Sometimes GenAI can’t come up with a summary because the notes were minimal. I figure those cost money. The red “Please provide message” is when I am asking GenAI to summarize notes and there are no notes.
Do the instances when Glide AI fails because of no input cost updates?
The cost seems to be really high (compared to other AI providers) if their was a ‘floor’ for every row where it costs 1 update for 580 tokens when the going published rate is 1 update for 10,000 tokens.
Since a column is by nature an aggregate total of the Generative operations across the set of rows - it seems like it might make sense to apply the published rate across the set of operations.
In my example of 100 rows and 50 actually GenAI success - charging 50 updates which are valued by Glides as 500,000 tokens when I have actually only used 33,000 in the worst case seems…excessive.
So my use case is providing an AI summary of notes across a variety of different class objects. Everything can have notes, some objects have scores of notes, others with none. I am trialing out different prompts to get different results. Then using JSON to attach meta-data to the “note” to get even trickier/smarter.
Having a floor of 1 update per 580 tokens vs the 10,000 advertised seems off.
Yeah I think the wording there can be better, but based on how they charge updates for API query, I suspect this is the case and it seems like it is.
Also I’m not sure I understand your 580 number. I think as long as you make a call, you’re guaranteed to have 1 update already, and it scales up every time you pass a 10k threshold.
The “each run uses 1 update per 10K tokens” note seems stuoendously misleading. You used about 50 tokens per “run” or say 150 tokens for 3 ‘runs’. 150 isn’t close to 30,000 tokens.
The definition of ‘run’ is probably the most misleading - column v row.
