon, either.

Believe me, I'd benefit if the policies and guidelines said otherwise. I could probably wring out two or three articles about my own research by using a peer-reviewed publication as the basic hook and then arXiv preprints to flesh out the details... but sadly for me, that's not the way things go. XOR'easter (talk) 18:06, 18 January 2025 (UTC)

"…the concept of prompt engineering is not even particularly controversial in the scientific community…" That's not even remotely true, though. I'm deeply confused as to how you reach that impression.

"…was published in NeurIPS 2022 and thus peer-reviewed." The authors didn't follow the standard practice in computer science of updating arXiv metadata to point to the peer-reviewed version; all the same, it's on me for missing. That said, it's not clear which version of the paper was submitted to NeurIPS, nor to what extent the specific claims in that paper were subject to peer review. To that end, review in NeurIPS is more heavily tilted towards the basis of originality and expected impact than traditional review, owing in part to the demanding schedule that reviewers are given to evaluate dozens of talks.

That leaves the evidentiary basis for the extraordinary claim made by this article, namely that prompt engineering is real and has a scientific basis, resting on a conference paper written entirely by authors with a significant financial interest in the outcome of those claims. From reading WP:MNA, I don't think it's at all necessary to pretend that prompt engineering is scientifically valid, given the sparsity of evidence to support that. cgranade (talk) 19:25, 22 January 2025 (UTC)

I tagged this for questionable source reliability because of all the arXiv cites. Unreviewed preprints are equivalent to blog posts. Do we have anything more solid that can be used here for claims currently cited to preprints? - David Gerard (talk) 10:41, 24 October 2024 (UTC)

The quality of AI research papers is, at best, wildly uneven. For this page (like any other AI page) I think it's extremely important to restrict to peer-reviewed publications and secondary sources; material relying only on primary sources or preprints should just be removed. Gumshoe2 (talk) 19:54, 18 January 2025 (UTC)

I agree. Doing that would just be living up to the standards we try to apply across the board, and there's no reason to slack off on those standards here. XOR'easter (talk) 00:17, 21 January 2025 (UTC)

Yes, I agree. Anything here should sourced to multiple reliable sources. Anything sourced to unreliable sources and anything sourced to just a single source can be removed. Asparagusstar (talk) 16:23, 21 January 2025 (UTC)

I've deleted some subsections with poor sourcing. Just a start, there are still many problems. Gumshoe2 (talk) 16:48, 22 January 2025 (UTC)

Good removals. XOR'easter (talk) 18:06, 22 January 2025 (UTC)

What we need, at a minimum, is a secondary source like a review article that has passed a referee process in a reputable journal or conference (not MDPI trash or a WASET Potemkin meeting). XOR'easter (talk) 02:26, 25 January 2025 (UTC)

Having a CoT is used more and more in "reasoning" or "thinking" models trained with reinforcement learning as in DeepSeek-R1 & R1-zero, and heavily suspected in OpenAI o1, o3, etc., and Gemini 2 thinking. This is distinct from the prompting technique, as this isn't included in the prompt or pre-prompt, but is part of the model's output, shaped by its loss function. Describing this technique would maybe require creating a 'Reflection (Artificial intelligence)' article (or similar) where this kind of multi-pass CoT, alongside recent techniques purely in latent space, could be described? Or maybe this is already covered elsewhere? — Preceding unsigned comment added by Hplotter (talk • contribs) 13:32, 4 February 2025 (UTC)

This is becoming notable, I guess there could be an article on the concept of reflection. One detail, avoid the uppercase: "Reflection (artificial intelligence)", not "Reflection (Artificial intelligence)". May be worth also updating the article automated reasoning. Alenoach (talk) 00:14, 5 February 2025 (UTC)

This idea has been on my mind for six months, and I was planning to write a similar proposal today, but you beat me to it. Please, let's work on the draft together: Reflection (artificial intelligence) - Wikipedia TheTeslak (talk) 09:55, 5 February 2025 (UTC)

Thanks for your inputs @Alenoach @TheTeslak and creating the article, I'll comment directly there. Hplotter (talk) 15:41, 5 February 2025 (UTC)

