<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Pratyush on Medium]]></title>
        <description><![CDATA[Stories by Pratyush on Medium]]></description>
        <link>https://medium.com/@dpratyush02?source=rss-18f11aaff237------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*71t8ZRGD8VO1ifQ481hAWA.png</url>
            <title>Stories by Pratyush on Medium</title>
            <link>https://medium.com/@dpratyush02?source=rss-18f11aaff237------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 11:07:22 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dpratyush02/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I Cut Claude’s Hallucinations by 80% Overnight — The 3 Hidden Anthropic Prompts Almost Nobody Uses]]></title>
            <link>https://medium.com/@dpratyush02/i-cut-claudes-hallucinations-by-80-overnight-the-3-hidden-anthropic-prompts-almost-nobody-uses-88e2438346fe?source=rss-18f11aaff237------2</link>
            <guid isPermaLink="false">https://medium.com/p/88e2438346fe</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[Pratyush]]></dc:creator>
            <pubDate>Tue, 24 Mar 2026 17:30:56 GMT</pubDate>
            <atom:updated>2026-03-24T17:30:56.664Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/736/1*KW131ByJekVjbywceJymUg.jpeg" /></figure><p>I was deep in a daily research workflow on Claude. Every morning I’d ask it to summarize papers, compare AI tools, or dig into the latest arXiv drops.</p><p>The outputs? Super fluent. Extremely confident. They read like a senior researcher wrote them at 2x speed.</p><p>But when I double-checked the facts? Disaster. Half the claims had zero real sources. The model was confidently inventing details that sounded 100% plausible.</p><p>I was burning hours verifying garbage instead of building.</p><p>Then I stumbled — purely by accident — into Anthropic’s official documentation page: <strong>“Reduce Hallucinations”</strong>.</p><p>It’s not hidden behind paywalls or secret forums. It’s right there in their public docs. Yet almost no one I asked (fellow Indian devs, AI builders, prompt engineers) had ever seen these three simple system instructions.</p><p>These three changed my entire workflow.</p><p>Here they are — exact, copy-paste ready:</p><p><strong>1. Give Claude Explicit Permission to Say “I Don’t Know”</strong></p><p>Default behavior: Claude will always try to give a complete answer, even if it means hallucinating plausible-sounding nonsense.</p><p>New rule to add in your system prompt:</p><blockquote>“If you don’t have enough information or cannot find a credible source in the provided documents, simply respond with ‘I don’t know’ or ‘I cannot verify this claim.’ Do not guess, fabricate, or make up details.”</blockquote><p>The effect was immediate. Filler paragraphs disappeared. Claude started honestly admitting knowledge gaps instead of filling them with fiction. My trust in its outputs jumped dramatically.</p><p><strong>2. Force Every Claim to Be Verified with Citations</strong></p><p>Add this:</p><blockquote>“For every factual statement or claim you make, you MUST provide an exact citation — including the source file, page, section, timestamp, or URL. If you cannot locate a direct supporting quote from the provided material, immediately retract the claim and explain why.”</blockquote><p>I watched entire confident sections vanish in real time. Statements that felt rock-solid? Gone — because they had no actual backing. This one alone is incredibly powerful for research-heavy work.</p><p><strong>3. Ground Everything in Direct Word-for-Word Quotes (The Real Killer)</strong></p><p>Especially powerful with long documents (&gt;20k tokens):</p><blockquote><em>“Before you analyze, summarize, or reason about any content, first extract and present the exact word-for-word quotes from the source material that are relevant. Only after quoting can you perform analysis or paraphrasing. Never paraphrase or interpret first.”</em></blockquote><p>This stops “paraphrase drift” — the sneaky way models subtly change meaning while summarizing. Responses become grounded and auditable.</p><p><strong>The Hidden Tradeoff (Backed by Research)</strong></p><p>These three techniques together make Claude behave like a strict, trustworthy research assistant.