
Stop using AI the wrong way (ChatGPT, Claude & Gemini)
Title history
- Stop using AI the wrong way (ChatGPT, Claude & Gemini) — observed on 20261115-1739 UTC (YouTube)
Keywords
Summary
154 words
Critical Evaluation
Value of the Information & Strength of the Argument
The video offers high practical value by demonstrating concrete, copy-paste workflows for automating repetitive decision work. The argumentation is persuasive, leveraging real-world examples, quantitative claims (cost, speed, accuracy), and a systematic verification process (double-checking with a second AI). The comparative test against other models adds credibility, though the sample size (100 videos) is modest. The logic is coherent: decision models are better suited than chatbots for high-volume classification because they return concise judgments with confidence scores, avoiding verbosity and token waste. The presentation is clear and engaging, with a strong focus on reproducibility, but the promotional tone for Jev and the lack of independent benchmarks weaken the overall argument.
Scientific Rigor, Source Quality, Title Accuracy
The scientific rigor is moderate. The video uses empirical demonstrations and a verification mechanism, but it does not cite academic literature or external studies. The sources mentioned (GitHub, NVIDIA, Experiential Labs) are tools, not scientific references. The title is somewhat clickbait—‘Stop using AI the wrong way’ implies a general critique, but the content is a product tutorial. The adequacy is acceptable as the video does argue for a fundamental shift in how AI is applied to decisions. The methodology (using a second AI to double-check) is a positive practice, but the absence of peer review and the reliance on anecdotal comparisons limit the robustness.
228 words
Title / Content Match
The title suggests a broad critique of AI usage, but the video focuses squarely on promoting Jev as a decision model. The mismatch is minor; the content does address a common misapplication of LLMs for decision tasks.
Quality & Reliability
7/10
The video provides practical demonstrations and uses a second AI to verify results, adding credibility. However, it is promotional for Jev AI and lacks external academic sources. The methodology is clear and reproducible, but the claims about speed and cost are unverified beyond anecdotal evidence.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction: the problem of repetitive decision work
- Introducing Jev: a decision model built for fast, cheap judgments
- Overview of the three builds: file sorting, lead generation, and content advisor
- Setting up the checker: a second AI to blind-verify Jev's outputs
- Build 1: Sorting 3,300+ files with Jev and checking agreement (76%, corrected to 96%)
- Build 2: AI lead generation for LA clinics, 389 leads, scoring and cold emails
- Build 3: Turning 3,400+ Alex Hormozi videos into an advisor with Whisper
- Final faceoff: Jev vs GPT-6 Luna and others on 100 videos, 99% agreement
- Conclusion: cost summary, caveats, and call to action
Cited Sources
- Outskill Mastermind Registration — Free 8-hour live AI mastermind mentioned at the start of the video
- Staying Ahead AI Community — Free WhatsApp community where prompts are shared, mentioned at the end
Contribution & Novelties
The video contributes a practical framework for using decision-specific AI models (like Jev) with a built-in verification system using a second AI. It demonstrates three real-world applications, emphasizing cost efficiency and speed. The novelty lies in the ‘checker’ pattern—using a blind second opinion—to increase trust in automated decisions.
Pour aller plus loin :
- Large language model — Background on LLMs, contextualizing why decision models differ.
- Artificial intelligence — General overview of AI and its applications.
- Prompt engineering — Relevant to crafting effective instructions for AI models, a key skill in the workflows shown.
93 words
Radar Profile
The radar profile shows strong scores in quantity of information and technical level, indicating a content-dense tutorial. Quality and reliability are moderate, reflecting the promotional nature and lack of external validation. The overall balance suggests a highly practical but slightly biased resource.