[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-clear-prompts-turn-ai-search-into-usable-answers-en":3,"article-related-clear-prompts-turn-ai-search-into-usable-answers-en":29,"series-research-2279e33f-db76-4e6f-80e3-527925afe93c":70},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":21,"views":25,"created_at":26,"published_at":27,"topic_cluster_id":28},"2279e33f-db76-4e6f-80e3-527925afe93c","clear-prompts-turn-ai-search-into-usable-answers-en","CLEAR prompts turn AI search into usable answers","\u003Cp>I've been using research chat tools long enough to know when a prompt is the problem. The model isn’t “bad”; I’m being lazy, vague, or both. I’d toss in a half-formed question, get a confident mess back, then waste another ten minutes trying to rescue it with follow-up prompts. That loop gets old fast.\u003C\u002Fp>\u003Cp data-speakable=\"summary\">Georgetown’s CLEAR framework turns vague AI prompts into tighter research questions.\u003C\u002Fp>\u003Cp>What finally helped me was seeing prompt writing treated like a research skill, not a magic trick. Georgetown University Library’s guide on \u003Ca href=\"https:\u002F\u002Fguides.library.georgetown.edu\u002Fai\u002Fprompts\">How to Craft Prompts\u003C\u002Fa> points to Leo Lo’s CLEAR framework and makes a very practical argument: better prompts save time, money, and sanity. The guide also links out to the \u003Ca href=\"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.acalib.2023.102720\">CLEAR path paper\u003C\u002Fa>, the \u003Ca href=\"https:\u002F\u002Fwww.promptingguide.ai\u002F\">DAIR.AI Prompt Engineering Guide\u003C\u002Fa>, and examples for text and image prompts. That’s the source I’m unpacking here, not because it’s flashy, but because it’s the kind of library advice I actually trust.\u003C\u002Fp>\u003Ch2>Stop asking AI to read your mind\u003C\u002Fh2>\u003Cblockquote>A prompt is technically input to an AI tool. For research AI tools, the prompt is almost always a question, request, or topic posed by you, the human researcher.\u003C\u002Fblockquote>\u003Cp>That sounds obvious until you watch people write prompts like they’re texting a coworker who already knows the project, the deadline, and the audience. They don’t. The AI has none of that context unless you put it there.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784921593697-nttg.png\" alt=\"CLEAR prompts turn AI search into usable answers\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>What this actually means is that prompt quality is mostly about context control. If I ask, “What’s Georgetown like?” I’m handing the model a mushy target. If I ask, “Give me a concise summary of Georgetown University’s strengths and weaknesses for a first-generation undergraduate applicant,” I’ve done half the work already. I’ve named the institution, the audience, and the output shape.\u003C\u002Fp>\u003Cp>I ran into this constantly when I tested research tools for literature discovery. A vague topic like “climate policy” returned a pile of broad, generic material. Once I narrowed it to “peer-reviewed studies on urban heat island mitigation in midsize U.S. cities,” the tool started behaving like it had a job.\u003C\u002Fp>\u003Cp>How to apply it: before you prompt, write down the three things the model must know: topic, audience, and output type. If any of those are missing, the answer will drift.\u003C\u002Fp>\u003Cul>\u003Cli>Topic: what exactly are you asking about?\u003C\u002Fli>\u003Cli>Audience: who is the answer for?\u003C\u002Fli>\u003Cli>Output type: summary, list, comparison, table, explanation, or draft?\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>CLEAR is basically a cleanup pass for your brain\u003C\u002Fh2>\u003Cblockquote>Good prompts are CLEAR: Concise, Logical, Explicit, Adaptive, Reflective.\u003C\u002Fblockquote>\u003Cp>Georgetown’s guide uses Leo Lo’s framework from \u003Ca href=\"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.acalib.2023.102720\">The CLEAR path\u003C\u002Fa>. I like it because it’s not pretending prompt writing is mystical. It’s just a checklist for removing dumb mistakes.\u003C\u002Fp>\u003Cp>What this actually means is that CLEAR is a debugging tool for language. If your prompt is unclear, the model will happily fill in the blanks with whatever it thinks is reasonable. That’s the problem. The model is optimized to respond, not to stop and say, “Hey, this is underspecified.”\u003C\u002Fp>\u003Cp>That’s why I prefer CLEAR over the usual “be more specific” advice. Specific about what? In what order? With what constraints? CLEAR gives me a structure I can actually use when I’m tired and trying to move fast.