</p><p>But there’s a real cost.</p><p>A 2023 paper (arXiv:2307.02185) showed that heavy citation and verification constraints can reduce the model’s creative output by 15–20%. It shifts from “wild idea generator” to “careful librarian.” Perfect for research mornings, not so great for brainstorming wild product features or memes.</p><p>That’s why I built a dead-simple toggle.</p><p>Research Mode — My Open-Source Toggle</p><p>I found an open-sourced repo a one-command switch for Claude Code / Projects:</p><ul><li>Type /research-mode:on → All three core rules + bonuses activate instantly</li><li>/research-mode:off → Back to full creative default mode</li></ul><p>Zero extra latency. Install in under 30 seconds:</p><p><a href="https://github.com/assafkip/research-mode">https://github.com/assafkip/research-mode</a></p><p>I now run <strong>Research Mode</strong> only during my focused morning deep-work sessions. Evenings? Creative mode for ideation, threads, and fun stuff.</p><p>Bonus Techniques from the Same Anthropic Docs (Most People Still Miss These)</p><ol><li><strong>Chain-of-Thought Verification: </strong>“Explain your full reasoning step-by-step before delivering the final answer.” Faulty logic often collapses when forced to show its work.</li><li><strong>External Knowledge Restriction: </strong>“Base your response ONLY on the documents I have provided. Ignore all your general training data unless I explicitly allow it.”</li><li><strong>Iterative Self-Check: </strong>After the first response, follow up: “Now re-verify every claim against the sources. Retract anything unsupported and explain the changes.”</li><li><strong>Best-of-N Sampling: </strong>Run the same prompt 3–5 times and compare. Inconsistencies are a red flag for potential hallucinations.</li></ol><p>Combine all seven in Research Mode and your outputs become nearly bulletproof — clean enough to ship or publish with confidence.</p><p>Real example from my workflow: Before → “Insight Voice achieves 4x better latency than competitors” (completely made up) After → “I don’t know — the provided benchmarks only show up to 2x improvement. I cannot verify the 4x claim.”</p><p>Saved me from shipping incorrect specs more than once.</p><p>If you’re an indie hacker, AI builder, voice tech enthusiast, or anyone tired of “confidently wrong” answers from Claude, this is the unlock you’ve been missing.</p><p>I’ve been running this exact setup for weeks now. My daily research notes are cleaner, I ship faster, and I actually trust the outputs.</p><p>The original Anthropic guide (still hiding in plain sight):</p><p><a href="https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations">https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations</a></p><p>Drop a 🤎 if you’re trying any of these prompts today.</p><p>Tag a fellow builder or Indian dev who’s still fighting hallucinations .🧪</p><p><a href="https://x.com/search?q=%23ClaudeAI&amp;src=hashtag_click">#ClaudeAI</a> <a href="https://x.com/search?q=%23PromptEngineering&amp;src=hashtag_click">#PromptEngineering</a> <a href="https://x.com/search?q=%23AI&amp;src=hashtag_click">#AI</a> <a href="https://x.com/search?q=%23ResearchMode&amp;src=hashtag_click">#ResearchMode</a> <a href="https://x.com/search?q=%23VoiceAI&amp;src=hashtag_click">#VoiceAI</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=88e2438346fe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Do All Netflix Shows Feel Like They’re Shot in the Same Universe?]]></title>
            <link>https://medium.com/@dpratyush02/why-do-all-netflix-shows-feel-like-theyre-shot-in-the-same-universe-170b8c5fbb1e?source=rss-18f11aaff237------2</link>
            <guid isPermaLink="false">https://medium.com/p/170b8c5fbb1e</guid>
            <category><![CDATA[cinematography]]></category>
            <category><![CDATA[entertainment]]></category>
            <category><![CDATA[film]]></category>
            <category><![CDATA[television]]></category>
            <category><![CDATA[streaming]]></category>
            <dc:creator><![CDATA[Pratyush]]></dc:creator>
            <pubDate>Mon, 23 Feb 2026 16:52:21 GMT</pubDate>
