a16z Partner Says Judgment Training is Scarce in the AI Era
a16z partner Tim Sullivan stated that after AI reduces content production costs, the surge of low-quality "slop" content and the ethical panic surrounding AI creation are not new phenomena; similar controversies have arisen from the decline in media costs throughout history, from 18th-century London Grub Street, cheap newspapers, popular novels, television, blogs to social media.
He believes that an oversupply of content does not automatically mean a decline in cultural quality. Historically, low-cost production has expanded both the scale of creators and audiences, generating a large amount of derivative content, and providing space for experimentation for a few new forms, new authors, and new distribution mechanisms; AI has simply accelerated this cost compression process to text, images, videos, and code.
The article cites social science perspectives indicating that "taste" is not a single aesthetic talent but may consist of various mechanisms: professional judgment in recognizing quality, the ability to access cross-domain information, experience in filtering potential successes amid uncertainty, and evaluation frameworks formed through long-term feedback.
Sullivan mentioned that research from Columbia University shows that whether a work becomes a hit is highly influenced by social factors and path dependence, making it difficult even for experts to reliably predict the final winners. Experts may be better at assessing certain clear quality dimensions but cannot fully foresee whether a work will gain network effects among specific groups, platforms, and time points.
He referenced sociologist Ron Burt's "structural hole" theory: the most innovative employees are usually not deeply embedded in a single tight-knit community but are positioned between previously disconnected groups. This "intermediary" position allows individuals to access different information, languages, and problem frameworks, thereby forming new combinations that others may not see.
AI can quickly search, summarize, and reorganize information across fields, but it is difficult to fully replicate the tacit knowledge, trust, shared experiences, and long-term relational responsibilities that humans accumulate in multiple social networks. Therefore, while AI can expand the breadth of information, it cannot automatically generate the judgment that comes from real community participation.
Sullivan believes that after the widespread adoption of AI, the high-value roles of humans will lean more towards verification, selection, and judgment; these abilities rely on apprenticeship, repeated feedback, and professional practice. If companies reduce entry-level positions and training processes to save costs with AI, there may be a future talent gap in mature judgment capable of reviewing AI outputs.
In market mechanisms, AI will lower the prices of initial drafts, basic designs, and routine analyses, but will raise the value of credible filtering, editorial responsibility, domain validation, community connections, and final decision-making. Beneficiaries include organizations with high-quality feedback systems, real user networks, professional review mechanisms, and apprenticeship training capabilities; the pressured parties are content producers relying solely on bulk generation, lacking verification standards and brand trust.
Source: Public Information
ABAB AI Insight
Sullivan's core rebuttal is not that "slop does not exist," but rather to place it back in the context of technological history. Cheap writers of Grub Street, penny papers of the 19th century, popular novels, television, blogs, and social media have all raised concerns among critics that low-cost content would destroy public culture; yet each round of production expansion has also restructured the power dynamics between authors, editors, distributors, and audiences. The uniqueness of AI lies in its ability to not only lower distribution costs but also bring the first-step production costs of writing, drawing, editing, translating, and programming close to zero.
As a result, the capital path shifts from "production scarcity" to "filtering scarcity." When texts, images, and code can be generated infinitely, value no longer primarily comes from completing initial drafts, but from who can set tasks, verify facts, identify errors, bear consequences, and maintain stable distribution relationships. Content platforms will continue to gain traffic from scaled generation, but corporate clients, professional users, and high-trust consumers will be more willing to pay for editing, review, vertical knowledge, copyright clarity, and accountability.
Burt's structural hole theory provides a more actionable explanation than "innate taste": the advantage of innovators often comes from being positioned between different communities rather than simply having more information. AI can read materials from different fields, but it lacks a natural social position; it cannot personally attend industry conferences, bear the costs of collaborative failures, or build trust in multiple long-term relationships. Thus, human "taste" is not just a preference but a comparative ability and sense of responsibility formed after moving across communities.
This belongs to technological substitution. AI first replaces standardizable initial draft labor, not final judgment; however, if companies eliminate entry-level positions, apprenticeship feedback, and real project experience, they will weaken the infrastructure for cultivating future evaluators. The mechanism is that judgment is not a static skill but a capability formed after seeing errors, receiving feedback, and observing how seasoned individuals weigh decisions. Short-term cost reductions in training can improve efficiency, but in the long run, they may lead organizations to lose the ability to recognize when AI makes mistakes.
ABAB News · Cognitive Laws
The cheaper the content, the more expensive the judgment.
AI can search across fields, while humans must bear the consequences across communities.
Cutting apprenticeship costs will ultimately lead to a judgment gap.