I had similar experiences; ChatGPT is basically a extremely fancy 'suggest next word'-bot such as found on smart-phone keyboards. But because scientific writing is based on extremely detailed subject knowledge you hit the limitiation of ChatGPT very quickly. A general writer of blogs or websites can work with the more ambiguous, vague, or 'good enough' writings from ChatGPT. But because ChatGPT is not a knowledge base it makes errors, invents sources, etc. in a way that makes it useful for general outlining only. What ChatGPT does very well is what you show later in the video, taking on the role of 'co-author'. If you give it a draft text and ask it to revise it for clarity and brevity, you get some very great edits of you text.

One time I asked it to write a few paragraphs about statistical discrimination with references, and it got it so correct. Some other times it shows the same message you showed in the video.

Just wanted to say that this was an excellent post. I am an academic and write papers all of your points. Thank you.

Thank you very much for this great video which was very informative! I think the current limitation of ChatGPT is to use general information rather than detailed information from research papers. I think that in the future (not far from now), we can imagine that we provide a bunch of key papers to ChatGPT, we give it time to "read/digest" all papers and later on, ChatGPT can provide detailed references and comparisons about these papers. As a researcher, I think that it is only the beginning and ChatGPT can really revolutionalize the research community in the coming years.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Yes. I think you may be right about that.

@HimmelReiner

Жыл бұрын

I wrote already two papers for last week and submitted to top tier journals. Also wrote a book chapter and few reviews for research papers from journals. Now I have more time to write comments

@theobserver8881

Жыл бұрын

Revolutionise? Or take over?

@husseinjawad9298

Жыл бұрын

​@@HimmelReiner please can you help me about the citation of references that chatgpt provide, because all the references it provide is not correct.

I used chatGPT as an "editor" for a couple of awkward sentences in my manuscript (it will be my first and I've been having the grossest writer's block of my life). It made some good suggestions, but if anything they were only useful as examples of how to possibly reword the sentences. That writer's block was brutal. It's been a year since we started writing. Just recently discovered your channel in an attempt to get my *s* together. Looking forward to binging your content!

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for commenting. Good luck with your manuscript. I hope my videos help.

Dr. Karen, you are setting an example for early-career researchers. Thank you very much!

@ScientificWritingwithKarenLMcK

Жыл бұрын

I’m glad you think so!

You've done a marvellous job of the updating us! Thank you

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks!

Great experiments. Thank you!

Google is working on a model called "Sparrow" capable of providing citations. It is expected to go public between March and May. ChatGPT may also have improved similarly by that time.

Thank you for this very helpful and informative video!

Excellent work. Thanks much.

Thank you, this is the video which I was looking after reading about ChatGTP in news paper recently

@ScientificWritingwithKarenLMcK

Жыл бұрын

Glad it was helpful!

This is very educational and informative . Love from Kenya 💖💖💖💖

Very informative video ,thank you.

Thank you, dear. Learnt something

Thanks for this video👏

Thank you for introducing great resource

Thank you for the detailed information Dr. Karen

@ScientificWritingwithKarenLMcK

Жыл бұрын

Glad it was helpful!

Thank you for a great demonstration!

@ScientificWritingwithKarenLMcK

Жыл бұрын

You're very welcome!

Awesome video!

Thank a lot Karen. This is really helpful.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Glad it was helpful!

I guess it will be very beneficial to researchers whose mother tongue is not english. Additionally,I think it will create, huge momentum in academia. I believe using it will allow scholars to finish their research paper in no time ( a bit of exaggeration) but compared to what was used to be followed to write a paper. At the same time, journals may have to decrease their acceptance rate and put higher restrictions on the received manuscripts. As the number of people submitting papers will start increasing after the launching of ChatGPT. Another advantage also will be related to focusing entirely on the research. And the person will not have to worry about writing nor how long it will take to write different drafts. Looking at the whole thing from another perspective, this CHATGPT may cause a decrease in the profit of companies who work in the fields of plagiarism and writing check software. As people will mostly not buy these software as from now on, they will be capable of writing correct English. Thank you for such a great video.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for your comment.

