This video delivers a sharp critique of a recently released prompt resource from OpenAI for ChatGPT, labeling it as "terrible" and a significant step backward for AI education. The speaker argues that the provided prompts are overly simplistic, generic, and lack the necessary context to be useful for professional applications, using a flawed example for GDPR compliance to illustrate the point. He warns that such poor educational materials encourage a superficial understanding of AI, which could trap workers in a "messy middle" of adoption, leaving them unprepared for the technology's exponential advancement. The speaker advocates for a workflow-centric approach to AI training, where teams identify specific pain points and learn to build complex, context-rich prompts to solve them. He calls on AI model makers to provide better resources with clear progression paths and announces he is creating his own superior prompt pack in response.
Critiquing Poor AI Education and Advocating for a Workflow-Centric Approach
0:00 Criticism of ChatGPT's New Resource: The speaker immediately criticizes OpenAI's new prompt resource as "absolutely terrible," highlighting the poor state of AI education from industry leaders. The prompts are condemned for being generic one-to-two-line examples that are not useful for serious applications.
0:32 Deconstructing a Flawed Prompt: An example prompt for engineers on GDPR/CCPA compliance is analyzed. The speaker points out that it fails to include critical context such as data schema, storage locations, or the existing tech stack, treating the LLM like a simple search engine rather than an intelligence.
1:34 The Danger of the "Messy Middle": The speaker expresses concern that low-quality resources will trap knowledge workers in a "messy middle" of AI adoption. They will learn just enough to be ineffective and will ultimately be left behind by AI's exponential progress.
2:09 AI Requires Continuous Learning: A key takeaway is that AI is not a "one-and-done" software adoption but a "moving train." Users must commit to continuous and deep learning to keep pace, as superficial prompting skills will quickly become obsolete.
4:11 A Better Approach to AI Training: The speaker outlines a superior method for AI education. Instead of generic prompts, training should be grounded in real-world use cases by first identifying a team's specific pain points and manual workflows, then building tailored AI solutions for them.
5:53 Most AI Potential is Untapped: The speaker asserts that most organizations are only scratching the surface of AI's capabilities, often limiting its use to simple tasks like drafting emails. He estimates that 80-90% of the potential AI opportunities within workflows remain untouched.
6:53 It's the Method, Not the Model: The specific AI model (e.g., Copilot, ChatGPT, Claude) is less important than how it is used. Even less advanced models can be powerful when applied thoughtfully to well-defined workflows. The primary gap is in human training and strategy.
8:44 A Call for Better Education: Model makers like OpenAI are urged to provide better educational resources that offer clear on-ramps for beginners but also teach scalable principles (like the importance of context and goals) to help users advance.
10:22 Speculating on OpenAI's Motive: The speaker suggests the prompt pack feels like a "defensive gesture" created merely to check a box for enterprise sales contracts, rather than a genuine effort to educate users.
11:03 The Bar is Rising Exponentially: To illustrate AI's rapid advancement, the speaker cites an example where Anthropic's Sonnet 4.5 model autonomously rebuilt a version of Slack by writing 11,000 lines of code in 30 hours. This demonstrates the new standard for productivity that workers and teams must prepare for.