Ask HN: Are you reconsidering building on top of OpenAI Products/APIs?

2 points by NameError 2 years ago | 2 comments

My view of OpenAI has previously been that they're basically a successful startup, building and selling products - because despite the odd nonprofit ownership structure, they seemed to be -acting- more like a startup than anything else.

I guess I was a bit confused and off-put by the shift from "open" to a more corporate looking direction (MSFT partnership, paid plans etc). But I basically just accepted that they're a product company now, and they do make best-in-class AI products at the moment.

Given yesterday's news, it looks like a lot of insiders were unhappy about that direction and things are potentially going to change. If you're building on top of OpenAI, are you worried? Worried that the APIs will go away, or that time invested into systems like plugins/GPTs will be wasted after OpenAI pulls support for their more product-y things?

campaign_tools 2 years ago |

I think sound advice is not to rely too much on any one company now. I founded an AI-based startup that is harnessing LLMs with RAG to do cool stuff. My guiding philosophy when designing it was to be agnostic as much as possible. And so far that's worked out.

I can send my prompts to multiple different LLMs. My embeddings are mine, for the most part. I have the raw content, so I can always re-embed them if I decide I no longer want to use Cohere or OpenAI. There's also self-hosted vector embeddings using things like Huggingface models.

And in terms of vector databases for RAG? Those are mostly open source, so I'm not worried there. The truth is that LLMs are in an arms-race right now, so staying flexible is just a good business model.

If anything the firing of Sam just confirmed that I have made the right decision by maintaining flexibility. Relying too much on any one of these companies early on could be dangerous.

mvkel 2 years ago |

Not reconsidering at all. Building right now is simply riding the wave.

The alternative is sitting on the sidelines and waiting until things "settle," but will miss out on years of experiencing the evolution of these apis.

It's skill building as much as anything else.

Nobody can monetize wrapper apps at any meaningful scale anyway because rate limits and costs per prompt are too high.