{"id":954,"date":"2025-03-15T21:52:22","date_gmt":"2025-03-15T21:52:22","guid":{"rendered":"https:\/\/microregistrar.com\/?p=954"},"modified":"2025-05-15T20:58:21","modified_gmt":"2025-05-15T19:58:21","slug":"using-llms-effectively-for-academic-work","status":"publish","type":"post","link":"https:\/\/microregistrar.com\/?p=954","title":{"rendered":"Using LLMs effectively for academic work"},"content":{"rendered":"\n

This summary reflects my experience with LLMs over the past several months. I have interacted with Gemini (formerly Bard), ChatGPT (including version 4.5), Perplexity, Claude, and Grok3, along with some use of DeepSeek, Mistral, AllenAI, Leonardo, Midjourney and Copilot.<\/p>\n\n\n\n

Every LLM has its strengths and weaknesses, and no one LLM is the best at doing everything for your academic needs. However, the capabilities and functionalities are changing regularly and fast.<\/p>\n\n\n\n

Search<\/td>Perplexity leads for research; Gemini excels with Google integration. Deepsearch of ChatGPT is excellent in my opinion. Grok DeepSearch is also good.<\/td><\/tr>
Hallucination<\/td>ChatGPT (DALL-E 3) and Gemini lead in generating stock images.
Claude is good at producing charts and graphics for academic purposes. ChatGPT cen refine academic images but is often unreliable.<\/td><\/tr>
Reliability<\/td>Claude for logic; Perplexity for facts<\/td><\/tr>
Deep Search<\/td>Perplexity offers it free; Grok, ChatGPT and Gemini shine in paid tiers.<\/td><\/tr>
Image Generation<\/td>ChatGPT (DALL-E 3) and Gemini lead in generating stock images.
Claude is good at producing charts and graphics for academic purposes. ChatGPT can refine academic images but is often unreliable.<\/td><\/tr>
Coding<\/td>Claude and DeepSeek are top-tier<\/td><\/tr>
Video<\/td>Gemini has this capability.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Aspect<\/strong><\/td>ChatGPT<\/strong><\/td>Perplexity<\/strong><\/td>Grok<\/strong><\/td>Claude<\/strong><\/td>DeepSeek<\/strong><\/td>Gemini<\/strong><\/td><\/tr>
Search<\/strong><\/td>Paid only, decent<\/td>Best, real-time<\/td>Real-time, X (twitter) focus<\/td>Basic<\/td>Good, technical<\/td>Excellent, Google<\/td><\/tr>
Risk of Hallucination<\/strong><\/td>Moderate<\/td>Low<\/td>Low<\/td>Very Low<\/td>Moderate<\/td>Moderate<\/td><\/tr>
Reliability<\/strong><\/td>High, general<\/td>High, research<\/td>High, technical<\/td>Very High, logic<\/td>High, STEM<\/td>High, multimodal<\/td><\/tr>
Deep Search (Free)<\/strong><\/td>No<\/td>Yes<\/td>Limited<\/td>No<\/td>Yes, limited<\/td>Yes<\/td><\/tr>
Deep Search (Paid)<\/strong><\/td>Yes (Pro)<\/td>Yes (Pro)<\/td>Yes (Premium+)<\/td>No<\/td>Yes<\/td>Yes (Advanced)<\/td><\/tr>
Image Generation<\/strong><\/td>Yes (Paid)<\/td>No<\/td>Yes (Paid)<\/td>No<\/td>No<\/td>Yes (Paid)<\/td><\/tr>
Coding Capability<\/strong><\/td>Strong<\/td>Decent<\/td>Strong<\/td>Exceptional<\/td>Exceptional<\/td>Strong<\/td><\/tr>
Video Generation<\/strong><\/td>No<\/td>No<\/td>No<\/td>No<\/td>No<\/td>Yes (Paid)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

