{
  "id": 20533549,
  "title": "Your AI startup may be a very fast nonsense factory",
  "url": "https://urgent.news/2026/11/18/your-ai-startup-may-be-a-very-fast-nonsense-factory",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-18T02:00:17.000Z",
  "source": {
    "name": "e27",
    "slug": "e27",
    "url": "https://e27.co/your-ai-startup-may-be-a-very-fast-nonsense-factory-20261109/"
  },
  "original_language": "en",
  "account": "I, Faheem Aizat Kamsan, a self-taught AI builder with a paramedic background, discovered that providing vague instructions to AI can lead to incorrect outputs with great confidence. During my work on OnTheRice between October 7th and 11th, I pushed beyond headline summaries to analyze emerging signals, historical market behavior, and current prices. I also focused on prediction scoring and made signals easier to search and filter. However, the challenging part was determining which outputs were worthy of survival. My research assistant continuously updates the database, which helps compare saved observations with new evidence, but it also creates a headache as every box is labeled \"important,\" and nobody remembers the contents. The retrieval-augmented generation research shows that combining generation with retrieved material is crucial, but the models can still miss relevant information buried deep within long inputs. To address this, I want the database to retain a trail of records, showing what produced them, how they changed, and who was involved. This way, a polished summary won't erase the messy origins of the data. My platform teaches me to question confident answers, comparing cryptocurrency prices displayed on public pages and explicitly stating that those comparisons are not verified one-hour returns. The distinction helps prevent confident stories based on unverified data. I also tightened the requirements for directional calls, which now must include a starting price, direction, and deadline. While AI can be a helpful team member, I still need to manage it and ensure it doesn't produce false information. My mentor, Daniel Tan, brings a marketing perspective to my world of building, reminding me to communicate with people about my creations. My cousin, Aidil Aqmal, handles security and scrutiny, ensuring the platform stands up to examination. I now spend more time checking what deserves publication instead of repeatedly collecting material. AI can inform my decision-making process, but ultimately, my responsibility remains to ensure the information is accurate and truthful.",
  "summary": "I know the temptation. I am Faheem Aizat Kamsan, a self-taught AI builder with a paramedic background. Give me a stubborn problem and I will keep building. Give AI a vague instruction and it may build the wrong thing with magnificent confidence. Between 7 and 11 October, my work on OnTheRice became a lesson in […] The post Your AI startup may be a very fast nonsense factory appeared first on e27 .",
  "key_points": [
    "Self-taught AI builder creates platform to analyze AI outputs",
    "Database retains trail of records to prevent erasure of data origins",
    "AI used to inform decision-making, but accuracy ultimately user's responsibility"
  ],
  "editors_take": "Prioritizing data provenance and rigorous verification over automated generation shifts the developer's role from a content collector to a critical gatekeeper against the inherent risks of confident AI hallucinations.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}