登录
First Look 2025: Gemini for STEM, DeepSeek for Everything Else
First Look 2025: Gemini for STEM, DeepSeek for Everything Else
The semester always starts the same way. New classes, new tabs, and the same question: which AI should I actually use when the work gets real. After testing both tools with problem sets, lab PDFs, emails, and drafts, a pattern kept showing up. Gemini handles STEM with more structure and fewer surprises. DeepSeek moves faster on writing and day-to-day tasks.
Why Gemini belongs in STEM
You already live in JupyterHub and Google Colab for labs, at least I do as a data science major. Gemini fits your routine without asking you to learn a new app or change how you turn in work. Think of it as a lab partner that sits inside your notebook. You type a plain sentence about what you are trying to do, and it gives you code, an explanation, or both, right next to your data and plots. Because everything stays in the same notebook, it is easy to show your work to a TA and to remember why a change fixed the problem.
On JupyterHub, the appeal is that nothing about your class workflow changes. You still open the same notebook your instructor gave you, run the same cells, and submit the same file at the end. The only difference is that you can ask for help in place. Instead of leaving the notebook to search for a StackOverflow post, you can write a short prompt in a cell that explains the situation. For example, you might say that a KeyError is coming from a pandas selection and ask for two safe ways to fix it. Gemini responds with the explanation and short code you can run immediately. If it works, you keep the cell as part of your record. If it needs a tweak, you edit it and add a sentence explaining what you changed. The point is that the fixing and the learning happen in the same space where you already code, so your reasoning ends up in your submission rather than scattered through browser tabs.
Colab feels just as natural. You open a notebook, connect to a runtime, and talk to Gemini the way you would talk to a classmate who already knows your toolkit. If you say you have a CSV in Drive and want to fill missing values with the median, Colab can receive that request and return runnable cells that import the right libraries, read the file, do the cleaning, and make a simple plot. Because those cells are regular Python, you can immediately adjust column names, change the fill strategy, or add a comment in your own words. For team projects, this helps you get to a shared starting point quickly. Everyone sees the same code and the same comments, and you can divide the next steps without arguing about setup.
The best part for STEM courses is how this supports the “show your work” expectation. In statistics or data science, points are often tied to the justification, not just the final number. When Gemini suggests an approach, you can ask it to include a one or two line reason. If you are testing a model, you can request a tiny checklist that explains when the test is appropriate and what to look for in the output. When you run the cell and get results, you type a short note beneath it that says what you see and whether it matches the assumption. That habit turns a notebook into a readable story that graders appreciate, and it makes studying easier when you look back later.
Gemini also helps with common classroom pain points. When a plot will not render the way you expect, you can ask for a clear axis label and a sentence that states the main takeaway. When a loop is slow, you can ask for a vectorized pandas version and a quick timing check to prove it is faster. When you are stuck on a bug right before lab ends, you can paste the stack trace into a cell and ask for a short, plain English summary with two paths to try. None of this replaces learning the tools. It just removes the friction that keeps you from getting to the idea you want to test.
There are a few good habits to keep this useful and fair. After any code that Gemini suggests, ask for a brief explanation in everyday language so you can repeat it out loud later. Run the code and add one comment of your own about what happened. Most important, only submit work you understand. If a line of code looks mysterious, ask Gemini to rewrite it in a simpler form or to annotate it with comments. Your goal is to make the notebook readable to a future you who is tired and studying for an exam.
If you already rely on JupyterHub and Colab, you do not need a new mental model to use Gemini. You stay in the notebook, keep your code and explanations together, and move faster from problem to test to result. For a typical week in a STEM class, that means less time fighting setup and more time thinking about the question your instructor actually cares about.
Why DeepSeek carries the rest of your week
DeepSeek works best as your “all-courses” helper. Where Gemini sits inside your STEM notebooks, DeepSeek meets you in the places you write, read, and plan. You open a blank doc, a Canvas discussion, or an email, and instead of staring at the cursor you ask for a rough first pass you can shape. The goal is not to sound like a robot. It is to get past the awkward start and move quickly toward a draft that sounds like you.
Start with messy notes. Paste the scattered lines from your phone, a few quotes from last night’s reading, and a sentence about what your instructor actually asked. Tell DeepSeek the audience and the tone. It will turn that pile into a short outline with topic sentences you can approve or reject. Once the outline feels right, ask it to expand one section at a time so you stay in control of the voice. When a paragraph feels wooden, ask for a lighter, student-y rewrite that keeps your claims but changes the phrasing. If a citation is required, ask for help formatting the sources you already have rather than inventing new ones. You are steering the ship. DeepSeek is the wind that gets you moving.
For shorter writing, like emails to professors or internship applications, think of DeepSeek as a coach that keeps you clear and polite without sounding stiff. You say what you need, when you need it, and any details that might matter. It returns a two or three sentence draft you can paste into your mail app and tweak to match your voice. The same pattern works for résumés and cover letters. Feed it a bullet list of your experiences, the role you are applying for, and the skills the posting mentions. It will suggest a structure and accomplishment lines with verbs and numbers you can verify. If you have a draft already, ask for a pass that tightens the verbs and cuts filler so the page breathes.