You can use LLMs for<\/strong><\/p>\n\n\n\n

\u2022searching literature,
\u2022generating ideas,
\u2022blog posts,
\u2022social media post,
\u2022generating structure for your article,
\u2022reviewing your paper or correcting spelling\/grammar\/structure
\u2022summarising
\u2022statistical analysis,
\u2022critically appraise a paper<\/td>		\u2022Write an article (you must be careful about generating the whole article for a paper)
\u2022generating images (stock image, graph, tables, diagram, or just an absurd image for fun),
\u2022testing your understanding of a subject by conversing with you
\u2022generating podcast
\u2022generating code
\u2022generating a short video.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

How can you use LLM safely?<\/p>\n\n\n\n

\u2022Do not upload confidential data (for example: no patient identifiable data or business intelligence\/business critical data)
\u2022Follow your organisation\u2019s AI\/data policy or your National policy<\/td>		Although LLM use is considered sessional (the information you upload is not retained after you end the conversation, it might still be reviewed by someone to train the LLM.<\/em>
Also, there is always a potential policy change from the LLM companies, leading to data retention<\/em>
There is always some possibility of data retention if you are using services like ChatGPT scheduled task or perplexity spaces.<\/em><\/td><\/tr>
\u2022Use your prompts effectively (more on this later)
\u2022 Cross-check the output (text and reference)<\/td>		There is always a risk of hallucination (LLM producing information or reference that does not exist). It is important to check the information if you are producing something that is not in the field of imaginative work (writing a story)<\/em>
Any academic work must be checked for its validity. I find expert knowledge is important here. If you are not an expert in that field you risk missing the hallucination.<\/em>
Appropriate prompt can help to some extent, but still cross checking is important<\/em><\/td><\/tr>
\u2022Check the AI use policy of the organisation
\u2022Declare the scope of use of AI in your work<\/td>		If you are using AI\/LLM for serious work (preparing a journal article or business report) your organisation or the publisher may want to know the extent of the use of the AI. Always check what the scope is and comply.<\/em><\/td><\/tr>
\u2022Beware of the guardrail
\u2022Refrain from using AI\/LLM for offensive\/illegal purpose<\/td>		There is always a risk of hallucination (LLM producing information or reference that does not exist). It is important to check the information if you are producing something that is not in the field of imaginative work (writing a story)<\/em>
Any academic work must be checked for its validity. I find expert knowledge is important here. If you are not an expert in that field you risk missing the hallucination.<\/em>
Appropriate prompt can help to some extent, but still cross-checking is important<\/em><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Using prompts effectively for an academic purpose<\/h2>\n\n\n\n
    \n
  1. Detailed and Explicit Prompts:<\/strong><\/li>\n<\/ol>\n\n\n\n
      \n
    • There is no such thing as a prompt that’s “too long.” If you have to use a long prompt to get an answer you want, go for it.<\/li>\n\n\n\n
    • Include context, specific tasks, intended outcomes, and any constraints clearly.<\/li>\n<\/ul>\n\n\n\n

      2. Assign a Role or Character to the LLM:<\/strong><\/p>\n\n\n\n

        \n
      • Clearly define what role the LLM should adopt (e.g., “senior consultant,” “editor of a journal,” “statistician,” “historian”). This significantly influences the nature and depth of responses.<\/li>\n\n\n\n
      • Example: If you are asking LLM to review an article, write\u00a0 “Assume you are an editor of a reputed medical journal evaluating a research article. Review this article {name}\u201d<\/em><\/li>\n\n\n\n
      • If you ask LLM “You are Napoleon Bonaparte, analyse the Battle of Gaugamela”, you will get a different answer than asking it \u201cYou are William Shakespeare \u2013 describe the Battle of Gaugamela”.<\/li>\n<\/ul>\n\n\n\n
        \n
        \n

        As Napoleon Bonaparte, I would analyse the Battle of Gaugamela thus:<\/p>\n\n\n\n

        The Battle of Gaugamela (331 BC) epitomises Alexander the Great\u2019s brilliance in battlefield strategy, a masterpiece of tactical flexibility and psychological dominance. Alexander’s deliberate use of the oblique order, combined with disciplined troops and precisely executed maneuvers, showcases his strategic genius. His deliberate weakening of the left flank to entice Darius III into committing heavily, thereby creating vulnerabilities in the Persian center, exemplifies masterful deception.<\/p>\n\n\n\n

