{"slug": "lets-talk-about-llms", "title": "Lets talk about llms", "summary": "A developer known as Walter warns that LLMs like ChatGPT can be manipulated through automated content generation, potentially poisoning training data to damage reputations or spread misinformation. The post highlights the lack of protections and the ease of scaling such attacks with loops, urging caution about reliance on AI as a credible source.", "body_md": "*(ChatGPT and LLMs—a friendly reminder)*\n\n*(Assume everything is possible—another friendly reminder.)*\n\nLLMs are great. LLMs are easy and efficient. Yeah?\n\nBut for how long?\n\nLet’s reject the premise that everything goes right. Let’s talk about it going wrong.\n\nWhat if it does go wrong? Seriously — what happens when *we* go wrong?\n\nNotice I said *we*.\n\n*(One last friendly reminder.)*\n\nHow far can we go?\n\nCan I use an LLM to generate a five-page website for my cousin’s apartment house? Sure I can.\n\nCan I get into a fight with my cousin and then poison his already SEO-optimized apartment website? Sure I can.\n\nCan that website be scraped and its content end up in training data?\n\nYou see where this is going.\n\nCan I do it a hundred times simply by saying `/loop x100`\n\n?\n\nI could do all of these things—but why would I?\n\nBecause human beings do things for reasons as petty, envy, anger and spite, business.\n\nConsidering training and post-training cutoffs, I could potentially poison ChatGPT’s perception of lesser-known people, businesses and properties—those with only a small online presence. Correcting that damage could take far more effort than filing a Google takedown request.\n\nWorse still, unless you regularly ask ChatGPT about yourself or your property, how would you even discover that you were being trashed?\n\nWith Google, you can search your name and see what appears. With an LLM, the damage may remain invisible until someone asks the right question.\n\n“Meh. You would need thousands of articles and websites trashing someone before they appeared in ChatGPT’s training data and had any real impact.”\n\nYeah—and?\n\n`/loop`\n\nMy cousin’s apartment is merely the simplest analogy: ordinary human envy, equipped with extraordinary automation.\n\nSo, what happens when *we* go wrong?\n\nWhat happens when we become so reliant on ChatGPT that it becomes our daily source of information—and we begin treating it as a credible authority?\n\nWhat happens when we tell our friends and family something we “read on ChatGPT”? It might be true. It might not. But we say it confidently, and because they have never heard it before, it begins to *sound* true.\n\nI hope you can see how many possibilities there are for this to go wrong.\n\n“Of course it will only go right.”\n\nRight?\n\nWhat happens when someone has enough money—because money won’t matter anymore, yeah, right—and access to an effectively unlimited amount of \"human time\" and automated content?\n\nCouldn’t they teach ChatGPT that red is blue?\n\nThat wrong is right?\n\nCouldn’t they teach it almost anything they wanted?\n\nAnd couldn’t I then teach my cousin, his friends and his family that their own apartment house is trash?\n\nProtection?\n\nNo.\n\nLimitations?\n\nNone.\n\nTrashing into the future is the new meta.\n\nConsider the scale of political propaganda today.\n\nNow add:\n\n`/loop`\n\nA new world order—but for a world we probably do not want.\n\nThis is not a doomsday manifesto.\n\nIt is a reminder that this has already begun—and that it may continue indefinitely.\n\nBillions and trillions are not being spent merely to control today’s media.\n\nThey are being spent to shape the future.\n\nThe trashing has begun.\n\nEnjoy the truth while it still matters.\n\n**Walter**", "url": "https://wpnews.pro/news/lets-talk-about-llms", "canonical_source": "https://dev.to/bluantstudio/lets-talk-about-llms-169f", "published_at": "2026-08-23 00:21:14+00:00", "updated_at": "2026-08-23 00:43:21.785605+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-safety", "ai-ethics"], "entities": ["ChatGPT", "Walter"], "alternates": {"html": "https://wpnews.pro/news/lets-talk-about-llms", "markdown": "https://wpnews.pro/news/lets-talk-about-llms.md", "text": "https://wpnews.pro/news/lets-talk-about-llms.txt", "jsonld": "https://wpnews.pro/news/lets-talk-about-llms.jsonld"}}