Thanks for sharing

Another great video!

@ScientificWritingwithKarenLMcK

Жыл бұрын

Glad you enjoyed it!

Excellent!!!

If you ask for references in the initial question I found it would often list them. Also if you ask again for references in a different way maybe directly referring to the original passage it might work as well.

@marcfleischmann269

Жыл бұрын

I had the problem that the links for the references it listed often would not work.

@olavihekandjo2928

Жыл бұрын

@@marcfleischmann269 that because it makes up examples of references and doesn't actually give you its resources.

Starting to feel like you are the co-supervisor I have never met. Thanks for another good video. I am currently writing up the introduction for my thesis, but I think I will pass on help from ChatGPT, as I am worried the generated text will show up in a plagiarism detection engine at some point ;-)

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks!

@ScientificWritingwithKarenLMcK

Жыл бұрын

I asked ChatGPT if it is plagiarism to use A.I.-generated text and got this reply: "It is generally not considered acceptable to use the text generated by ChatGPT or any other language model as your own without proper attribution. Doing so could be considered plagiarism, which is the act of using the work of others without giving proper credit. If you want to use text generated by ChatGPT or any other language model in your own work, it is important to properly cite the source and make it clear that the text was generated by a machine. This will help to avoid any misunderstanding or confusion about the origin of the text and ensure that you are giving proper credit to the creators of the language model. In general, it is always a good idea to be careful about using the work of others and to give proper credit when it is due. Plagiarism can have serious consequences, including academic penalties, loss of reputation, and legal action."

@BRIANIGANI

Жыл бұрын

This lady is simply breathtakingly amazing. She has laid the foundation for me concerning writing papers. And as for chatGP, hmm

@alejandrarodriguezsanchez6667

Жыл бұрын

you are touching on the most important aspect of this technology. will people be sued by the developers of chatGPT for plagiarism?

Yeah I already tried this many time ✌️ amazing

Karen, fantastic presentation. I do a lot of work with science writers, particularly helping individuals with their NIH and NSF grant submissions, and I provide both commentary and editorial suggestions. In that sense, I function as a "non-artificial intelligence" in this role, as do colleagues and collaborators, as you note at the end of your vid. Having ChatGPT serve a similar role, where you enter text YOU have written and ask it for corrections and suggestions, seems to me a perfectly acceptable way to use it, and certainly not as ethically troubling as asking the engine to write your paper for you. Besides being unethical, having the AI write for you is also hazardous, because it can be very very wrong. However, I think writers who use ChatGPT as a "copy editor" should ALWAYS ask it why it made the changes it did, and to explain the principles behind those changes. In that role, it can serve as educator -- assuming, of course, that it is right. Finally, you have a new subscriber!!

@ScientificWritingwithKarenLMcK

Жыл бұрын

Karl, thanks for your thoughtful comment. Good suggestion about asking ChatGPT for the principles behind suggested changes. Thanks for subscribing!

great video

Great video, Karen! Guess with SciSpace, we might already be heading in this direction. They call themselves the ChatGPT for research.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for commenting!

Do you know that at the end of their response… You can put one word: “continue” and it will continue on usually for two or three continues that will fill even additional information in that you haven’t asked yet…?

Great video. I wanted knew there were limitations on the use of chatgpt for research, but didn't this it was this bad until I washed this video. One has to wonder if this is actually artificial intelligence. More impressed by Midjourney.

I just take help from chatgpt for structuring my research which is pretty good. The point here is to be specific with the choice of words we ask

excellent

Thanks for sharing. By the way, here is a peer reviewed paper which has been published by ChatGPT as a second author: "Open artificial intelligence platforms in nursing education: Tools for academic progress or abuse?"

@ScientificWritingwithKarenLMcK

Жыл бұрын

Interesting. I'll check it out.

