Ever since my career began as a AI Engineer, I have noticed a trend in the AI start-ups. All they say they are building the cutting edge thing for XYZ purposes, but all it is doing is calling OpenAI key in the background, I do not like this at all. This does not mean I have some problem with API, the whole point of tools is that people use it, but half of the time that thing hallucinates, sometimes do not even work.
And the PMs of these startups are a different breed of human beings, like literal human POS. They know nothing about how AI is working under the hood, they say they want only results but they have to understand, AI is not AGI yet . They want unrealistic results just from OpenAI key, I think this is not even possible.
Last month I got a task to process some files, to calibrate prompts so that AI gives consistent result as a normal engineer I reduced the temperature of the AI and added seed values (they have been working for a year they do not know about temperature of AI) to improve the determinism of the AI. And sometimes their RAG pipeline was failing to fetch the PDF content. I looked at the backend code which was calling all that they are not following standard practice of chunking and extracting data they are just giving pdf to the AI and then expecting it to work somehow. I do not know honestly how have they survived till now.
TLDR ; Do not work for any AI startup honestly until they are really building something very unique and productive not calling other API keys and expecting it to work with their fingers crossed.
Top comments (0)