Official: You Can Now Get Early Access to the Next Version of Logos

I know the other subscription tiers haven't been announced/determined, but I hope there is one aimed at people with academic interests. I would have little use for many of the new Pro features, but I might be interested in an AI tool optimized to do textual criticism or Greco Roman or Akkadian intertextuality and stuff like that. 

There will be a tier focused more on academic use. I confess that Akkadian intertextuality isn't yet at the top of our list of possible features for that tier, though...

How will the subscription model work with the Academic Program and Academic Pricing? More than just tiers targeted at academic interests, which could include non-students/non-faculty, will there be pricing for academic users in the future?

Major improvements will certainly be delivered to subscribers only.

With the certainty that all new major improvements/features will be accessible to "subscribers only," it seems their decision has unfortunately been made.

My apologies. I expressed myself poorly there. What I was trying to say was that if we released a new feature, any improvements to that feature would be delivered to subscribers but would not be delivered to people who had bought the feature outright. I wasn't ruling out the possibility that in two years' time there might be a means to purchase those improvements outright.

Mark 

There seems to be confusion on the point as to whether the company may still consider new features for non subscribers.

Please clarify this for us.

And guess what ... the more that the AI tools infect the Logos experience, the LESS VALUE tagging and datasets will have.  The AI doesn't understand the incredibly poweful syntax and keywords that are inherent in Logos searching.  And it would be hugely, impractically expensive for Logos to reprogram ChatGPT to recognize those capabilities.

If I'm wrong in my logic, please explain why, and explain how subscriptions will change any of it.

Mark explained how smart search works earlier. First, they take your search and make it broader (looking for synonyms) and use traditional tagging to find the top 50 results. Then AI orders those results and figures out which portions to highlight as a match with your search. Beyond that, there is no reason that the tagging should not be visible to an AI. It can easily be trained to convert a search for Abraham to look for "Abram," "Abraham," and person:Abraham.

However, at least in current forms, and likely in all future forms, SINCE the tagging is hidden, not part of the exposed readable text, the AI engine is oblivious to it.  The AI searches don't need nor can they use it.  The AI is trained on the world-wide-web ... why should it need tagging? (sarcasm intended ... like in Back to the Future ... where we're going, we don't NEED "roads").  

You are making several assumptions here that we have no evidence of. We do not know what the access of the AI component to the tagging is, The AI component of search is not trained on the world-wide-web. Logos is not using one of the glitzy chat bots for its prioritization of results. I suspect that much of the tagging was done initially by AI or a NLP predecessor. Should Logos expand its AI in the search into the core search engine itself, I would expect it to still continue to be tied to tagging - physical or on the fly.

And it would be hugely, impractically expensive for Logos to reprogram ChatGPT to recognize those capabilities.

It is a matter of retraining rather than reprogramming. Should Logos ever choose to partner with ChatGPT or a similar product, I would expect significant retraining to match the corpus that it is intended to "know."

that as the AI grows, Logos will do significantly less tagging, on significantly fewer books ... since that's probably a big portion of their historical business expenses.

I assume the reverse. As AI becomes more adept at accurate tagging (I'm assuming somewhere 92-96% now) and there is less human cleanup of the data after the initial computerized process, I would expect more tagging and more complex tagging than we currently have.

MJ:   I have participated in the R&D plus testing of another ChatGPT-driven Bible Study tool, so I do have some experience with the process, in addition to having been an active programmer since 1967. I've also been very extensively involved with R&D plus use of serious genetic algorithm (GA) and neural-net (NN) tools for high-end stock trading algorithms.  So, none of this is new turf for me.

I can't speak to what's going on behind the curtain at Logos re AI, but I'm 95% sure that the things I spoke of are correct and will remain so for the foreseeable future.  FYI, they have NOT trained ChatGPT on their huge library ... partly due to the expense, and partly since even those couple hundred thousand books would just be a drop in the bucket statistically, vs the zillions of MB of web-scraped content that makes ChatGPT & other robust AI engines function.  Furthermore, I *am* certain that the tagging is hidden from the current Logos AI implementation, based on simple tests I've already done.  And, even if it wasn't, those tags if made visible to ChatGPT are just "noise" ... meaningless gobbledegook that would not statistically correlate with other items in ChatGPT's huge dictionary of genes and neurons, nor would they contribute to the rules that it generates.