DeepSeek is also useful for readings and study sessions outside STEM. When an article is dense, copy a small section and ask for a plain-English summary with the author’s claim, the evidence they use, and one question you could bring to discussion. If you have midterms coming up, give it the list of topics and the date, then ask for a week plan that spreads practice in a way that fits your schedule. It can turn your own lecture notes into a one-page study sheet with key terms, two example questions, and a short checklist of things to review. Because the raw material comes from you, the output stays close to what your instructor emphasized.
Group work gets easier when someone can break the ice. Share a prompt and your team’s rough ideas, then have DeepSeek propose a project outline with roles and deadlines. Nobody is locked in by the first draft, but it gives everyone something concrete to react to. When it is your turn to compile slides, you can paste each section’s notes and ask for a slide-friendly version with a title, three concise points, and a line you could say out loud. If a slide looks crowded, ask for the same content reduced to what absolutely must be on the screen, then put the rest in the speaker notes.
Good habits make all of this work for you rather than against you. Keep your prompts short and specific about the task, the audience, and the length. Bring your own sources and quotes so the content stays grounded. After DeepSeek produces text, read it once for accuracy, once for voice, and once for assignment fit. Add a sentence in your own words wherever a claim might be questioned. If your class requires disclosure, include a brief note that you used an AI assistant for drafting or editing. The aim is to learn faster and present your thinking clearly, not to hand in something you cannot explain.
If Gemini feels like a partner that lives inside your notebooks, DeepSeek is the partner that lives in your docs, emails, and to-do lists. It helps you start, helps you organize, and helps you finish with a clean draft that still sounds like you. On a typical week with readings, short responses, a club update, and a job application, that means less time stalled at the top of the page and more time refining ideas you actually care about.
Head to head in real situations
Use Gemini when you are in a notebook and want help that stays next to your code, plots, and results. It fits JupyterHub and Colab, explains errors in plain language, and helps you turn fixes into clean, graded cells. Use DeepSeek when you are writing or organizing outside the notebook. It is best for turning notes into outlines, polishing emails and essays, and shaping slides or study sheets.
If you are choosing for a single task, ask where the work lives. Code and data in a notebook point to Gemini. Drafts, messages, and planning docs point to DeepSeek. Most weeks you will touch both, with Gemini speeding up lab work and DeepSeek helping you finish clear, readable writing.
The simple rule to remember
Both tools can do a little of everything. The point is not to force a single model to be your entire workflow. The point is to use each one where it makes your life easier. If you start the semester with that split in mind, you will spend less time wrestling with your tools and more time getting your work done.
On April 20, 2020, the Digital Innovation Lab (DIL) will hold a Colloquium on Computational Poetics and Data Aesthetics, as an extended investigation and celebration of the relationships between humans and machines and the works of art, literature, and visual knowledge they produce together.
Mary Currin, Family Recruitment Coordinator at UNC, applies her data management and research expertise to medical fields, but her work in the program required her to branch out. “The Digital Humanities Certificate offered me the chance to explore data in a different context,” she says. “The majority of my courses were taken in the Information and Library Science Department, and I also had the opportunity to complete a placement at the State Archives.” Her work incorporated studies such as programming, metadata, and data curation.
Carlos Vázquez Cruz worked to complete an online edition of the picaresque gay novel from Puerto Rican writer Juan Antonio Rodríguez Pagán, Ondergraund.com. The novel began on a blog, but Pagán died before he was able to complete the full story. “While taking the courses leading to the DH Certification, my idea of completing Rodríguez Pagán’s work changed drastically,” says Carlos. In an effort to honor the “hypertext version initially proposed by Rodríguez Pagán,” Carlos added hyperlinks and tooltips. “But this is not the end of it,” he says. “I would like to join a DH community in the future to turn this project into a hypermedia novel.”
Kristen Foote, a PhD student in French and Francophone Studies, has been working in the digital humanities for years, most recently in the Digital Innovation Lab on campus, gaining experience with digital pedagogy, text analysis, and the administration of digital humanities programs. She is currently pursuing a fellowship in France, continuing her work to develop a digital corpus in order to analyze a disputed sixteenth-century French text.
Laurel Foote-Hudson, PhD student in English and Comparative Literature, has worked with several disciplines and digital tools while pursuing the certificate. “Over the years, I have worked with methods such as cultural and literary analysis,” she says, “but this program provided me with a new context to pursue several of the technical skills I have, but have not directly applied to my research.” Her first project involved establishing “a database and related taxonomy to build a comparison-focused data visualization (using tools such as D3 and Tableau),” while her second project saw “the development of an original short demo of an adaptation of the Japanese kabuki play Chūshingura using the Unity 3D game engine.”
This year marks the 