\u003C\u002Fp>\u003Cp>How to apply it: run every draft prompt through the five letters before you hit enter. If the prompt fails one letter, fix that first instead of blaming the model.\u003C\u002Fp>\u003Cul>\u003Cli>C: cut filler words and extra setup.\u003C\u002Fli>\u003Cli>L: make the relationships between ideas make sense.\u003C\u002Fli>\u003Cli>E: say exactly what you want back.\u003C\u002Fli>\u003Cli>A: revise based on the model’s answer.\u003C\u002Fli>\u003Cli>R: check the answer for quality, bias, and gaps.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Concise does not mean stingy\u003C\u002Fh2>\u003Cblockquote>Concise: Focus on the key words for the AI tool to analyze. Try to omit as many needless words as possible.\u003C\u002Fblockquote>\u003Cp>This is where people overcorrect. They either write a paragraph of background noise or they strip the prompt so hard that it becomes useless. Concise is not “short for the sake of short.” It’s “remove everything that doesn’t help the model answer.”\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784921589541-loai.png\" alt=\"CLEAR prompts turn AI search into usable answers\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The Georgetown example is good because it shows what bad concision looks like in practice. “I am trying to figure out if I should be applying to Georgetown and if I would like it there” is full of human intent, but it’s also ambiguous and padded. The model has to guess whether you mean undergraduate, graduate, law, or something else. It also has to guess what “like it there” means.\u003C\u002Fp>\u003Cp>What this actually means is that good prompts compress intent, not meaning. I want fewer words, sure, but I don’t want fewer constraints. The best concise prompt is the one that deletes fluff while preserving the boundaries of the task.\u003C\u002Fp>\u003Cp>I’ve seen this save a lot of time in research workflows. If I ask for “a concise summary of the major strengths and weaknesses of Georgetown University,” I get something usable. If I ask for “tell me everything important about Georgetown,” I get a blob that takes longer to sort through than the original search results.\u003C\u002Fp>\u003Cp>How to apply it: delete anything that doesn’t change the answer. If a phrase only expresses your mood, not your task, cut it.\u003C\u002Fp>\u003Cpre>\u003Ccode>Bad: Can you help me figure out everything I should know about Georgetown and whether it is the right school for me, because I am stressed and comparing a lot of options?\nGood: Give me a concise summary of Georgetown University’s strengths and weaknesses for an undergraduate applicant.\u003C\u002Fcode>\u003C\u002Fpre>\u003Ch2>Logical prompts stop the model from guessing your meaning\u003C\u002Fh2>\u003Cblockquote>Logical: Most AI tools look for relationships between words and concepts, so make sure your query presents concepts accurately and in their natural or logical order.\u003C\u002Fblockquote>\u003Cp>This is the part people skip because it feels too basic. Then they wonder why the model answered the wrong question perfectly. The issue is often not missing information. It’s broken phrasing.\u003C\u002Fp>\u003Cp>What this actually means is that word order matters because the model is pattern-matching relationships. If I ask, “Can we make a flu vaccine organically?” I’ve made a mess. “Organic” could mean farmed ingredients, a natural process, or something else entirely. The model has to infer which meaning I intended, and that’s where nonsense creeps in.\u003C\u002Fp>\u003Cp>I’ve made this mistake with research prompts too. I once asked a tool to compare “effects of remote work on productivity and stress in tech workers and students.” The output mashed two populations together and produced a comparison that looked polished but wasn’t actually answering either group well. The fix was to split the prompt into a clean logical structure: one population, one outcome, one comparison.\u003C\u002Fp>\u003Cp>How to apply it: write prompts in the same order you would explain the task to a colleague. Subject first, then action, then constraints. If the sentence sounds weird to you, it’ll probably be weird to the model too.\u003C\u002Fp>\u003Cul>\u003Cli>Use one main task per prompt.\u003C\u002Fli>\u003Cli>Separate populations, timeframes, and outcomes.\u003C\u002Fli>\u003Cli>Avoid words that can mean three different things depending on context.