@geene3

Жыл бұрын

Its an editorial article, its not original research

Thank you for ur video. So, can we using chatGPT in publishing articles in departments engineering , review , and PhD students? Thank you again

Great video 👍🏼 I agree with you, the inability for ChatGPT to cite articles is a big no no. The ethics involving the use of AI in one’s work will be interesting. For now, I think chat GPT is good for generating possibilities and ideas, though one will have to use his or her own judgement on the responses.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for sharing your opinion!

@michelle.coates

Жыл бұрын

It can cite. Just tell it to and how many sentences each reference.

@harryhahne

Жыл бұрын

@Michelle Coates unfortunately many times when I press it for sources, Chat-GPT makes up citations to non-existent articles. It attributes a viewpoint to scholars who explicitly rejected that view in writing. You have to fact check so much of what it writes

@angiem9116

Жыл бұрын

@@harryhahne I have checked all of the references ChatGPT gives me, and none of them have been accurate. I only use AI to summarize articles so I understand better the content. I paraphrase in my own words if I want to make sure what I say is really in the article.

nice video, good challenging of the AI, so as you can see, since you are a smart person you can see how limited is this AI an not at all so powerful how people mean it. It is tho a great help for people who need to focus more on data rather than fancy writing.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for commenting. I think each person must determine for themselves whether AI tools are a help to them for writing or other purposes. For example, a non-native speaker of English may be helped by ChatGPT to improve their writing skills.

Thanks

One can always use outside websites to check for plagiarism or similarity with other authors. Newer AIs interface with more powerful current data generators and analytics such as Wolfram Alpha.

Very useful video

@ScientificWritingwithKarenLMcK

Жыл бұрын

Glad you think so!

Is it safe to use it just to paraphrase my own words because Turnitin can catch AI text? The thing is I don't speak English that well, so I use ChatGPT for paraphrasing my own text into something more academic. Will I be caught by Turnitin in this case?

Karen - I enjoyed this demonstration of what ChatGPT can and cannot do. Thanks very much for sharing the results of your experiments with it. One aspect which bugged me was (circa 10:30 onwards) when you asked it to reword your findings on the annual variation of Nitrogen concentration. It seemed very keen on "fluctuated" - rather than "varied" (for example). Fluctuation suggests some noisy ups and downs to me, whereas your initial wording conveyed to me a systematic decrease from a high in January to a minimum in mid July and then increase from August to December. That reads more like an annual cycle? Cheers.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for your comment. I’ve also noticed ChatGPT favors certain words and phrases. However, I think fluctuate means to go from one level or state to another- up or down. So its use here does not seem wrong to me. Also, to claim an annual cycle really requires more than one year of data-although that may be the case here. You’re right, though, to scrutinize word choice and be sure it’s what you really mean.

@tassiedevil2200

Жыл бұрын

@@ScientificWritingwithKarenLMcK @Scientific Writing with Karen L. McKee Thanks for your reply. Re "fluctuate" - I had noticed that you didn't disapprove of that choice. I guess we differ on the appropriateness in this context - for example both temperature and day-length vary through the year, but I'd not apply "fluctuate" to the latter. I admit that it might be a (rash?) leap for an AI (or a scientist) to infer an annual cycle here. When you asked it to describe the results with less explicit mention of months it recognised that it could comply by dropping the February and March information, and if it had used "varied" I suppose that I would have interpreted "reaching a high in January and a low in mid-July" the way I did the original - as implicitly reporting a progressive decrease from January to July, whereas with "fluctuations" I'd tend to wonder if it was simply reporting extremes of a noisy sequence. In both cases, I realise I'm making speculative inferences based on assumptions about what was not said: e.g. in the original narrative that if the data for April through June did not fit the narrative of the decrease over January to March then this would have been remarked on. Of course, in reality there might be missing data etc. ChatGPT is being more cautious, and (strictly more correct given the "data" provided) to not offer "decreasing from a high in January to a low in mid-July, " to match what it was told about the "increase from August to December". It might be interesting to see if it would make that more informative statement if also told it that concentrations fell in April and May and June. Getting off-topic (away from paper writing) it might be interesting to see how ChatGPT responds if asked to suggest factors that might cause these variations. Cheers. You might enjoy the KZread by #coolworlds of ChatGPT taking a College Astrophysics Exam.

