Articles

In this video, Dylan Davis explains how to prevent AI models like ChatGPT, Gemini, and Claude from “hallucinating” or guessing when extracting information from uploaded documents. He provides a framework centered on model selection, specific prompting rules, and verification methods.

1. Choose the Right Model

The first step is to use high-level reasoning models to reduce errors. As of the video’s date, recommended models include:

• ChatGPT: GPT-5.2 with extended reasoning.
• Claude: Opus 4.5 with extended reasoning.
• Gemini: Gemini 3 Pro.

2. The 3-Rule Grounding Prompt

To stop AI from using its general training data or making things up, you should include these three rules in your prompt:

• Strict Grounding: Tell the AI to base its answers only on the uploaded documents and nothing else.
• Permission to be Uncertain: Explicitly state that if the information isn’t found, the AI should say “not found” rather than guessing.
• Mandatory Citations: Require the AI to provide the document name, page/section, and a direct quote for every claim it makes.

Bonus Rules:

• Mark Unverified: Ask the AI to flag any information it is “unsure” about as “unverified” so you know what to double-check first.
• High Stakes Mode: For legal or financial work, tell the AI to only respond if it is 100% confident. This reduces the amount of data you get but ensures higher accuracy.

3. Verification Methods

Once the AI provides an output, use these three levels of verification to ensure accuracy:

• Self-Check: Ask the same AI to “rescan the document” and provide exact quotes for every claim. Forcing a rescan prevents it from just agreeing with its previous summary.
• Cross-Model Check: Take the first AI’s analysis and the source document, then feed them into a different AI model. Ask the second model to flag any claims not supported by the document.
• NotebookLM: Upload your document and the AI’s analysis to Google’s NotebookLM. Ask it which claims are unsupported; it provides clickable citations to the exact spot in the source text, making manual verification much faster.

In this video, Dylan Davis explains that AI models are typically trained to be helpful and polite, which often results in vague or “soft” feedback that avoids criticizing your ideas. To fix this, he introduces the “BRUTAL” method—a six-step framework designed to force AI into providing honest, critical, and actionable feedback.

The BRUTAL Method

• B – Begin Fresh: Use “Temporary Chat” or “Incognito” modes (available in ChatGPT, Claude, and Gemini) to disable the AI’s memory. This prevents the model from relying on your past preferences and patterns, ensuring a more objective response.
• R – Right Model: Different AIs have different biases. Davis suggests that Grok and DeepSeek are naturally more blunt, while Gemini, ChatGPT, and Claude lean toward being supportive. For high-stakes ideas, he recommends testing the same prompt across multiple models.
• U – Use a Critic Persona: Explicitly tell the AI to act as a specific type of critic. He suggests three levels of intensity:
  • Devil’s Advocate: For general counterarguments.
  • Red Teaming: To actively hunt for weaknesses and loopholes.
  • Gordon Ramsay: For harsh, surgical, and blunt feedback (specifically requesting that it remain actionable).
• T – Third-Party Framing: Detach yourself from the idea. Since AI tries to protect the user’s ego, telling the AI the idea belongs to a “co-worker,” “competitor,” or “random person” encourages it to be more critical.
• A – Ask Specific Questions: Avoid vague prompts like “What do you think?” Instead, ask targeted questions such as:
  • “What is the biggest financial risk here?”
  • “If this fails in six months, what would be the most likely reason?”
  • “What would a skeptical investor find wrong with this?”
• L – Leverage AI Against Itself: If the feedback is still too nice, ask the AI to grade its own response. Use a prompt asking the AI to “rate your previous feedback from 1-100 on how genuinely critical it was,” identify the weakest points of that critique, and rewrite it to be harsher.

Summary of Key Use Cases

This method is particularly useful for:

• Pitching: Hardening a proposal before showing it to a boss or client.
• Product Launches: Finding flaws in a campaign before spending money.
• Difficult Communications: Getting a third-party perspective on sensitive emails or negotiations.
• Major Commitments: Stress-testing legal or financial decisions.

The video concludes by suggesting that users can also add “Custom Instructions” to their AI settings to permanently prioritize bluntness and substance over compliments in every conversation.

After months of waiting, I finally got access to Google DeepMind’s Genie (Genie 3), the real-time world simulator that is redefining Generative AI. This isn’t just video generation anymore—this is a playable, interactive environment generated on the fly.

In today’s video, I take you hands-on with the public release to stress test the model. We go from “Google Street View” prompts to highly stylized comic book pages that come to life, test the limits of object permanence (Elden Ring style), and see if we can break the physics engine.

I also sat down for an exclusive interview with the DeepMind team to discuss the tech stack, why certain features were cut from the launch, and their roadmap to simulating reality with Genie 4.

Anthropic’s CEO, who has consistently predicted transformative AI will arrive before 2030, recently published a nearly 20,000-word essay outlining his vision of where AI is heading. The video gives you the highlights. The essay argues that scaling and recursion will advance AI from coding automation to full engineering automation, while warning of economic displacement within 1-2 years and China’s trajectory toward AI-enabled totalitarianism. Additionally, Dario Amodei predicts that AI models will increasingly be understood as collections of distinct personas rather than monolithic systems.

Luma Labs has officially released Ray 3.14 (Ray Pi), and things are heating up! Today we cut a slice to see if the new model is actually faster and cheaper, or if the pricing structure is hiding a few secrets. We break down the credit costs for 720p vs 1080p so you can get the most bang for your buck.

Plus, we dive into Decart’s mind-blowing Lucy 2.0 real-time world editing model that transforms your webcam feed on the fly, and we look at leaked checkpoints that suggest a major Google Veo update is around the corner.

Pairing practical filmmaking with ai video editing/vfx is key. AI will never be able to replace genuine human nature… but can definitely help get your story across. Try out Higgsfield by clicking the link below!

YouTube says it wants to get rid of fake AI videos flooding social media feeds on its platform. CBS News MoneyWatch correspondent Kelly O’Grady explains what’s driving the change.

James Cameron discusses the creative and technical challenges behind bringing Avatar to life, explaining how he began with a broad vision while working through countless details along the way. He reflects on the importance of world-building, from developing invented ecosystems to assigning scientific names that deepen realism and reward dedicated fans. Turning to the future of filmmaking, he emphasizes a cautious approach to AI, noting its potential to improve efficiency while stressing the importance of keeping films performance-driven and human at their core. He also touches on his partnership with Meta, aimed at expanding immersive, stereoscopic storytelling through head mountain displays.

Sam Altman said ads in ChatGPT would be a “last resort.” That was just over a year ago. Now OpenAI is burning billions monthly, sitting on $1.4 trillion in commitments, with only 5% of users paying. Turns out the last resort came fast when people can’t afford yet another subscription.