Nick Tarazona, MD’s Post

Did you know that chatGPT can generate graphs for information taken from sources that it cites? I asked chatGPT to generate a graph of estimated honeybee population as an example. It includes links to 4 citations, generated a graph that I could customize the look and style of. Totally changes the way we'll do research.

For me to understand the need for creative services in this market, I’ve decided that it’s time to use some science here. Science in ChatGPT, to be correct. I’m going to share with you why I need 271 responses for the survey, but before that - please fill out the survey and grab a free 15-minute consultation with me https://tally.so/r/w8KAoA So, here we go, let’s use ChatGPT for that (I’m no scientist here, sorry) and ask it for “Sample size in research methodology”. I've entered the numbers that I have researched prior and asked ChatGPT to conduct calculations and give me the number that will be suitable for this experiment. At least now I have a clear goal and it makes it a bit easier. P.S. I've even created a progress bar in Notion, which is quite useful as well, here's the formula if you need it (use the Formula type for the cell).

With the rise of more open-source Large Language Models (LLMs) such as Llama-2 and the quantization of models, it’s becoming easier to customize them for our own use cases. Although I have been using ChatGPT to check for spelling and grammar in text, this is the first time I'm exploring the concept of 'rewriting text in the style of'. I believe this can truly open up a lot of new applications in the FMCG industry, especially when tapping into proprietary data sources with the use of RAG. Let’s explore!

Your insights are invaluable to my research on ChatGPT; I kindly request your participation in filling out the questionnaire, contributing to the depth and quality of my study. https://lnkd.in/d7w6uMya

How is ChatGPT’s behavior changing over time? A recent Stanford University and UC Berkeley study examined the evolution of various GPT generations and their responses over time. The findings revealed a decrease in GPT's performance as newer versions are developed. This highlights the need for constant monitoring and assessment of LLMs in real-world applications by service providers. As end-users, it is crucial to exercise caution and verify the accuracy of the answers provided by these systems before accepting them blindly. Have you experienced something similar with ChatGPT or similar tools? Source: https://lnkd.in/e_q-f5gn

Both #ChatGPT and #Gemini can accurately triage critical and urgent patients in the Emergency Severity Index (ESI) 1&2 groups at a high rate. Furthermore, ChatGPT is more successful in ESI triage for all patients. These results suggest that large language models can assist in accurately triaging patients in the Emergency Room https://lnkd.in/dDYpvPzj

Could #ChatGPT 3.5 be capable of writing the discussion/conclusion of a #researchpaper? 🤔 To find it out, I based my mini experiment on the introduction ChatGPT had generated for me one week ago. Again, I aimed at the ‘moves’ it would apply to organize the ideas in the text since my purpose would be descriptive and linguistic. I was interested in detecting whether it would apply the macro and microstructural organization identified by Peacock’s model (2002) based on Dudley-Evans’ (1994). So…what happened? 😶 Although the result respected the organization I was looking for, it showed certain limitations unlike the introduction. For example, ChatGPT 3.5 could not return significant findings, graphs, or data since such input had not been provided previously. As a result, the lack of data weakened the microstructure of the concluding section of the text: it did not provide limitations about the research and recommendations for future works. I can infer that ChatGPT 3.5 will be capable of writing a discussion/conclusion for a research paper only if given enough data to do so. Otherwise, it will not follow the ‘moves’ expected on each one of the sections of the discussion. In case you would like to read the prompts, the resulting text, and the final analysis, click here: https://t.ly/3OluJ Did you expect a result like this in this mini experiment? Feel free to comment! 🙂 #medicalwriting #chatgptprompts #medicaltranslation #cardiology #researcharticle #generativeaitools #generativeai #translation

Today, I attended a talk about applying the findings of this paper to results at my place of business: https://lnkd.in/dQuYe3rT Here's the short version: ChatGPT outperforms automatic scoring techniques for a number of tasks, as measured in terms of how well its output correlates to human evaluation. That bit is really, really important - correlation to human evaluation. On the one hand, ChatGPT is the best system among those evaluated. On the other hand, the correlation between ChatGPT and humans *never exceeded 0.6*. In practice, this means that ChatGPT cannot be relied upon to give human-like feedback on... well, pretty much anything. More distressingly, it was suggested that a benefit of ChatGPT was that it never disagreed with itself, as compared to human evaluators who may disagree between themselves as to how a specific example should be scored. As a piece of research, this is interesting. As a technology to be implemented in lieu of human annotation or oversight, it's terrifying. As a technology to be implemented for quantitative evaluation, it's catastrophic. I can't properly get into full details of my thoughts here - maybe I need to start a blog, or write a book or something - but as a piece of enduring guidance and advice, please remember: the most precious resource on this planet is the human mind. Any attempt to replicate, subvert, or replace it, without full understanding of what is being replicated, subverted, or replaced, is doomed not only to failure, but collateral damage that cannot be calculated until the dust settles.

SciSpace a really interesting addition to ChatGPT, allowing you to access, summarise and challenge academic material in minutes. Have you had a look Lex Lang, Matt Warne and Dave Presky?

Are medical studies being written with ChatGPT? Well, we all know ChatGPT overuses the word "delve". Look below at how often the word 'delve' is used in papers on PubMed (2023 was the first full year of ChatGPT).