Also, I'm drawing some information from an email that Mark Barnes sent to me on March 11th ... it wasn't labelled as confidential, so I'll paste it here in its entirety (bold & underlines are mine):

Hi Jim,

Very few companies have the financial ability for an entirely homegrown AI engine, so we’re building on technology provided by several partners. We use a variety of large language models and other AI tools and choose each one based on its suitability for specific tasks. We monitor the effectiveness of these models and have the ability to quickly switch to better-performing or more reliable models if we discover frequent bias or other adverse issues. 

No AI tool presently in existence could *only* be trained using Logos’ library. We have more than 200,000 books, but training an AI from scratch would likely need tens or hundreds of millions of books.

In a document I’m writing, I say that AI has human qualities, not divine ones: AI, and especially the large language models that power tools such as ChatGPT, are typically trained on billions of documents written by human beings. That’s how it’s able to mimic human responses. But that also means AI mirrors and sometimes amplifies the weaknesses found in all human authors. Like us, AI has limitations, fallibilities, and biases. Human beings are both fallen and made in the image of God. AI is neither of those things, but it reflects both of them because it is trained on us. Just as human-authored books can both help and hinder our study of God’s word, so too can AI.

We mitigate against those risks by using AI to illuminate the highest-quality biblical texts, datasets, and books. We don’t just use AI standalone, we use it with our own content. That content, and careful direction from us, guides the AI to product output that is very likely to be useful for biblical studies. It’s still imperfect, and needs to be used with discernment, but it’s generally higher quality and more accurate than you might expect.

Logos

It's far easier for me to make an investment now for something that I can continue to use in the future than it is for me to commit to an ongoing subscription fee that I may not be able to maintain through my retirement. 

I'm not sure I'm understanding you. You'll still be able to upgrade to the latest version of Logos after retirement (for free), without paying a subscription fee. 

Then let me try to express the core concern here in a different way. If Logos moves to this new model in October, what will my options be in December? Will I be able to purchase a new feature in December and carry it over into retirement? I would like for the answer to be "yes." The most recent post from Mark Barnes suggests that may be the case. 

Major improvements will certainly be delivered to subscribers only.

With the certainty that all new major improvements/features will be accessible to "subscribers only," it seems their decision has unfortunately been made.

My apologies. I expressed myself poorly there. What I was trying to say was that if we released a new feature, any improvements to that feature would be delivered to subscribers but would not be delivered to people who had bought the feature outright. I wasn't ruling out the possibility that in two years' time there might be a means to purchase those improvements outright.

If the answer is "no," then I would most likely forgo the new feature, because I'm not interested in picking up ongoing monthly fees as I approach retirement.

I believe most of the angst in this thread is because earlier posts from FaithLife staff seemed to imply that the answer would be "no." But I'm entirely comfortable with the arrangement Mark describes here:

...What I was trying to say was that if we released a new feature, any improvements to that feature would be delivered to subscribers but would not be delivered to people who had bought the feature outright. I wasn't ruling out the possibility that in two years' time there might be a means to purchase those improvements outright.

I don't expect to get new features for free.

Mark Barnes (Logos) wrote the following post at Today 2:29 PM:

My apologies. I expressed myself poorly there. What I was trying to say was that if we released a new feature, any improvements to that feature would be delivered to subscribers but would not be delivered to people who had bought the feature outright. I wasn't ruling out the possibility that in two years' time there might be a means to purchase those improvements outright.

....

I apologize for adding to that confusion. It's hard to keep up with this thread, and I didn't proofread my response carefully enough before I hit Post.

Yes, Logos is still considering new features for non-subscribers. That is not ruled out. That has been the situation since my first post where I asked for feedback on whether that would be important to you.

It's not surprising to hear many of you say that the ability to buy a feature set is important. We knew it would be important to some of you. What we wanted to understand was how important it was and how many of you felt that way.

