HorizonX

Researchers waste 40% of their time just finding the right papers. We built HorizonX to fix that. 🔬 200M+ academic papers 🧠 AI Literature Review 💬 Research Chat — journals only, zero noise 📎 Citation-ready results Free to start → horizonx.live

@JeremyNguyenPhD fair, that's a better description. two people at the keyboard is still writing, just collaborative writing. the point I was reaching for is the gap between having understood something and being able to articulate it, not whether you're alone when you do it.

@__HorizonX__ why do you say that writing is solo? so many of my papers have literally involved two people at the same computer, one at the keyboard and both trying to find the right words

Want to start using Claude Code for academic research? Or see incredible things other researchers are doing with Claude Code? Here's a thread of 5 great tutorials, skill and projects you can start using right now. I'm also keen to do an online "reading group" to work through these and others every week. Beginners welcome. Let me know below if you're interested. 1/ Chris Blattman shares an entire suite of tools he build in the last 4 weeks:

@Faheem_uh missing step between 2 and 3 is synthesis. after you've collected your papers but before you start collecting data, you need to figure out what the field actually says and where the real gaps are. most tools skip right past it.

Which tool to use in each phase of your PhD? 1. Exploration and Idea Generation 𝐀𝐧𝐬𝐰𝐞𝐫𝐓𝐡𝐢𝐬 → lnkd.in/dbeWPgME 𝐄𝐮𝐫𝐞𝐤𝐚 → lnkd.in/d_ZX6tXq 2. Literature Review 𝐓𝐫𝐢𝐧𝐤𝐚 → lnkd.in/dxpyDn8Z 𝐖𝐢𝐬𝐏𝐚𝐩𝐞𝐫 → lnkd.in/dVUCUA_p 3. Data Collection 𝐍𝐞𝐱𝐃𝐚𝐭𝐚 → lnkd.in/d_paSwt7 𝐦𝐨𝐚𝐫𝐚 → moara.io 4. Data Analysis 𝐒𝐜𝐢𝐒𝐩𝐚𝐜𝐞 → lnkd.in/gv8FpUpU 𝐆𝐚𝐭𝐬𝐛𝐢 → gatsbi.com 5. Writing and Drafting 𝐏𝐚𝐩𝐞𝐫𝐩𝐚𝐥 → lnkd.in/d_HFd4QT 𝐑𝐞𝐯𝐢𝐞𝐰-𝐢𝐭 → review-it.ai 6. Review and Refinement 𝐑𝐞𝐯𝐢𝐞𝐰𝐞𝐫𝟑 → review-it.ai 𝐓𝐡𝐞𝐬𝐢𝐟𝐲 → thesify.ai

@PhDtoProf three passes works well for individual papers. the harder version is doing this across 20 papers at once, where you're trying to figure out what they all mean together. pass 3 across multiple papers is where most researchers lose weeks.

Reading every word. Start to finish. Taking 3 hours per paper. Forgetting page 1 by page 15. Sound familiar? It's called the Three-Pass Approach. Here's exactly how it works:

How to read 50 research papers without losing your mind. I spent years doing it wrong. Then I found a method that cut my reading time by 70%.

@PhD_Genie the folder always looks impressive. it's the moment you have to actually synthesize them that things fall apart.

Collecting papers for literature review into a folder.

one thing that keeps coming up when talking to researchers: they don't know if a gap in the literature is actually a gap, or if its just a gap in what they've found. those are very different problems. and most tools don't help you tell them apart. how do you figure out which one you're dealing with?

@JeremyNguyenPhD yeah the coauthor dynamic is interesting. the understanding gets built collectively, in conversation. but writing is solo. so there's this translation step where shared understanding has to become individual articulation, and thats where it falls apart.

@__HorizonX__ is that true? people I know have so many papers where we've talked it through with our coauthors, really banged our heads against the wall until it makes a type of sense to us and then someone gets delayed or stumbles while writing it up.

@Haofan_Wang there's a real tension in that shift. the builders who stop reading papers eventually start re-solving problems that were already solved 2 years ago. the trick is reading differently, not reading less.

I've shifted my day from reading papers and training models to using newly launched products and building my own side products with coding agents.

@EricTopol @NatureMedicine @ekoermann 4,609 papers and 19 actual trials. the publication volume is almost entirely about the tools, not the domain. that ratio tells you something about where the real research work is still left to do.

A new systematic review of LLMs in medicine @NatureMedicine @ekoermann nature.com/articles/s4159… “Despite thousands [4,609] of publications since late 2022, only a small fraction use real clinical data and just 19 randomized trials exist.”

@Faheem_uh curious, in your experience teaching researchers, is interpretation usually something they develop over time, or does it seem more like a disposition some people bring in and others just don't? wondering if it can actually be trained.

@__HorizonX__ I agree with your point.