@ScientificWritingwithKarenLMcK

Жыл бұрын

I do use "fluctuation" as you do--to mean an irregular or unpredictable variation in something. The example I gave was not entirely clear on this point, so I didn't want to unfairly ding ChatGPT for using "fluctuation". An author could easily revise the chatbot's output to use "variation" if that fits the observed pattern better than "fluctuation".

@tassiedevil2200

Жыл бұрын

@@ScientificWritingwithKarenLMcK Point taken. I also agree that a critical reader could easily adapt the chatbot's output. Its efforts could also spark new lines of research - particularly for one with a relevant background knowledge. (Pasteur's: "chance favours only the prepared mind"?). My last comment was also partly on the point that I would have been more impressed if ChatGPT were to infer/propose (or ask?) whether the signals January to July showed a steady decline - that sort of inferential reasoning seems a bit beyond it? As I commented, it did very efficiently drop references to specific months as asked - but (and as I said, it is just my reading) by using fluctuate instead of vary, I felt it lost some information that the researchers would have preserved in that edit. I can imagine that with the wide range of things it "knows", ChatGPT could probably come up with lots of correlation-based suggestions about the cause of the variations in N concentration (provided it knew which hemisphere the research site was in) but I think that's still rather mechanical correlation, not inference. I've enjoyed the conversation.

I wrote a comment last night, but don't see it. I was able to elicit legitimate citations by prompting chatgpt to write a intro paragraph about ____ following the principles of scientific writing. Then I asked ti to use the apa style to add citations and it did just that along with lit cited. There were only 3. You definitely have to check everything it writes but it can be used to accelerate a first draft to great effect

The answer is "no" ... BUT... it can and will definitely help and support you to write much better papers. Because most scientific people are not necessary excellent writers, the process of writing a good top paper is a slow, painful, and exhausting one. Yet it's rewarding because it helps ordering ideas, seeing more clearly what we're doing, and the pain stops once the paper is accepted and read. ChatGPT is a little bit like that amazing bunch of advisors and directors we wished we had at hand 24/7 to advise us on the hundreds of insecure little questions we have when writing. The only negative thing, ChatGPT might take away the natural process of "maturity" that struggling brings to a growing scientist.

it helps overcome writer's block

Dear Karen. You can ask chatgpt to provide the refererences of a setence ( It provides the citation though)

@ScientificWritingwithKarenLMcK

Жыл бұрын

You are right. ChatGPT will provide its source for individual sentences. However, the cited sources are not primary sources (i.e., peer-reviewed journal articles), but instead secondary sources such as NOAA, USGS, or Encyclopedia Britannica web pages with general information about various topics. You would not want to cite such secondary sources in your scientific paper. For example, when I asked: Can you give a reference for this sentence: "In addition, mangroves help filter pollutants and excess nutrients out of the water, improving water quality?", I got this response: "Here are a couple of sources that support the statement that mangroves help filter pollutants and excess nutrients out of the water, improving water quality: 'Mangroves are known for their ability to filter pollutants, excess nutrients, and sediment from the water, improving water quality.' (Source: United States Geological Survey, www.usgs.gov/centers/pacific-coast-science-center/science/mangroves-and-coastal-wetlands-protection-and-restoration) 'Mangroves have an important role in maintaining water quality by filtering out pollutants and excess nutrients.' (Source: National Oceanic and Atmospheric Administration, oceanservice.noaa.gov/facts/mangroves.html The USGS link is broken, and the NOAA site provides no statement about filtering nutrients and pollutants, just a generic description of a mangrove forest. When I ask for a summary of a journal article on the topic, ChatGPT responds: "I'm sorry, but I'm unable to access specific research papers or other external sources of information."

