The War on Slop – swyx

Key Concepts

• Slop: Low-quality, inauthentic, or inaccurate content generated by humans or AI. The antithesis of “kino” (high-quality, engaging content).
• Kino: High-quality, engaging content, often used as a benchmark against “slop.”
• Brandolini’s Law: The amount of energy needed to refute misinformation is an order of magnitude greater than the energy needed to produce it.
• Fix’s Law of Anti-Slop: The amount of taste needed to fight slop is an order of magnitude greater than that needed to produce it.
• Computer Use (Devon): Autonomous AI agents capable of operating complex applications, including IDEs, for automation.
• Code Maps: Utilizing AI to scale code-based understanding, aiding in identifying and combating code slop.
• Modularity (Greg Brockman’s concept): Maintaining clear boundaries between human-designed components and AI-generated code.
• Context Rot: The degradation of information and understanding over time, addressed by utilizing sub-agents.

The War on Slop: A Summary of the AI Engineer Summit Keynote

This keynote, delivered at the AI Engineer Summit, centers on a critical issue facing the rapidly expanding AI landscape: the proliferation of “slop” – low-quality, inauthentic, or inaccurate content. The speaker frames this as an “asymmetric war” requiring a conscious and concerted effort to maintain quality and taste within the industry.

The Evolution of the AI Engineer Summit & the Rise of Slop

The speaker outlines the evolution of the AI Engineer Summit, starting with a focus on the emergence of the “AI engineer” role, expanding to include leadership, then concentrating on “model labs.” Recent summits have broadened the scope to encompass AI designers and code, with a strong emphasis on curation. The speaker notes that while growth is important, it must not come at the expense of quality. The concept of “slop” has become increasingly relevant, even being considered for the Oxford English Dictionary’s 2024 Word of the Year (losing to “brain rot”).

Defining Slop & Distinguishing it from Quality (“Kino”)

The speaker clarifies that “slop” isn’t exclusive to AI-generated content; humans are equally capable of producing it. The opposite of slop is “kino,” representing high-quality, engaging work. Examples are provided to illustrate the contrast: Netflix’s “K-pop demon hunters” (kino) versus “Electric St” (slop), and the varying quality of Sora videos (referencing Paul Rambles’ work). The speaker emphasizes that even successful trends can degenerate into slop over time, referencing a now-defunct AI slides company (which used to be used for keynotes) as an example of a promising idea that failed. The infamous final season of “Game of Thrones” is also cited as a human-generated example of slop.

Slop Across Domains: Models, Ideas, and Code

The issue of slop extends beyond content creation. The speaker highlights how the same startup idea can manifest as either “kino” or “slop,” depending on execution. Different approaches to “vibe coding” are presented, with one considered superior. The speaker also points to the tension between exponential growth charts, noting that one feels more authentic (“keeno”) than the other (“slop”).

Specifically regarding code, the speaker cites examples from this year: the potential for two engineers to create tech debt equivalent to 50, and instances of exposing private data of millions of users. Even claims of models running autonomously for extended periods (30-60 hours) are deemed “sloppy” without accompanying information about code quality. The speaker advocates for “autonomy without accountability” – a parallel to “no taxation without representation.”

Addressing Slop with AI: Laws & Tools

The speaker introduces “Fix’s Law of Anti-Slop,” stating that the “amount of taste needed to fight slop is an order of magnitude bigger than that needed to produce it.” This acknowledges the difficulty in combating low-quality content. However, the speaker also proposes that AI can be used to fight slop.

Several tools and approaches are highlighted:

• AI News: The speaker’s side project, a newsletter that intentionally refrains from publishing when there’s no significant news, demonstrating a commitment to quality over quantity.
• Prompting for Quality: Mahash and Barry’s work demonstrates that explicitly instructing AI models not to produce slop significantly improves output.
• Code Maps: Utilizing AI to scale code-based understanding, helping identify and address code slop. The speaker directs attendees to the Cognition team for further details.
• Computer Use (Devon): Autonomous AI agents like Devon can automate complex tasks, including website updates, reducing the potential for human error and slop.
• Sub-Agents for Context Rot: Addressing the degradation of information over time through the use of sub-agents.
• Modularity (Greg Brockman): Maintaining clear boundaries between human-designed components and AI-generated code, as advocated by Greg Brockman, promotes quality and control.

Brandolini’s Law & the Asymmetric War

The speaker draws a parallel to Brandolini’s Law – the disproportionate effort required to refute misinformation compared to creating it – emphasizing the uphill battle against slop. The decreasing cost of generating tokens (dropping 100-1000x annually) further exacerbates the problem, making it easier to produce low-quality content.

The Call to Action: “No More Slop”

The keynote culminates in a rallying cry: “No more slop.” The speaker encourages the audience to adopt this mantra in various scenarios: when faced with demands for quantity over quality from their superiors, when confronted with engagement-bait algorithms, and in all their work. The speaker emphasizes the importance of taste and elevating standards within the AI community.

Conclusion

The keynote serves as a critical intervention within the AI community, urging a conscious rejection of “slop” and a commitment to quality, authenticity, and taste. The speaker presents a compelling argument that while AI can contribute to the problem, it can also be a powerful tool in combating it, provided that a focus on quality and accountability remains paramount. The core takeaway is a call to action – a collective commitment to raising the bar and prioritizing substance over superficiality in the rapidly evolving world of AI.