It looks like the NPOV template was removed, but none of the POV issues have been addressed. This page still seems to take the strong and non-consensus stance that, rather than being a pseudoscience, prompt engineering is an actual and valid avenue of scientific research. It is not, that view is a highly biased one, and the content as stated on this page is extremely dubious as a result. I don't want to engage in an edit war by re-instating the template without further discussion, but I would strongly suggest that the neutrality issues be fixed, possibly by removing the article entirely in favor of one that discusses the pseudo-scientific belief rather than adopting it; the opening paragraph to Creationism may be an interesting example to follow in doing so. cgranade (talk) 20:59, 18 February 2025 (UTC)

I concur and have restored the tag - David Gerard (talk) 01:35, 10 March 2025 (UTC)

The idea that LLMs are "Stochastic parrots" is not NPOV and does not represent the scientific consensus. It is also unclear what is a Stochastic parrot. The extraordinary claim here is that prompt engineering is a pseudoscience which seems to be entirely the opinion of you @Cgranade who seems to believe prompt engineering is comparable to Creationism, and that the pseudoscientific status is anything more than your own opinion.

While I can prompt engineer with locally running LLMs and see the effect directly, I still await the day when I can ask God about how He created mammals. pony in a strange land (talk) 02:22, 14 March 2025 (UTC)

the idea that LLMs understand anything (which is the opposite of a claim of stochastic parrots) is dramatically less representative of scientific consensus, however. As such I don't think you're approaching the topic in good faith User:LittleBitMoreKSP (GreebleNeebleGombleWomptus???) 03:27, 27 April 2025 (UTC)

If LLMs couldn't understand anything, it would be easy for experts to craft a private benchmark on which LLMs don't perform significantly better than random. But LLMs performance on benchmarks whose data is original and unavailable on the internet has been steadily improving since 2021 when the stochastic parrots hypothesis was introduced.

It's unfortunate though that there isn't a poll of experts on the topic. My impression is that the stochastic parrots hypothesis is popular among AI ethicists, but not among AI researchers. Alenoach (talk) 14:00, 27 April 2025 (UTC)

This disparity could in theory be due to difference in the usage of the word "understand"-- what does it mean for an LLM to "understand" anything? What definition of "understand" is most applicable with LLMs?

For me personally, if "understand" is used in the sense I typically use it (where we reference a sentient actor comprehending and processing the context and details of a topic before trying to perform actions related to them), LLMs do not do this, even if we ignore the "sentient" clause, they do not perform any comprehension or processing prior to writing their responses-- they write on-the-fly, word by word, without prior abstract thought or processing. So far, anything trying to emulate understanding (such as Reason features of recent chat LLMs like ChatGPT or DeepSeek) involve simply writing notes at itself for some period of time using this same on-the-fly system to work through a problem more reliably. I believe this is the most common usage of "understand" in this context in the greater human populace, but I have no sources to back this up.

However, if someone used a definition of understanding such that it did not rely on these internal conscious processes, then it depends, and with the right definition of "understand", could very well be said that LLMs can "understand". User:LittleBitMoreKSP (GreebleNeebleGombleWomptus???) 16:01, 27 April 2025 (UTC)

There are indeed two main ways to define understanding. One is related to capabilities and how information is processed. In this sense, understanding a concept is about how efficiently you process information related to this concept. The other definition is perhaps less common but used in the Chinese Room thought experiment. It is focused on the feeling of understanding, and is related to the notion of sentience. I think the stochastic parrot hypothesis was originally related to the first notion, by stating that LLMs don't really process semantics but just make some superficial pattern matching. But over time, interpretations of the hypothesis may have gradually shifted toward the second definition. At least, that's my understanding of it.