@allandueck3256

Жыл бұрын

Agree with Karen on this. GPT is trained on the common crawl, a vast collection of webpages, so it reflects the general knowledge of the internet very well. It constructs its arguments based on linkages between words. It is not specifically referring to any one of these pages when it produces a response. It’s citations are derived the same way. Occasionally it will cite real research in a field you are digging into because it has a high probability of being linked to the content but it does not have the context on if that research is confirmatory, disconfirmatory, or a something sensationalist written by someone outside of the field altogether (false balance in AI emerges). If this AI is trained on research papers we would have a great resource,p that would kill many a publishers business model, but for now, gpt is helpful for general knowledge, not evidence based authorship. As an example, it’s explanation for how lift is generated by an airfoil aligns with a widely circulated explanation, but is, from a scientific perspective, strictly wrong.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Yes, ChatGPT can summarize well-known research papers such as Watson and Crick's 1953 paper on DNA, but it does so using what others have said about it rather than using the 1953 paper itself. Here is ChatGPT's summary of that paper. "James Watson and Francis Crick's 1953 paper, "Molecular Structure of Nucleic Acids: A Structure for Deoxyribose Nucleic Acid," described the discovery of the double helix structure of DNA. This groundbreaking discovery revolutionized our understanding of genetics and has had far-reaching implications for fields ranging from evolutionary biology to medicine. In the paper, Watson and Crick describe the structure of DNA as a double helix, in which two strands of nucleotides are wound around each other. They proposed that the structure of DNA was determined by the specific sequence of its nucleotide bases, and that this sequence could be used to store genetic information. They also suggested that DNA replication occurred through the separation of the two strands of the double helix and the synthesis of new strands complementary to the original ones. The discovery of the double helix structure of DNA was a major milestone in the field of biology, and Watson and Crick's paper is still widely regarded as a classic in the scientific literature." This is a very general description lacking the structural details found in the paper. When I asked for sources for the last sentence, here is what ChatGPT provided: "Here are some references that support the statement that the discovery of the double helix structure of DNA was a major milestone in the field of biology and that Watson and Crick's paper is still widely regarded as a classic in the scientific literature: "The discovery of the double helix structure of DNA is one of the most significant scientific discoveries of the 20th century, and its impact on biology, medicine, and other fields has been immeasurable." (National Human Genome Research Institute, "The Double Helix: A Turning Point in the History of Genetics," www.genome.gov/the-double-helix) *This link is broken.* "The discovery of the double helix structure of DNA is one of the most important scientific discoveries of the 20th century, and it has had a profound impact on our understanding of genetics, evolution, and medicine." (Discovery Education, "DNA: The Double Helix," www.discoveryeducation.com/science/dna-the-double-helix) *This link is broken.* "The discovery of the double helix structure of DNA is one of the most significant scientific discoveries of all time, and Watson and Crick's paper describing the structure is a classic in the scientific literature." (Encyclopedia Britannica, "James Watson," www.britannica.com/biography/James-D-Watson)" *This link is broken.*

when we verify the provided refrences we find then virtual , they do not exixt

It can provide citations and references (even compliant with any style you would want) but it is a bit tricky

@sadaf3945

Жыл бұрын

How

@firefly5819

Жыл бұрын

I also want to know how 😢

I have watched this video a few times to make sure my comments would be reflective of the topic at hand. Prompt engineering would give you better results. learning to write better prompts would give you better results the tool is only as good as the information given to the AI. I am sure you are well versed in your given field but this is an emerging technology that can be neither good or bad but dependent upon the user.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for your comment. I wanted to show other scientific writers what I experienced with ChatGPT. Few of them are likely to be "prompt engineers" and, like me, will need to learn how best to use this technology for writing, if at all. You are correct that technology is neither good nor bad, but there will still be ethical questions about how that technology is used.