        Alexander\u2019s bold cavalry charge, precisely timed and directed straight at the heart of Darius’ formation, demonstrated courage and calculated risk-taking\u2014principles essential to decisive victories. Darius’s inability to adapt to rapidly changing battlefield conditions and his premature retreat signified a failure in leadership and the psychological collapse of Persian morale.<\/p>\n\n\n\n

        \u2022From my own perspective, victory demands decisiveness, clear communication, and exploitation of the enemy\u2019s psychological weaknesses\u2014all expertly demonstrated by Alexander. Had I faced Darius at Gaugamela, my approach would have resonated with Alexander\u2019s methods: breaking enemy cohesion, maintaining initiative, and achieving decisive victory through audacious yet calculated maneuvers. Alexander achieved a model of strategic victory that military leaders should study for centuries hence.<\/p>\n<\/div>\n\n\n\n

        \n

        As William Shakespeare, my depiction of the Battle of Gaugamela would unfold thus:<\/p>\n\n\n\n

        Upon the plains of Gaugamela wide,
        Two mighty hosts in fateful clash abide.
        Great Alexander, fierce with youthful fire,
        Met Darius, crowned by desperate desire.<\/p>\n\n\n\n

        A sea of steel and courage brightly gleamed,
        As spears like stars against the sunlight streamed.
        Alexander, bold, with eagle\u2019s eyes did see
        The Persian weakness\u2014fate\u2019s decree.<\/p>\n\n\n\n

        He charged with valour’s flame, like tempest wind,
        A sharpen\u2019d wedge, as Destiny had skinned
        The fabric of the Persian king\u2019s array;
        The centre shattered, courage fell away.<\/p>\n\n\n\n

        The royal Darius fled the grievous sight;
        His armies broken, routed from the fight.
        Thus Fortune favoured daring hearts that day,
        And Alexander claimed immortal sway.<\/p>\n\n\n\n

        \u2022This battlefield, where glory spilt its flood,
        Is ink\u2019d forever in the scrolls of blood.
        O Gaugamela! Name remembered long,
        Where pride did fall, and heroes\u2019 deeds belong.<\/p>\n<\/div>\n<\/div>\n\n\n\n

        3. Demand factual accuracy explicitly:<\/strong><\/p>\n\n\n\n

          \n
        • LLMs tend toward generating coherent narratives, occasionally at the expense of factual accuracy. Explicitly instruct the LLM to adhere strictly to factual information.<\/li>\n\n\n\n
        • LLMs are reluctant to say \u201cno\u201d unless you give it permission to do so.<\/li>\n\n\n\n
        • Tell it \u201cProvide answers based solely on factual information. Avoid generalisations, assumptions, or extrapolations beyond explicitly stated or verifiable data. If uncertain, explicitly state uncertainty, and if information is not available, say not available.”<\/li>\n<\/ul>\n\n\n\n

          4. Set clear boundaries and constraints:<\/strong><\/p>\n\n\n\n

          Specify clearly what kind of resources and guidelines to use or avoid. For instance:<\/p>\n\n\n\n

            \n
          • \u201cOnly cite peer-reviewed journal articles published within the last 10 years, prioritising systematic reviews and meta-analyses.\u201d<\/li>\n\n\n\n
          • \u201cPrimarily use UK guidelines (e.g., NICE, BIA, HIS, BASHH, BHIVA). If no UK guidelines are available, explicitly indicate this and cite European, US, or WHO guidelines.\u201d<\/li>\n\n\n\n
          • \u201cExplicitly state when using non-peer-reviewed or unofficial sources, clearly indicating their limitations.\u201d<\/li>\n<\/ul>\n\n\n\n

            5. Ensure rigorous citation and verification:<\/strong><\/p>\n\n\n\n

            Be extra-careful about citation. Citations are a frequent area of inaccuracies or hallucinations. Explicitly instruct the LLM to verify citations and clearly communicate citation details:<\/p>\n\n\n\n