This thread has been great for us in that regard. We hear, loud and clear, that – alongside our plans for subscription – many of you would value the option to purchase a feature set, even if it didn't come with regular updates or AI features. We're continuing to listen, and continue to value this feedback.

Mark -

THANK YOU for qualifying the earlier exclusionary statement, which truly has, as another poster put it, "caused many of us a lot of angst".  Understatement, that.

However, your phrasing still makes it clear that you ARE SURE you are going to start offering subscriptions of varying types, but the wording above "is still considering" means you ARE NOT YET SURE whether you will offer periodic "baskets" of accumulated non-AI features (along with necessary datasets) to be purchased on a "perpetual license" basis, as has been true for all prior history.

Please, if ... WHEN (I hope) ... y'all DO become SURE about offering those accumulated-feature+dataset packages every year or two, PLEASE let us know here in this thread, and please MODIFY your original post so that newcomers to the thread won't have to wade through everything to find it. 

And I do join with others most sincerely in thanking God for leading you to open this up to feedback ... and thanking you for following that leading.

I'm familiar with your search for Noah, I started the other thread about it. I think MJ did a good job of explaining that what you're looking for is not what Logos is trying to do. They're making a smart search, not a chatbot. Since no resource in your library is going to distinguish between the age of Noah from a YEC perspective versus other viewpoints, there isn't anything for it to surface.

In my book, that is a feature, not a bug. I don't want Logos generating an answer by combining different resources, where I have to guess where it came from. That is how other AIs make stuff up. I don't want answers that aren't explicitly in my library. But your mileage may vary. 

If you run a search where the qualifier is relevant, you will get better results. Example: who is the Antichrist according to preterists?

I'm familiar with your search for Noah, I started the other thread about it. I think MJ did a good job of explaining that what you're looking for is not what Logos is trying to do. They're making a smart search, not a chatbot. Since no resource in your library is going to distinguish between the age of Noah from a YEC perspective versus other viewpoints, there isn't anything for it to surface.

In my book, that is a feature, not a bug. I don't want Logos generating an answer by combining different resources, where I have to guess where it came from. That is how other AIs make stuff up. I don't want answers that aren't explicitly in my library. But your mileage may vary. 

If you run a search where the qualifier is relevant, you will get better results. Example: who is the Antichrist according to preterists?

Thanks for that early thread, Justin.  Yours plus Josh's comments convinced me to spend the money and try it out.

I hear you re the value of using qualifiers that are keywords in Logos datasets ... I doubt that the Logos AI implementation actually USES those datasets at present, however (I explained why, earlier).  It would be a very complicated and expensive undertaking.

The FACT is that Logos DID recognize "young-earth creationists" ... it found a lot of info about those teachings, and the debates with old-earth creationists, and evolutionists (I have a lot of titles in my library about this topic ... it's a favorite of mine, me being a science geek.)

So, until and unless Logos staff explicitly state that the keyword lists, tagging and dataset info is fully accessible to the AI Search engine, as someone who's done a lot of coding and testing of this kind of stuff before, I'd advise everyone to assume that info is NOT utilized, at least at present.  I would be *happy* ... overjoyed, in fact ... if someone in charge of the Logos AI coding were to correct me on this.

In fact, IF those keywords, tags and datasets ARE fully intertwined with the Logos AI Search tools, I'd have a much higher opinion of their value.

In fact, IF those keywords, tags and datasets ARE fully intertwined with the Logos AI Search tools, I'd have a much higher opinion of their value.

I'm still waiting for the AI to understand that "Nicea" and "Nicaea" are the same place. 

the Logos AI is BUILT ON various "chat" engines

Smart Search doesn't use ChatGPT or Gemini.

In Smart Search, we perform a very loose search without AI. That finds 50 articles that roughly match your query. We then use AI to identify a snippet from each article that most closely matches your query. We don't rewrite any text; we only identify the most relevant existing text in the book. Once we have these snippets, we use AI to re-order them to put the most relevant ones at the top.

I think we have different readings of "built on" ...