I don't quite know how to react to this. I fear that this is going to make it very hard to evaluate writing submitted by students to meet a course requirement..

It created references for me! I asked for a paper on the effects of dams on sediment transport. Then i asked for references and it provided ten. Then i told it that they should be within the text. It apologized and rewrote the same text with the references incorporated.

@ScientificWritingwithKarenLMcK

Жыл бұрын

You're right. If you ask for a paper on a topic, ChatGPT will provide one. However, if you ask for ten peer-reviewed journal articles (on how coastal wetlands respond to sea-level rise), it says it can't because it's "too time consuming" and suggests using a search engine such as Web of Science, Scopus, JSTOR, and Google Scholar. Depends on how you ask the question.

Good afternoon. I think you need to create context with the AI before write your questions. This interaction can give you better answers.

All you need are the exact prompts to produce exactly what you need. You can even ask for variables to be included, sample data, statistical formulas, and how you will solve them step-by-step. ASK AND IT SHALL BE GIVEN. Soon, people will hire people who are called PROMPT ENGINEER.

@MMABeijing

Жыл бұрын

prompt engineer, you nailed it

cannot get in-text citations Another AI provided citations and references

Does this app follow OCAR sentence elements?

@ScientificWritingwithKarenLMcK

Жыл бұрын

Here is what ChatGPT says in response to your question: “I see. "ChatGPT" is the nickname for the GPT-3 model that was fine-tuned for chatbot-like text generation. It is not specifically designed to follow any particular story structure, such as OCAR. Instead, it is a general-purpose language model that can generate text in a wide variety of styles and formats, depending on the specific prompts and context it is given. So, it would be possible to use ChatGPT to generate text that follows the OCAR structure, but it would not do so automatically or inherently. You would need to provide specific prompts and examples to guide ChatGPT in generating text that follows the OCAR structure. Does this help clarify things?”

What if there's an specific science chat, with only papers and books?

@ScientificWritingwithKarenLMcK

Жыл бұрын

I think it will still require a lot of human input because chatbots cannot evaluate the validity (or relevance) of the information they use to generate text. But, who knows what the future holds?

Thanks for this interesting video. It is possible to use chat gpt in other languages? I mean Spanish.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Here’s what ChatGPT says: “Yes, ChatGPT can handle languages other than English, but its proficiency may vary depending on the language. As it is trained on a large dataset of text in various languages, it can understand and generate text in multiple languages, including Spanish, French, German, Italian, and many others. However, its proficiency in a particular language may be lower than in English as it is primarily trained on English text.”

Hello Karen, please any new information concerning ChatGPT since this video? Thanks for your experiment. Very helpful. I also want to know if ChatGPT is accepted for scientific papers? How to reference it ? Please, I’m currently writing my thesis and would like to keep in touch with you via email or phone for more advise Many thanks

@ScientificWritingwithKarenLMcK

Жыл бұрын

Yes. I addressed these questions in this video: Will Scientific Journals Accept Papers Containing AI-Generated Text?kzread.info/dash/bejne/ZHiYztyhZKytqrg.html

Interesting - this seems like a useful tool but you make some good points about being cautious. I wonder if this AI can perform statistics?

@ScientificWritingwithKarenLMcK

Жыл бұрын

ChatGPT can suggest statistical tests if given an experimental design. Here's what it said when I asked if it could perform a statistical test: "Yes, I can certainly help you perform statistical tests on your experimental results. However, it's important to choose the appropriate statistical test based on the type of data you have collected and the research questions you are trying to answer. To perform a statistical test, I will need to know the following information: The type of data you have collected (e.g., continuous, categorical, ordinal) The research question you are trying to answer The number of samples or groups you have Whether the data is normally distributed Whether the samples or groups are independent or related With this information, I can help you choose an appropriate statistical test and walk you through the steps of how to perform the test using a statistical software package or a statistical calculator. Please let me know if you have any further questions."