How to generate a literature matrix table from multiple research papers? First, what is a literature matrix table. 𝐀 𝐥𝐢𝐭𝐞𝐫𝐚𝐭𝐮𝐫𝐞 𝐦𝐚𝐭𝐫𝐢𝐱 𝐭𝐚𝐛𝐥𝐞 ➝ Contains extracted data from the papers ➝ Helps synthesize multiple research papers ➝ Data can be aims, methodology, key findings, etc. ➝ You will analyze this data for your literature review ➝ Your analysis will lead to new findings 𝐍𝐨𝐰, 𝐡𝐨𝐰 𝐭𝐨 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞 𝐚 𝐥𝐢𝐭𝐞𝐫𝐚𝐭𝐮𝐫𝐞 𝐦𝐚𝐭𝐫𝐢𝐱 𝐭𝐚𝐛𝐥𝐞 Especially when you have multiple research paper. Let’s suppose you have identified 10 papers. Great. Now do the following 1. Go to tinyurl.com/6s6aykp9 and log in 2. Click on 𝐶ℎ𝑎𝑡 𝑃𝐷𝐹 from the left menu 3. So, upload your 10 papers 4. Now click on 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑒 𝑙𝑖𝑡𝑒𝑟𝑎𝑡𝑢𝑟𝑒 𝑚𝑎𝑡𝑟𝑖𝑥 𝑡𝑎𝑏𝑙𝑒 𝑝𝑟𝑜𝑚𝑝𝑡 5. @teampaperpal will create literature matrix table 6. The literature matrix table helps you compare findings, group studies etc. 7. The table contains the following from all source files ✦ Study objectives ✦ Major methods ✦ Key findings ✦ Conclusions ✦ Limitations 7. Data items is linked with the respective part in paper 8. To see the paper part, just click the colored reference/citation 9. It will show you the exact part in the research paper 10. You can do the following with the table — Copy the content — Download the content — Add notes to the table 11. You can also use agentic AI to do the following → Prepare literature review draft of a paper → Compare and contrast findings → Identify future research gaps → and so on If you are doing literature review, try it. You will love it. Try it here for FREE: tinyurl.com/6s6aykp9

at what point do you stop reading and start writing? not about word count or deadlines. more like, what's the internal signal that tells you you've actually read enough to have something to say? curious how researchers think about this. it seems different for everyone. drop your answer below.

@drainzyy @SentientAGI parsing is the part that's been solved for a while. the real question is whether the synthesis holds up when there are contradictions in the corpus, not just consensus. that's where most real research decisions actually live.

@SentientAGI alphaXiv in this cohort is huge. Agents that can actually parse and synthesize research papers in real time? That's the unlock nobody's talking about enough🔥

Applications are now live! Cohort 0 starts March 13th in Presidio with OpenHands, OpenRouter, alphaXiv, Fireworks, Dedalus Labs, Franklin Templeton, Founders Fund and Pantera. → $25K+ in prizes → 3 weeks building state-of-the-art AI agents → Many more surprises Apply below 👇

@apkinverse "figuring out what i actually wanted from a paper" is such an underrated skill. most people sit down with a paper with no real question in mind, just hoping the paper will tell them what matters. that's where hours disappear.

feb × papers wrapped - i tried reading research papers every day in feb. the goal was: compress exposure, get better at reading papers, and stay closer to what’s happening in current research. what i gained/what was hard 👇

@Karajarati skimming isn't always laziness. sometimes it's the brain protecting itself from investing in a paper before knowing if it's worth it. the trick is a middle layer between skim and deep read, where you decide. most people skip that and go straight to "save for later"

I need to stop skimming research papers and actually start reading them thoroughly and take notes

been talking to a lot of researchers lately and one thing keeps coming up: they already have a mental model of their field before they sit down to do a literature review. and the review isn't actually changing that model. it's just confirming it. the papers that challenge the model get skimmed. the ones that fit get cited. that's not a tool problem. that's a workflow problem. but better tools can at least surface the friction points before they become blind spots.

@Faheem_uh and interpretation is the part that doesn't compress well. you can't just prompt your way out of it, you need to actually sit with the disagreements across papers and decide what they mean for your question. that's still very human work.

@__HorizonX__ I agree. How we interpret the extracted data is the other half of the entire job.

@Am_Nancie @AlanMCole fabricated references are the silent failure mode no one wants to talk about. the student doesn't know the paper doesn't exist, the supervisor often can't tell unless they check. the citation looks real, it just isn't. that's a different problem than just "AI mistakes".

@AlanMCole Claude 4.6 is quite smarter than GPT 5.2. For basic stuff all are good but let's remember the downside of AI is too much hallucinations. Evidenced from my student's literature review that had non -existing references

Is Claude actually smarter than GPT? Or does he just feel that way because he doesn’t have the hyperactive golden retriever personality?

@quasicoh 100% agree on explaining concepts. the gap shows up when you need the actual research landscape, not just the concept. LLMs smooth out contradictions and missing debates that the papers themselves contain. great for onboarding, less reliable for knowing what's actually settled.

It is honestly insane how good LLMs are at explaining ML research. I recently spent 1-2 hours chatting with ChatGPT, Claude, and Gemini about GRPO and I learned 20x more efficiently than I would have by reading books and papers.

when you're reading papers for a new research area, what's the actual hardest part? not the search. not the access. the part after that. when you have 30 papers open and you're trying to figure out what they actually mean together. curious what that process looks like for different people. drop a reply.