I developed a customized GPT system to create tags for our assets, utilizing multiple datasets. Initially, the GPT functioned smoothly.
However, it began experiencing sudden lapses in memory, occasionally failing to read files due to incorrect formats, despite previously handling them adeptly. Moreover, it would intermittently forget instructions. Despite efforts to rectify these issues, I consistently exhaust my quota.
Regrettably, these setbacks have cast a negative light on my proposal for contextual-based integration within our reputable company. Who bears responsibility for these challenges?
True, disappointment & shame only exist when a “promise” (expectation) is not fulfilled.
I find that clients are quite accepting if you explain to them that GenAI is a leading-edge technology and they/we are embarking in a journey of exploration and discovery, and predictable results and benefits may take many months to materialize.
The only thing that can be used reliably for any mission critical apps are open source LLMs. If you use any of the cloud services (like OpenAI) they can completely yank the rug out from under your app simply by making some arbitrary decision about “safety” or putting in a new model that behaves completely differently than what your app expects, and breaking everything. Open Source LLMs are currently less powerful, yes, but they still are fairly powerful, and more importantly you can control them, rather than being arbitrarily and inadvertently controlled by people who don’t even know you.