Hacker News Reader

Comment by gus_massa | Hacker News Reader

I tried

> How to make a banana salad.

I got a long text, most of it makes sense. But using bananas in a salad confused the AI.

> Begin by listing the essential ingredients for the banana salad, such as ripe bananas, diced avocado, sliced bell peppers, and crunchy nuts like almonds or cashews. Mention using organic produce and high-quality oils to ensure optimal taste and nutrition.

I think asking explicitly for the list of ingredients is useful, I should have thought about that. But some of the ingredients are strange, but I never ate a banana salad anyway. (sorry for that, but I always use banana for test.)

My question is how are you making this? Is this AI generated? How are you sure that the long "optimized" version is better?

View on HN

Thank you for trying. Yes, this is AI-generated. We fine-tuned a special LLM with a huge dataset of prompts with a Monte Carlo algorithm to explore the solution space. We have evaluated every single prompt in our datasets before fine-tuning them with LLM metrics.

So, you've created an LLM specifically for generating prompts for other LLMs? A meta-LLM?

That tickles my engineering receptors in a pleasing way (), but my financial side has questions. The most pressing: if this is successful and useful, why wouldn't the LLMs that you're targeting just replicate your creation and transparently put it between users and their own LLM?

In other words, what is your moat here?

This is a great question. Actually, I have two opinions about this: - It is not economically useful for foundational model providers to build this, or at least integrate it with their products, because it will cost them double. They would prefer RLHF and something intrinsic to their models. - Our dataset is proprietary. We have to collect and write a huge number of prompts to fine-tune it well. And our model is not 100% LLM, we use a version of the evolutionary algorithm to control the quality.

I would say we have a tiny moat of data and algorithm. But at the end of the day, it is about delivering something people find useful. If we could be replaced, let it be. But it works for now.

> The most pressing: if this is successful and useful, why wouldn't the LLMs that you're targeting just replicate your creation and transparently put it between users and their own LLM?

> In other words, what is your moat here?

He is using other people platforms to create a business, layers on top of LLMs won't have a moat, not even training the LLM is guaranteeing a moat. Why on earth everybody asks this on Show HN? The world doesn't need more big techs with moats, we need a healthy ecosystem where multiple players can work together without killing each other and become the only option. No moat, please.

If you can make a better banana salad than the computer suggested, then and only then so you have the right to criticize the computer's output.

Reply on news.ycombinator.com