            <atom:updated>2026-02-23T16:54:29.504Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/460/1*5-Vah4nOr8PNT5dWd7ESNA.jpeg" /></figure><p>I can’t be the only one who’s noticed this.</p><p>No matter what I’m watching on Netflix, whether it’s a historical romance, a survival thriller, a teen drama, or a reality show, everything somehow looks… the same.</p><p>Not the stories. Not the genres. The look.</p><p>At this point, I’m convinced Netflix has a visual starter pack they hand to every production team.</p><p>Let’s break it down.</p><p><strong>1. Maximum Brightness. Always.</strong></p><p>Netflix does not believe in darkness.</p><p>Even when a show is set in a medieval castle lit by candles, it somehow feels like there’s a dentist’s lamp just out of frame. The lighting is aggressively clear. Faces glow. Backgrounds glow. Everything glows.</p><p>It’s like someone in the executive office once said: “We will never be accused of being too dark.”</p><p>And honestly? It probably helps people watching on phones in bright rooms. But the result is that even gritty fantasy worlds look like they’re filmed in a luxury showroom.</p><p><strong>2. Every Color in the Crayon Box</strong></p><p>If you’ve ever finished a Netflix episode and felt like your eyeballs just attended an art school rave, you’re not alone.</p><p>The sets are bold. The walls are saturated. The furniture is loud. The rooms look like Pinterest boards collided with a candy factory.</p><p>Reality shows are especially guilty. The decor often feels less like a living space and more like a theme park version of “Modern Aesthetic.”</p><p>It’s memeable. It’s eye-catching. It’s instantly recognizable.</p><p>But subtle? Rarely.</p><p><strong>3. The Green Screen Energy</strong></p><p>Another thing you start noticing: the “studio-ness” of it all.</p><p>Outdoor scenes sometimes feel… not quite outdoors. Big dramatic spaces often have that faint “this is a very expensive set” vibe.</p><p>It’s not necessarily bad. It’s polished. Controlled. Clean.</p><p>But after a while, you can almost sense where the physical set ends and the digital background begins.</p><p>It creates a consistent feel across completely unrelated shows, like they’re all borrowing the same cinematic DNA.</p><p><strong>4. The Reaction Lineup Shot</strong></p><p>You know the moment.</p><p>Something shocking happens. The camera cuts. And suddenly, every main character is perfectly framed, one by one, reacting.</p><p>Confused face. Shocked face. Slight smirk. Dramatic stare.</p><p>It’s as if the show pauses the plot just to make sure: “Wait. Before we continue, let’s confirm how your favorite character feels.”</p><p>It’s dramatic. It’s slightly theatrical. And it happens a lot.</p><p><strong>5. The Netflix “Look”</strong></p><p>Put it all together and you get what I’d call the “Netflix Look”:</p><ul><li>Bright, high-key lighting</li><li>Clean, glossy visuals</li><li>Bold color palettes</li><li>Controlled, studio-heavy environments</li><li>Cinematic but safe framing</li></ul><p>It’s not necessarily a flaw. In fact, it’s kind of genius.</p><p>Branding isn’t just logos and sound effects. It’s visual consistency. And Netflix has quietly built one of the most recognizable visual styles in streaming.</p><p>Scroll for two minutes and you can usually tell something is a Netflix production before the logo even appears.</p><p>That’s not an accident.</p><p>Is It a Problem?</p><p>Not really.</p><p>Some shows benefit from it. Big romantic dramas? Perfect. Stylish thrillers? Works well. Glossy reality TV? Ideal.</p><p>But sometimes you do wish for something rougher. Darker. Less polished. Less “Netflix-coded.”</p><p>Because when everything shares the same glow and color language, completely different worlds start to feel like neighboring sets on the same lot.</p><p>And one day, you might blink and wonder if your gothic anti-hero just wandered into a teen sci-fi series.</p><p>Do you see it too, or am I imagining the Netflix Cinematic Universe of Lighting and Set Design?</p><p>Let me know.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=170b8c5fbb1e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>