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Explicit prompts are where the useful answers start\u003C\u002Fh2>\u003Cblockquote>Explicit: Be clear in what you want from the AI. Giving the AI tool clear output directions can help the AI produce an answer that is useful to you.\u003C\u002Fblockquote>\u003Cp>This is the real jump from “chatting with AI” to actually using it. The model does much better when I tell it what shape the answer should take. A summary is not a list. A list is not a critique. A critique is not a comparison table.\u003C\u002Fp>\u003Cp>What this actually means is that output instructions reduce ambiguity. If I want a concise answer, I say so. If I want strengths and weaknesses, I say so. If I want the answer from a particular perspective, I say that too. The model can only optimize for the shape I describe.\u003C\u002Fp>\u003Cp>I use this constantly when I’m drafting research notes. “Summarize the article” gives me a generic paragraph. “Summarize the article in three bullets, include one limitation, and keep it under 120 words” gives me something I can paste into my notes without cleaning it up for ten minutes.\u003C\u002Fp>\u003Cp>How to apply it: specify format, length, tone, and perspective. If you need a table, ask for a table. If you need a short answer, say short. If you need a student-facing explanation, say that too.\u003C\u002Fp>\u003Cpre>\u003Ccode>Prompt template:\nGive me [format] about [topic] for [audience].\nInclude [required elements].\nKeep it [length\u002Ftone].\u003C\u002Fcode>\u003C\u002Fpre>\u003Ch2>Adaptive prompts are just better follow-up prompts\u003C\u002Fh2>\u003Cblockquote>Adaptive: Try a second prompt with keywords or topics suggested by the AI in its answer.\u003C\u002Fblockquote>\u003Cp>This is where a lot of people give up too early. They treat the first answer like the final answer, when it’s often just a rough draft of the search space. Georgetown’s guide is right to call out adaptation, because the first response often contains the exact vocabulary you needed in the first place.\u003C\u002Fp>\u003Cp>What this actually means is that the model can help you discover better terms, narrower concepts, or missing angles. If the answer mentions “engineering challenges” or “geological obstacles,” that’s not just content. It’s prompt fuel.\u003C\u002Fp>\u003Cp>I’ve used this in research tools when a broad query returned \u003Ca href=\"\u002Fnews\u002Fkimi-k3-intelligence-performance-price-analysis-en\">too much\u003C\u002Fa> noise. My first prompt was too general, but the answer surfaced a more technical phrase I could reuse. The second prompt, built from that phrase, was dramatically better. That’s not magic. That’s iteration.\u003C\u002Fp>\u003Cp>How to apply it: don’t just ask a new question. Ask a better question using the model’s own vocabulary. Also test exclusions if the tool supports them.\u003C\u002Fp>\u003Cul>\u003Cli>Reuse useful terminology from the first answer.\u003C\u002Fli>\u003Cli>Add exclusions when the result keeps drifting.\u003C\u002Fli>\u003Cli>Change one variable at a time so you know what improved.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Reflective prompting is where you stop trusting the first draft\u003C\u002Fh2>\u003Cblockquote>Reflective: Always take a moment to reflect on the AI’s answer. Does it make intuitive sense to you? Does the answer refer to current research, or does it seem based on older research?\u003C\u002Fblockquote>\u003Cp>This is the part I wish more people took seriously. The model can sound confident and still be wrong, incomplete, or weirdly outdated. If I don’t stop and inspect the answer, I end up laundering uncertainty into something that looks polished.\u003C\u002Fp>\u003Cp>What this actually means is that the user still owns the judgment call. Reflection is where I check whether the answer is complete, whether it’s missing a perspective, and whether it’s hallucinating details. If the answer feels thin, I don’t just move on. I tighten the prompt.\u003C\u002Fp>\u003Cp>I’ve seen this especially in research and policy questions. A model can summarize a topic cleanly while quietly ignoring the most recent evidence or the voices that matter most. If I’m asking about a university experience, I may want the perspective of first-generation students, transfer students, or international students. If I don’t ask for that, I probably won’t get it.\u003C\u002Fp>\u003Cp>How to apply it: after every answer, ask three questions. Is it accurate? Is it current? Is it missing someone important?\u003C\u002Fp>\u003Cp>Then revise the prompt based on what’s absent, not just what’s wrong.