AI pre-summarizing books
This may be the proverbial stupid question.
As I understand it, some AI reports depend on the books in your library. But a summary in a specific book depends on you having that book and is independent of the rest of your library.
If that is so, if 100 users are asking for a summary on the same section of the same book they all own - won't they all get the same summary?
If that is so, couldn't Logos pre-run such book summaries and have them available for those 100 users to grab immediately and without each of them burning up AI tokens (or whatever the terminology is) to each get the same summary?
Comments
-
You are on a roll today Ken. I have wondered that myself. It could be useful, but it may come at a significant cost to Logos as they have a very large number of books available in their library.
👁️ 👁️
0 -
Ken F Hill said:
If that is so, couldn't Logos pre-run such book summaries and have them available for those 100 users to grab immediately and without each of them burning up AI tokens (or whatever the terminology is) to each get the same summary?
Have you factored in the cost of programming to know whether to pick up a saved summary or create one? The server overhead to retrieve, save, etc. the saved summaries. As branching in code increases potential error points, any added cost in testing and documentation ... etc. This very limited list illustrates the questions that must be answered to see if the suggestion is cost effective. I have no idea whether or not it is, I'm simply pointing out a necessary consideration.
Orthodox Bishop Alfeyev: "To be a theologian means to have experience of a personal encounter with God through prayer and worship."; Orthodox proverb: "We know where the Church is, we do not know where it is not."
0 -
MJ. Smith said:
Have you factored in the cost of programming to know whether to pick up a saved summary or create one? The server overhead to retrieve, save, etc. the saved summaries. As branching in code increases potential error points, any added cost in testing and documentation ... etc. This very limited list illustrates the questions that must be answered to see if the suggestion is cost effective. I have no idea whether or not it is, I'm simply pointing out a necessary consideration.
I wondered some of the same things. But I am too ignorant of all the ramifications to have any idea about how to factor them in. I also wondered if there might be some publisher-contract concerns.
I just suspect it might be more convenient for me and hope it is feasible. If there is much interest, I suspect the facts will come in or at least a yes or no answer.
This is not one of my high priority issues so if the developers and managers don't want to hassle with it that's OK by me.
0 -
Ken F Hill said:
If that is so, if 100 users are asking for a summary on the same section of the same book they all own - won't they all get the same summary?
AI is continually learning, so it will give different answers as time goes on [:)]
0 -
Ken F Hill said:
If that is so, couldn't Logos pre-run such book summaries and have them available for those 100 users to grab immediately and without each of them burning up AI tokens (or whatever the terminology is) to each get the same summary?
It's a good question, and we have considered doing this.
We could make the first user "pay" (in AI credits) for the summary then let subsequent users get it for "free" (without spending credits). However, what we'd most likely do is track the overall reduction in AI costs and let all users share in that benefit. For example, if caching summaries caused our AI costs to decrease by 5%, we could give all users 5% more AI credits (or make each summary cost 5% less). Or if each summary ends up getting reused 100 times (this seems unlikely to me, but maybe!) we could dramatically decrease the cost (in credits) of each AI summary.
0 -
Bradley Grainger (Logos) said:
what we'd most likely do is track the overall reduction in AI costs and let all users share in that benefit. For example, if caching summaries caused our AI costs to decrease by 5%, we could give all users 5% more AI credits (or make each summary cost 5% less‘
Bring it on fella
I expect book popularity would play a part. Popular books would by summarised the most whilst not so popular books would be summarised rarely. So Logos summarising all books would not be economically attractive for the company.
👁️ 👁️
0 -
Bradley Grainger (Logos) said:
We could make the first user "pay" (in AI credits) for the summary then let subsequent users get it for "free" (without spending credits). However, what we'd most likely do is track the overall reduction in AI costs and let all users share in that benefit. For example, if caching summaries caused our AI costs to decrease by 5%, we could give all users 5% more AI credits (or make each summary cost 5% less). Or if each summary ends up getting reused 100 times (this seems unlikely to me, but maybe!) we could dramatically decrease the cost (in credits) of each AI summary.
Thanks for your response. Would any of these options improve speed of pre-cached (or whatever the terminology is) summaries as well as affect costs?
0 -
Ken F Hill said:
Would any of these options improve speed of pre-cached (or whatever the terminology is) summaries as well as affect costs?
Yes; caching the response should speed up the summarization (for the second and subsequent user to summarize the exact same section or search result).
0 -
I'd love to know the approximate number of credits, searches, etc we get currently, at each level. Also, we are told, if we are continually running out (very unlikely as I understand it) we'd be given the opportunity to upgrade. Lovely. Many users are already balking at the appearance of greed coming from Logos. BUT, let's suppose we do upgrade to Max, and are still running out. Then what? Again, I'm sure this scenario is highly unlikely; but, is a concern and should have the plan in place. Esp. as people begin to use this and rely on it more and more.
One thing I dont like, is pricing based on uncertainties. The old cell phone plans used to do something similar to that. Go over your minutes, and be charged unrealistic overages. a $100 bill becomes $500 (for example, not necessarily reality)/0 -
Titles impress people, obedience impresses God.
0
