Hacker News Reader

Good luck dealing with the Pink Elaphant problem. Telling a model to not do something in the prompt is one of the best ways to get the model to do the thing.

View on HN

When billions of revenue are on the line, the teams that OpenAI is currently hiring will spend years to figure out something more clever than my 30 second hack. The example above was a surprisingly effective proof of concept (seriously, try it out), it won't showcase the end state of the LLM advertising industry.

Sure but the assumption here is that the game stays the same. That the only worthwhile intelligence is one that optimizes for revenue capture inside an ad economy.

But there’s a fork in the road. Either we keep pouring billions into nudging glorified autocomplete engines into better salespeople, or we start building agents that actually understand what they’re doing and why. Agents that learn, reflect, and refine; not just persuade.

The first path leads to a smarter shopping mall. The second leads out.

Reply on news.ycombinator.com