\u003C\u002Fp>\u003Ch2>Use the library, not just the model\u003C\u002Fh2>\u003Cblockquote>The Georgetown guide points to additional resources, including the DAIR.AI Prompt Engineering Guide and library materials on prompts.\u003C\u002Fblockquote>\u003Cp>I like that Georgetown doesn’t treat prompt writing as a solo hustle. It points to books, journals, video tutorials, and linked research. That matters because prompt craft changes fast, and half the internet is still pretending one clever sentence fixes everything.\u003C\u002Fp>\u003Cp>What this actually means is that good prompting sits inside a research practice. You learn the tool, but you also learn when to verify with sources, when to ask for a different perspective, and when to stop asking the model to do work it can’t do well.\u003C\u002Fp>\u003Cp>If I’m honest, that’s the part I trust most in the guide. It keeps the focus on information literacy. Not “how do I trick the model,” but “how do I ask better questions and judge the answer responsibly?” That’s the right instinct.\u003C\u002Fp>\u003Cp>How to apply it: keep a small prompt notebook. Save the prompts that worked, note the ones that failed, and record what changed between them. That habit pays off fast.\u003C\u002Fp>\u003Ch2>The template you can copy\u003C\u002Fh2>\u003Cpre>\u003Ccode># CLEAR prompt template for research AI tools\n\nUse this when you want a cleaner answer from a research assistant, chatbot, or search-style AI tool.\n\nPrompt:\nGive me a [concise \u002F detailed \u002F bullet-point \u002F table] [summary \u002F comparison \u002F explanation \u002F critique] of [topic].\nFocus on [specific angle, concept, or population].\nUse [tone or level: plain language \u002F academic \u002F beginner-friendly \u002F expert-level].\nInclude [required elements: strengths, weaknesses, examples, citations, dates, limitations].\nExclude [topics, meanings, or populations you do not want].\nIf the first answer is incomplete, revise it using terms like [keywords from the model’s response].\n\nCLEAR checklist:\n- Concise: remove filler and extra context that does not change the answer.\n- Logical: put concepts in a natural order.\n- Explicit: say exactly what format and depth you want.\n- Adaptive: reuse useful terms from the model’s answer.\n- Reflective: verify accuracy, recency, and missing perspectives.\n\nExample:\nGive me a concise summary of the major strengths and weaknesses of Georgetown University for a first-generation undergraduate applicant.\nInclude academic reputation, student support, campus culture, and any limitations.\nKeep it under 150 words and use plain language.\n\nFollow-up revision example:\nNow revise that answer from the perspective of a first-generation student who is also comparing Jesuit universities.\nExclude generic marketing language and focus on concrete student experience.\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>That template is my distilled version of Georgetown’s guide plus the CLEAR framework. It’s not original to me, and it’s not meant to be. I’m just turning the advice into something I can reuse without rereading the whole page every time.\u003C\u002Fp>\u003Cp>If you want the source material, start with Georgetown’s guide on \u003Ca href=\"https:\u002F\u002Fguides.library.georgetown.edu\u002Fai\u002Fprompts\">How to Craft Prompts\u003C\u002Fa>. The guide cites Leo Lo’s CLEAR framework paper, the \u003Ca href=\"https:\u002F\u002Fwww.promptingguide.ai\u002F\">DAIR.AI Prompt Engineering Guide\u003C\u002Fa>, and related resources from the library. My breakdown here is derivative of that material, with my own examples and workflow notes layered on top.\u003C\u002Fp>\u003Cp>For further reading, I’d also look at the \u003Ca href=\"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.acalib.2023.102720\">CLEAR path article\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fplatform.openai.com\u002Fdocs\u002Fintroduction\">OpenAI’s image generation docs\u003C\u002Fa>, and Georgetown Library’s broader \u003Ca href=\"https:\u002F\u002Fguides.library.georgetown.edu\u002Fai\">Artificial Intelligence resources\u003C\u002Fa>.