But anyway, the discussion with cgranade was mostly about whether prompt engineering improves in practice the results of LLMs. cgranade said that prompt engineering is a pseudo-science. When searching on Google, I don't see mainstream media articles calling it a pseudo-science. But if there is some well-reported doubt on its efficacy, we could add a few sentences on it. Notably, some techniques designed for earlier models like GPT-3.5 may not be so efficient with new models like GPT-4.5 or o3 (for example, models are now trained to "think step by step" by themselves for complex problems, without needing a special prompt by the user). And it is true that the idea of prompt engineering becoming a job has sparked enthusiasm at some point, but is now regarded with skepticism, as these techniques are sometimes less efficient with (or integrated in) newer models. Alenoach (talk) 17:09, 27 April 2025 (UTC)

AI researchers who work actively in the field are under an obvious conflict of interest, as their funding is benefiting from the current AI hype. They aren't going to actively shoot their funding in the foot right now. Many of us who have actually studied the computer science of LLMs understand them to be exactly what the stochastic parrot paper describes. Jgiesbrecht (talk) 04:51, 5 May 2025 (UTC)

Issues I'm seeing mentioned above in other posts relating to point of view, source overkill with tech company articles and prepublished papers with underreporting from standard Wikipedia sources, and other related issues may have a root cause in the form of lack of notability. Can we review whether this article meets notability requirements? Though there exists a lot of sources, much of it is primary, potentially biased, preprints, or poorly verifiable, with relatively few articles posting reliable secondary information on prompt engineering. Its very notability seems to be subject to bias-- whether prompt engineering is a thing that works and deserves the title of "engineering" has a lot of discourse over it despite its relative inconsequentiality. Could we roll this into other machine learning and LLM related topics, deleting the original article in the process?

An argument that could also be made for removing or merging this article is the fact that the hype over generative AI as a whole seems to be relatively transient (and is decreasing over time), and this subfacet has had less and less ability to stand on its own as time has gone on. I'm not sure its notability will stand the test of time and be a relevant and notable topic for the forseeable future, or if we as a civilization fell for a dot com bubble style hype train. Is it too early to make this call?

I'm not very good at Wikipediatizing, so I'm not sure if this is the right course of action, and if it is, I'm not certain that I know how to correctly propose an article for deletion or merging. Let me know of your thoughts, and, if you're bolder than me and more confident than me, I think proposing it for merging or deletion may be beneficial. User:LittleBitMoreKSP (GreebleNeebleGombleWomptus???) 22:26, 5 May 2025 (UTC)

It's easy to find mainstream news articles on prompt engineering, and there are Wikipedia articles on the topic in 31 languages. It's very unlikely that the topic would be considered not notable enough for an article. Alenoach (talk) 05:46, 7 May 2025 (UTC)

@Alenoach Currently a very small fraction of the sources cited here are the actually Wikipediable secondary unbiased sources that you mention, so I don't think this works until they're found and added to the article. Most of the sources here are prepublished (i.e. not yet peer reviewed) or primary sources from AI researchers and AI tech companies, which may exhibit biases, are typically primary sources (which, while not a sin, are not something to over-rely on when discussing notability), and often have a conflict of interest. As for the article existing in 31 other languages, I think this isn't really a great supporting evidence here on its own, and doesn't convince me given my concerns of recency bias in the reporting of prompt engineering (which I'm happy to see disproven because it is just a hunch on my part). User:LittleBitMoreKSP (GreebleNeebleGombleWomptus???) 14:44, 7 May 2025 (UTC)

is there a general name for this sort of technique? Arlo James Barnes 02:43, 8 November 2025 (UTC)

It's a kind of "prompt injection" or "jailbreak". I don't know more specific terms for this technique. Alenoach (talk) 12:24, 8 November 2025 (UTC)

There's a new article at Program of Thought Prompting that seems to replicate the 'Chain-of-thought' section here, or at least considers a subset of it. I think that there's space to merge it in, particular once duplication is considered. Klbrain (talk) 21:34, 14 March 2026 (UTC)

Support - the concept of "Program of Thought Prompting" does not seem notable enough to have its own article, though merging would require removing most of it to avoid making the "Prompt engineering" article bloated. Alenoach (talk) 22:15, 14 March 2026 (UTC)

I WP:BLARed it. SuperPianoMan9167 (talk) 02:06, 18 March 2026 (UTC)