@jamesorior

Жыл бұрын

Statistics is easiest thing Chat GPT can do. Once this AI is released to the world wide web forget your PhD degree because it will do everything in a peer reviewed paper and you won’t be able know if it wrong or not

Thank you for the informative video. Would you please share about plagiarism of the results generated by ChatGPT based on your experience and expertise?

@ScientificWritingwithKarenLMcK

Жыл бұрын

I provided some thoughts about the ethics of using Chat GPT at the end of the video. In general, if you use Chat GPT-generated text, you need to acknowledge this in some way in your paper, assuming its use is allowed. Some journals such as Science consider text written by Chat GPT to be plagiarized: www.science.org/doi/10.1126/science.adg7879. But other journals may allow it if acknowledged in some way. A rule of thumb would be to consider if you would list a person performing the same service (as Chat GPT) in the Acknowledgements section or if you would cite a tool used in the Methods section. *Always check your target journal's instructions to authors or contact the editor directly about how to do this.* Although a few researchers have listed Chat GPT as a co-author, journals such as Science and Nature say that AI tools should not be listed as co-authors. See this article for more information: www.nature.com/articles/d41586-023-00107-z

We tried out ChatGPT for kicks to see what it would do with graduate level corrosion science. Basically, it talks a good game, but it doesn't really get it. When we asked fairly common questions, it gave a verbose general answer, which sounded right but wasn't completely accurate and couldn't go deeper into the details. When we asked credible-sounding but scientifically implausible questions, it just rolled on generating text as if we were reasonable. It's pretty good at re-wording clunky writing and organizing a cover letter. It didn't root out all the passive tenses and awkward pronouns in our technical sample. It's a very helpful editing tool, but it's got a long way to go to achieve intelligence.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Thanks for sharing your analysis.

Many times it doesn't show correct information. Author name and article title doesn't match.

But it might help us to read a research paper in short time if we get it to summarize the paper.

I have used ChatGPT to look for scientific articles. It has given me tons of articles that surprisingly, don´t exist. I don´t know if you check all the references it gives you, but I do, and none of them existed. Sometimes one or two. Have some of you found the same issue?

@ScientificWritingwithKarenLMcK

Жыл бұрын

Sometimes I get links to online sources or web pages that no longer exist or that are summaries of published research rather than the original paper.

-I don;t see the big difference between the answers provided and those you could get from writing the questions on the google browser! -Giving a title to an essay had been a capability in older MS Word versions from the late 1990s, which gave suggestions of titles with the same efficency.

Chatgpt can't access publications, online books, the internet. Therefore, sadly it is not good for writing research papers.

@ScientificWritingwithKarenLMcK

Жыл бұрын

Yes, ChatGPT can't (as yet) access the primary literature. But as I point out in the video, ChatGPT can suggest titles for papers and help the author polish awkward passages.

@haetienne

Жыл бұрын

As of the recent update yesterday to chatgpt 4 it now has the ability to access the net but they are rolling it out slowly

@PeterPerez.

Жыл бұрын

@@ScientificWritingwithKarenLMcK completely agree with that statement, helps users rephrase words to a more desirable level

Many questions regarding the ethics of AI.. I do t think I am interested.

I donot think that CHAT GPT can write a scientific article which usually has several references and specific info is extracted from those articles. These responses are general and any scientist would throw away such an article. In science you need to cite and mention specific results and discuss them through comparisons with other research and find what has been done, what are the drawback and what more can be done in future. Plus it wont write methods section. Even introduction is very general , it cant even be used to write a review article. so chat gpt can be used to write a general script of youtube video or an artcle that a fifth grader can use no more than that

@ScientificWritingwithKarenLMcK

Жыл бұрын

That's basically what I found and showed in the video. But as several commenters have pointed out, using better "prompts" will get you closer to what's needed in a scientific paper. Whether you should use that output is another matter.