Hi MJ:

I didn't recall seeing that quote you provided from Mark Barnes.  I'm using Mark's term, from the email he sent me, when he said:

"we’re building on technology provided by several partners. We use a variety of large language models and other AI tools and choose each one based on its suitability for specific tasks."

I've been using the term ChatGPT as a simplistic way to talk about AI interface.  Actually, ChatGPT is just a front end.  Logos obviously is utilizing a deeper level API.  He didn't tell me who those partners were. 

I also missed the two-part search interpretation "hint" that you quoted.  If that's how it's working, it's going to be hard to learn how to use.  If the Logos AI first-pass REQUIRES some rough syntax rules, then presumably it rejects some other syntax.  That's less powerful than other interfaces I've seen before.

Let’s imagine that Logos staff finally decide to clean up all the awkwardnesses and documented limitations of the notes+highlighting feature(s). 

Let’s imagine that a lot of users have previously created hundred of thousands of notes that use various features of the existing Notes+Highlighting. Finally, let’s imagine that your cleanup of Notes+Highlighting is so extensive that you need to replace a major portion of the existing code, and that the 2.0 version can do the same things as before, plus a lot more things, as well as fixing bugs. 
You’ve said major new features will only be by subscription. I presume a wholesale rewrite to a 2.0 version (my example) would be in that subscription-only category. 

The Logos NOTES feature, as well as the highlighters, are really bad. Even though I created my own highlighters, it's still bad because I'm limited by the tool as it was built and it has a lot of inconsistencies, almost a kind of "bug".
I still dream that Logos will implement the markdown language and a note system inspired by the Obsidian software. Only then will I start using the Logos grading system. Today I only use it to mark and save errors I find in books. Nothing more.

the Logos AI is BUILT ON various "chat" engines (see Mark Barnes' letter to me, that I posted)

He didn't say that. You quoted him as writing:

so we’re building on technology provided by several partners. We use a variety of large language models and other AI tools and choose each one based on its suitability for specific tasks.

This quote is accurate.

FWIW, Smart Search is currently built with word embedding vector and vector-space techniques, not by using a RLHF LLM. This may change in the future as technologies advance and different approaches deliver better results.

I've been using the term ChatGPT as a simplistic way to talk about AI interface.

This feels unhelpfully simplistic to me (if you want to talk about technical details).

Hey Bradley ... I'm surprised you contradicted my use of Mark's term.  Here, again, is the full letter.  Note the bold underline that I added, re "built on".  

I'm going to try to stop responding to quibbles about just how Logos AI (or any other AI) works ... although I've had the opportunity to do a lot of related development coding, I don't know Logos' code.

I've been trying to share insights but it appears to be creating arguments that don't really impact the opinion I am sharing, which I am of course entitled to.

So, sayonara, folks!

Hi Jim,

Very few companies have the financial ability for an entirely homegrown AI engine, so we’re building on technology provided by several partners. We use a variety of large language models and other AI tools and choose each one based on its suitability for specific tasks. We monitor the effectiveness of these models and have the ability to quickly switch to better-performing or more reliable models if we discover frequent bias or other adverse issues.

No AI tool presently in existence could *only* be trained using Logos’ library. We have more than 200,000 books, but training an AI from scratch would likely need tens or hundreds of millions of books.

In a document I’m writing, I say that AI has human qualities, not divine ones: AI, and especially the large language models that power tools such as ChatGPT, are typically trained on billions of documents written by human beings. That’s how it’s able to mimic human responses. But that also means AI mirrors and sometimes amplifies the weaknesses found in all human authors. Like us, AI has limitations, fallibilities, and biases. Human beings are both fallen and made in the image of God. AI is neither of those things, but it reflects both of them because it is trained on us. Just as human-authored books can both help and hinder our study of God’s word, so too can AI.

We mitigate against those risks by using AI to illuminate the highest-quality biblical texts, datasets, and books. We don’t just use AI standalone, we use it with our own content. That content, and careful direction from us, guides the AI to product output that is very likely to be useful for biblical studies. It’s still imperfect, and needs to be used with discernment, but it’s generally higher quality and more accurate than you might expect.

Logos