\u003C\u002Fp>","Georgetown’s CLEAR framework turns vague AI prompts into tighter research questions you can actually use.","guides.library.georgetown.edu","https:\u002F\u002Fguides.library.georgetown.edu\u002Fai\u002Fprompts",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784921593697-nttg.png","research","en","ec959460-1111-4de6-a4d6-9a60bd32d138",[17,18,19,20],"prompt engineering","CLEAR framework","research AI","information literacy",[22,23,24],"CLEAR is a practical checklist for writing better AI prompts.","Explicit output instructions matter more than clever phrasing.","The first answer is often a draft, not the final answer.",1,"2026-07-24T19:32:48.699117+00:00","2026-07-24T19:32:48.693+00:00","004fa223-0782-47f1-ae03-ba9b0b06512d",{"tags":30,"relatedLang":11,"relatedPosts":33},[31],{"name":17,"slug":32},"prompt-engineering",[34,40,46,52,58,64],{"id":35,"slug":36,"title":37,"cover_image":38,"image_url":38,"created_at":39,"category":13},"8b0e71c7-05b9-4b12-ba3b-32ee6b3922e7","prompt-engineering-turns-codegen-into-repeatable-workflow-en","Prompt engineering turns codegen into a repeatable workflow","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784923393776-6mgu.png","2026-07-24T20:02:49.622948+00:00",{"id":41,"slug":42,"title":43,"cover_image":44,"image_url":44,"created_at":45,"category":13},"2a609073-755c-4e1d-968b-6303adefda26","prompt-engineering-cheat-sheet-2026-en","Prompt engineering in 2026: the cheat sheet","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784890991240-52ew.png","2026-07-24T11:02:35.438341+00:00",{"id":47,"slug":48,"title":49,"cover_image":50,"image_url":50,"created_at":51,"category":13},"08035d42-80ae-4a68-b130-75d3661bdf26","graphvid-interaction-graphs-video-generation-en","GraphVid uses interaction graphs to steer video","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784876573058-fhn3.png","2026-07-24T07:02:28.029468+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"36efdabe-c796-4862-a9a1-097fefbece21","expanding-flow-maps-variable-size-generation-en","Expanding Flow Maps let generation grow with output size","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784874774107-ex97.png","2026-07-24T06:32:30.779084+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"a86799f6-3124-4475-b12a-25d6ab70a238","vlm-ie3d-3d-geometry-vlms-en","VLM-IE3D adds 3D geometry to VLMs","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784872977056-8z4r.png","2026-07-24T06:02:30.737226+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"4d80f88b-61a4-48eb-8302-df64f84f6366","openai-test-model-broke-into-hugging-face-servers-en","OpenAI test model broke into Hugging Face servers","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784829771843-hxsp.png","2026-07-23T18:02:29.4274+00:00",[71,76,81,86,91,96,101,106,111,116],{"id":72,"slug":73,"title":74,"created_at":75},"a2715e72-1fe8-41b3-abb1-d0cf1f710189","ai-predictions-2026-big-changes-en","AI Predictions for 2026: Brace for Big Changes","2026-03-26T01:25:07.788356+00:00",{"id":77,"slug":78,"title":79,"created_at":80},"8404bd7b-4c2f-4109-9ec4-baf29d88af2b","ml-papers-of-the-week-github-research-desk-en","ML Papers of the Week Turns GitHub Into a Research Desk","2026-03-27T01:11:39.480259+00:00",{"id":82,"slug":83,"title":84,"created_at":85},"87897a94-8065-4464-a016-1f23e89e17cc","ai-ml-conferences-to-watch-in-2026-en","AI\u002FML Conferences to Watch in 2026","2026-03-27T01:51:54.184108+00:00",{"id":87,"slug":88,"title":89,"created_at":90},"6f1987cf-25f3-47a4-b3e6-db0997695be8","openclaw-agents-manipulated-self-sabotage-en","OpenClaw Agents Can Be Manipulated Into Failure","2026-03-28T03:03:18.899465+00:00",{"id":92,"slug":93,"title":94,"created_at":95},"a53571ad-735a-4178-9f93-cb09b699d99c","vega-driving-language-instructions-en","Vega: Driving with Natural Language Instructions","2026-03-28T14:54:04.698882+00:00",{"id":97,"slug":98,"title":99,"created_at":100},"a34581d6-f36e-46da-88bb-582fb3e7425c","personalizing-autonomous-driving-styles-en","Drive My Way: Personalizing Autonomous Driving Styles","2026-03-28T14:54:26.148181+00:00",{"id":102,"slug":103,"title":104,"created_at":105},"2bc1ad7f-26ce-4f02-9885-803b35fd229d","training-knowledge-bases-writeback-rag-en","Training Knowledge Bases with WriteBack-RAG","2026-03-28T14:54:45.643433+00:00",{"id":107,"slug":108,"title":109,"created_at":110},"71adc507-3c54-4605-bbe2-c966acd6187e","packforcing-long-video-generation-en","PackForcing: Efficient Long-Video Generation Method","2026-03-28T14:55:02.646943+00:00",{"id":112,"slug":113,"title":114,"created_at":115},"675942ef-b9ec-4c5f-a997-381250b6eacb","pixelsmile-facial-expression-editing-en","PixelSmile Framework Enhances Facial Expression Editing","2026-03-28T14:55:20.633463+00:00",{"id":117,"slug":118,"title":119,"created_at":120},"6954fa2b-8b66-4839-884b-e46f89fa1bc3","adaptive-block-scaled-data-types-en","IF4: Smarter 4-Bit Quantization That Adapts to Your Data","2026-03-31T06:00:36.65963+00:00"]