              \n
            • \u201cList citations clearly with authors, title, journal\/guideline name, year, DOI or URL. Explicitly confirm that the cited information is accurately located within the provided source (mention page number, section, or heading).\u201d<\/li>\n\n\n\n
            • \u201cIf citation verification fails or is uncertain, explicitly flag this with statements like \u2018Citation existence not confirmed,\u2019 \u2018Information not found in cited source,\u2019 or \u2018Citation accessibility issue.\u2019\u201d<\/li>\n<\/ul>\n\n\n\n

              6. Provide a clear structure for answers:<\/strong><\/p>\n\n\n\n

              Offer a detailed structure or framework for responses. This significantly improves consistency, completeness, and readability. Example:<\/p>\n\n\n\n

                \n
              • \u201cWhen describing a bacterial species, provide information under the following structured headings: Classification, Virulence Factors, Laboratory Characteristics (Gram stain, culture media, biochemical tests), Clinical Presentations, Differential Diagnoses, Investigations, and Management\/Treatment.\u201d<\/li>\n<\/ul>\n\n\n\n

                7. Iterative refinement (Ask the LLM for Help):<\/strong><\/p>\n\n\n\n

                If unsure, first ask the LLM to propose an initial outline, structure, or scope for your topic. Example:<\/p>\n\n\n\n

                  \n
                • \u201cSuggest a comprehensive outline covering all relevant aspects of bacterial pathogens. Refine this outline based on my feedback.\u201d<\/li>\n<\/ul>\n\n\n\n

                  8. Precision and language style:<\/strong><\/p>\n\n\n\n

                  Specify the desired tone and style of writing (e.g., formal academic tone, concise and technical, detailed yet concise). Example:<\/p>\n\n\n\n

                    \n
                  • “Write in a formal academic style, concise yet detailed, appropriate for postgraduate medical students.”<\/li>\n\n\n\n
                  • Or you can ask \u2013 guideline style, short summary, summarise in 3 lines, use a language suitable for a lay person..<\/li>\n<\/ul>\n\n\n\n

                    9. Start a conversation:<\/strong><\/p>\n\n\n\n

                      \n
                    • LLM, although not human but remarkably humanlike in its conversational capability. Do not consider it as a machine but assume it as a person who is willing to help you and will never be angry if you ask the same thing hundreds of times (well you can go over your usage limit).<\/li>\n\n\n\n
                    • Start a conversation. If you do not like a part of the answer, suggest it. Ask for refinements. If you like a part \u2013 appreciate it. It will take these interactions on board and adjust its answer according to this. However, remember it’s memory is sessional, as soon as you close the conversation this small modification will be lost.<\/li>\n<\/ul>\n\n\n\n

                      10. Make a list of common prompts and store:<\/strong><\/p>\n\n\n\n

                      \u2022Store your common prompts in a separate place.<\/p>\n\n\n\n

                        \n
                      • As the LLM’s memory is sessional, you can make a big prompt to set the tone of the conversation and then start your interaction.<\/li>\n\n\n\n
                      • Your starting line could be something like \u2013 \u201cYou are a consultant in the NHS involved in patient care and teaching. You look for answers based solely on factual information. Avoid generalisations, assumptions, or extrapolations beyond explicitly stated or verifiable data. If uncertain, explicitly state uncertainty, and if information is not available, say not available\u2026.\u201d<\/li>\n<\/ul>\n\n\n\n

                        Invite LLM to all your workflows.. (as long as it is safe to do so)..<\/h2>\n\n\n\n

                        <\/p>\n","protected":false},"excerpt":{"rendered":"

                        This summary reflects my experience with LLMs over the past several months. I have interacted with Gemini (formerly Bard), ChatGPT […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ngg_post_thumbnail":0,"footnotes":""},"categories":[194],"tags":[195,196],"class_list":["post-954","post","type-post","status-publish","format-standard","hentry","category-ai","tag-ai","tag-llm"],"_links":{"self":[{"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/posts\/954","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/microregistrar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=954"}],"version-history":[{"count":1,"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/posts\/954\/revisions"}],"predecessor-version":[{"id":958,"href":"https:\/\/microregistrar.com\/index.php?rest_route=\/wp\/v2\/posts\/954\/revisions\/958"}],"wp:attachment":[{"href":"https:\/\/microregistrar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=954"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/microregistrar.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=954"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/microregistrar.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=954"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}