This is just the start. Imagine 5 years from now. Overall it's about productivity. With AI those that use it will be more productive than those that do not.

If you play chess you know what is coming. Give it another 5 - 10 years and it will blow your mind.

@jamesorior

Жыл бұрын

Too long time 3 years at least

Then it will be cut copy paste papers , not new scientific invention papers

Making money is an art, you can only make money if you practice and master the art.... by the way I enjoyed watching your videos. a fellow creator...

Can you write that for a 12 year old as it might be useful for explaining it to members of Congress! :D

not really. i asked it simple questions and it returns me garbage

I found your video quite informative and useful. My issue with GPT3 is the gender and racial bias that it seems to have come with. Perhaps with the growth to GPT3.5 or GPT4 the gender and racial biases can be lessened. However, I am concerned about this as technology is dominated by white, heterosexual males.

In the end - the question of ethics in using AI generated text should be irrelevant in science. Who cares who wrote it. The important part is that the data was collected honestly and reported accurately, and the data analysis was appropriate. ChatGPT (or something similar) should be able to assist and improve scientific knowledge. I think everyone working in a science role knows how bad science is regarding data collection and analysis (especially in academia)

@ScientificWritingwithKarenLMcK

Жыл бұрын

Journal editors and publishers care who wrote a paper.

@michaellawrence3513

Жыл бұрын

I'm sure they do. So does the author - academia is a game of who publishes the most papers, which is part of the reason the system is flawed. AI has the potential to achieve a level of objectivity that science should have, but currently lacks, and focus the scientific community on the actual hypotheses, data, data analyses, and conclusions. From a perspective of society gaining collective knowledge - who cares who wrote an article

Nothing in this world is absolute. ChatGPT is no different. Over time, everyone, I mean everyone, will realize that it is better than MOST of us. It will take time and w a few tweaks, it will collar most of us.

As much as I like ChatGPT for coding, the information it provides regarding research is not reliable for me.

The wetland answer is limited be it correct, but insufficient for a scientific paper.

Short answer is no. Long answer is still no.

Thankfully it does not provide references so the students cannot fully rely on it. As an academic I do fear that students who are stressed will use this and many will be caught out. It is plain and simple plagiarism.

Totally ethical to use AI. If it is not, then use a typewriter

Your prompts are vague and contain no detail to exactly what you are asking. Prompt GPT to include in-text citations and a reference portion. Prompt GPT to ensure the references are most current and backed by credible sources.

No, not anything that has real science and math behind it.

This service is not available in all countries. may be paid not free

Basically this app will make future degrees worthless. As an employer I wouldn’t hire anyone with a degree after 2019

What did you expect? An Oracle...this is an early stage.....the world is complex and even confusing to people.

@ScientificWritingwithKarenLMcK

Жыл бұрын

I didn’t know what to expect. That’s why I tested what it could do.

@bjrnhjortshjandersen1286

Жыл бұрын

@@ScientificWritingwithKarenLMcK Sorry I sounded harsh....I just first got the impression that you judged it on a much too critical level....writing scientific papers is a very rigorous dicipline. And that likely would need an AI trained specifically in that kind of data. For a general text AI I think it is quite impressive if you don´t ask for very precise information. 🙂

Now some corporations have started using this chatbot when you try to apply jobs via their website..I am never impressed and never will chat with Bot...it' brainwashing, it makes you think the machine is smarter then humans when in fact they are DUMB!

members of congress😂

You can get ChatGPT to provide references. I asked: "How does the complexity of decoding increase as a function of the number of qubits? Include a reference." The response included this promising reference: "Duclos-Cianci, G., & Poulin, D. (2019). Linear programming decoding of the surface code. Physical Review Research, 1(1), 013002." Unfortunately, no such article exists! I have tried similar queries with similarly bogus results. I gave feedback to ChatGPT so they can hopefully fix this in the future. Without legitimate references, ChatGPT is of limited use for scientific work.