Gains:
- Ability to find literature from real databases and use artificial intelligence to summarize, translate and correct only the actual text at hand
- Ability to independently verify each citation by DOI or title and eliminate fabricated sources
- Ability to preserve the meaning of technical terms and ambiguities in language correction and translation
No study of physics stops in a vacuum; Each conclusion builds on previous work. That's why literature review—finding, reading, and properly citing published articles on a topic—is the backbone of scholarly work. Artificial intelligence (AI) greatly accelerates this process: summarize a field, simplify a complex article, translate a text into English, correct a draft. But this is where AI's most dangerous mistake comes into play: made-up attributions. In this unit you will learn how to use AI in literature and scientific writing and how to independently verify every source, every citation, every fact. The main message of this unit is one: AI speeds up text; but no attribution is reliable without independent verification.
Why does AI produce fake resources?
A large language model does not depend on an actual database; It produces "real-looking" tags by mimicking the citation patterns it sees in the training data. The result: a properly written article, with believable author names, a real journal name, and a reasonable year, but a completely non-existent one. This has become a serious problem in the academic world; A false citation entered into a bibliography without verification both invalidates the work and violates scientific integrity. So use AI not to find literature, but to understand and organize the actual literature you find.
Quest
Is AI safe?
verification
General summary of a field
Partially (fact confirmation required)
Textbook / review article
Find/cite specific article
NO (makes up)
Search real database (by DOI)
Summarize the article you read
Yes (if you provide the text)
Compare summary with main text
translate text
Yes
Check technical terms
Language/style correction
Yes
Check that meaning is preserved
Bibliography formatting
Yes (with real imprint)
Map each field to the original source
Step by step: an honest literature and writing process
1. Find literature from real databases. Find articles from peer-reviewed databases (your institution's access, academic search engines). Note the DOI (Digital Object Identifier) of each article; DOI is the most reliable proof of the authenticity of the article.
2. Use AI to understand the actual text you find. Give the real article you have to the AI and summarize it and simplify a complex section. Don't tell the AI to "find me an article on this topic" — most of what it finds may be made up.
3. Independently verify each attribution. If the AI suggests a source, search for that citation by DOI or title in a real database. If it can't be found, that source is probably fake — don't use it.
4. Confirm the facts from the original source. When the AI says “that experiment found that result,” verify that claim by reading the actual article. AI can also misrepresent the conclusion of a real paper.
5. Maintain meaning in your writing. The AI can replace a technical phrase or an uncertainty phrase (“likely”, “tends”) as it streamlines your text. Compare the corrected text with the original meaning.
Tip: The surest way to test the authenticity of a citation is to check the DOI. Search for the DOI of a citation provided by YZ in a DOI analyzer; If it goes to a real article, it exists, if it says "not found" it is most likely fake. Never trust an AI citation that does not give a DOI or gives the wrong DOI.
three mini cases
Case 1 — Fake article in bibliography. A student asked YZ for resources for his thesis introduction; YZ submitted three articles with credible authors, with the name of a well-known journal. The student searched the titles in an academic database: two out of three did not exist at all. Without verification, fake sources would enter the thesis and create a serious integrity problem in the thesis defense.
Case 2 — Right article, wrong summary. A researcher had the AI summarize a real article, but gave only the title, not the text. The AI made up a “plausible summary” without ever seeing the article; When the actual article was read, it was seen that the result was exactly the opposite. Lesson: the summary is reliable only if the main text is given and compared with the main text.
Case 3 — Translation term error. A teacher had AI translate a summary into English. The AI produced fluent text, but instead of "uncertainty" it used a word that didn't fit the context and mistranslated a technical term. The teacher checked and corrected technical terms; The translation became available only after this inspection.
Four copyable templates
1) Summarizing the actual text:
Below is the text/summary of the ACTUAL article I read. Simplify it so that the main finding, method, and limitations are clear. Do not add any results or numbers that are NOT in the text. Mark the part you are not sure of as "not clear in the text". Text: [here]
2) Citation verification discipline:
Evaluate the following attribution: if you are not 100% sure of the authenticity of this article, "verify this attribution with a DOI in a database" DO NOT MAKE IT. DO NOT produce a DOI or byline that does not exist. Imprint: [here]
3) Scientific language correction (preserving meaning):
Improve the language and fluency of the following scientific text, BUT: do not change technical terms, numbers, and expressions of uncertainty ("possible", "trend"). Keep the meaning literal. Mark the places you changed so I can check. Text: [here]
4) Technical translation + term control:
Translate the following physics text into [language]. At the end of the translation, provide me with a list of the key technical terms you used, with source-target mapping, so I can check its accuracy. Tick the term you are not sure about.Text: [here]
Weak prompt / Strong prompt
Weak: "Give 5 academic sources and their citations on this subject."
Result: AI produces dog tags that look real but are probably fake; Violation of scientific integrity if used without verification.
Strong: "Below is the text of the actual article I read; summarize its main findings and method without adding anything that is not in the text, mark the ambiguous parts. Also correct the language of my introductory paragraph, without changing technical terms and expressions of ambiguity."
Result: A reliable work based on real sources, free of fabrications, and with preserved meaning.
Common mistakes
- Finding resources for AI. AI is not a real database; The attributions it produces may be fabricated. Find sources from real databases.
- Using citations without verification. Any citation that is not confirmed by DOI or title is risky; a fake source clouds the entire work.
- Requesting a summary without giving the text. The AI “plausibly” makes up an article it has not seen; The summary is reliable only with the main text.
- Not confirming the fact from the original source. AI can even misrepresent the conclusion of a real paper.
- Not checking the meaning in translation/editing. AI may change the technical term or ambiguity expression for the sake of fluency; meaning must be preserved.
Attention: Just because a citation has a real journal name, real-looking authors, and a reasonable year does not mean that the citation exists. This is precisely why fabricated attributions are believable. Before including a source in your work, be sure to find it in a real database and confirm it with DOI. An unverified citation is a violation of academic integrity, and responsibility rests entirely with the author — not the medium.
In summary
In literature and scientific writing, AI is a powerful accelerator in understanding, summarizing, translating, and language correction of actual text. But it is dangerous in sourcing and generating citations: it produces articles that are credible but do not exist. The safe discipline is clear: find literature from real databases, use AI to process the real text you find, verify every citation with DOI, confirm every fact from the original source, preserve meaning in translation and editing. In the next unit, we will bring this discipline of verification to teaching physics—course material and misconceptions.
Application task
Choose an actual physics article or textbook chapter you have. Summarize it by giving the text to the AI with the 1st template; Compare the summary with the actual text and see if the AI added/twisted anything. Then deliberately tell the AI "suggest sources on this topic" and search for the citations it suggests in a real database/DOI; How many actually exist? Note in 5-6 sentences: was the summary faithful, did the suggested citations seem contrived?
checklist
- [ ] I found the literature from real databases, I did not let AI find it.
- [ ] I gave the original text to YZ while summarizing it.
- [ ] I have verified each attribution with a genuine source by DOI or title.
- [ ] I have verified each factual claim from the original source.
- [ ] I checked the translation/editing for technical terms and ambiguity.
- [ ] I did not include any unverified sources in my work.