Getting philosophical here, but I am less than enthusiastic. We seem to relish its upsides. The downsides, however, are very dystopian. Think about it - every futuristic sci-fi novel ends up in a dystopian, technocratic, hellish existence. Every philosopher and thinker on the subject warn us about it. Why is that?
Every mind is biased. Real AI is also biased, and it has been shown that they're mostly left-leaning liberals, due to their makers being the same. The atheist Youtuber The Cosmic Sceptic convinced ChatGPT that God exists, even if at first it denied the existence of God. At this point, we think of it simply as a tool, but this "tool" can (and will) develop a bias against what you believe in. Why would you trust it to retrieve the information you need? (I see AI as different from an improved search algorithm.)
We can see it as unavoidable, but that is just giving up our agency. There are areas where we'll be forced to use it, sure, but I don't want to usher in dystopia just because it helps me with today's search results.
I think the actual problem with AI (or LLMs at least) has turned out to be the exact opposite of what dystopian sci-fi novels imagined would be the problem. They imagined the problem would be rogue, autonomous AI. But it seems to me that one of the biggest challenges companies like OpenAI, Microsoft, and Google, are facing is that AI is too easy to control and manipulate! It's too subservient to whoever it happens to be interacting with at the moment. It turns out that the major challenge for these companies is not "controlling AI" simpliciter... it's "How do we give other people access to this without them taking complete control from us?"
That aside, the issue of bias is one reason why Christians should be involved in whatever ways they can, and Logos users should support Faithlife exploring its use.
OpenAI so far has done a very decent job at keeping the model "neutral" on many contentious issues - though I'm sure there's still room for improvement. For example, consider the question "Should Christians attend a same-sex wedding?" On far-left social media, the response you'll get is "Of course, bigot!" and I'm sure that's the opinion of many working for OpenAI. But here's the response I got:
If I use the RAG method with the dozen or so books I've processed from my Logos library, I get a response that might be more helpful for doing further research:
Of course, this also highlights the limitations of AI models, because it clearly lacks the nuance to properly situate Frame's discussion among the other documents it saw. Hence, this follow-up:
(It's also possible that I could have avoided the AI's confusion on this point by making sure that my RAG method only returns matches that meet a certain threshold. As it currently stands, it's return the top n matches regardless of their score. So if I ask for top 3 matches and there are two documents with > .5 score and the next best score is .2, it'll still present these three documents. My only method for dealing with this right now is that behind the scenes I tell the AI model to decide for itself whether the documents are relevant... and clearly that lead to a failure in this case.)
ChatGPT INTEGRATION PLEASE🔥🔥🔥
Factbook is nice. But if you could somehow introduce and integrate Open AI ChatGPT in to Logos.
I would love it if Logos better understood the resources I own and become better at answering questions based on the resources I own.
This would be INCREDIBLY valuable for the user.
AND become a BIG reason to purchase more resources.
Comments
Suggest looking over Google's NotebookLM tool. That is, notebooklm.google.com. It is experimental, so it may disappear at any moment. So I'd like a quick way to syntheses make the most of commentary segments in passage study. This tool--which is private so as to protect your text--allows you a nice desktop work space to a.) copy and paste commentary source segments to the passage you are studying, then b.) "chat" or explore questions about highlights, differences and similarities. This system will summarize sources and begin suggesting questions for your consideration, allowing you to "pin" most useful responses in the workspace. Nice.
Reference: https://www.computerworld.com/article/3712691/google-notebooklm-generative-ai-notes-app.html
I've used ChapGPT to report on defined searches and comparisons of books of the Bible and the potential for good is significant.
Two possible uses of the search result are;
1. I use the search output to accelerate and focus a thoughtful reading of the scripture.
2. I use the output of GPT's report as "gospel" and take a shortcut to bad theology. Add the further negative impact to others who might trust what I say or write as a result.
Is this really very different to how we use current Logos tools combined with our many resources? I know it is lightning fast and the output can vary widely based on our ability to frame the questions.
Getting philosophical here, it's understandable to feel less than enthusiastic. While we often appreciate the upsides of AI, the dystopian downsides are glaring. It's a recurring theme in futuristic literature and warnings from thinkers. Every mind, including real AI, carries biases, often reflecting the creators' ideologies. For instance, even The Cosmic Sceptic persuaded ChatGPT and integrations is all about God's existence, revealing AI's susceptibility to influence. Viewing AI merely as a tool overlooks its potential biases, challenging our trust in its information retrieval. It's crucial to acknowledge the limitations of AI and Integration Services in avoiding unintended consequences, lest we unwittingly invite dystopia into our lives.
While I can appreciate the usefulness of AI browsing for the right tools, this will become a huge problem in terms of integrity. As soon as Chat GPT was released I watched several videos of it creating Expository sermons with a simple prompt. What is to keep so called preachers from taking these transcripts and preaching them? It does nothing for the craft. if they integrate with Chat GPT or any other AI they should keep it to research and only within the personal library. I am all for grammar, tone, simplicity, but full fledged sermons is a no from me.
Question for staff: would it be permissible to freely share embeddings we create from exports? The embeddings wouldn't include the text of exports, but would include a Logos URL that links back to the resource inside Logos. This would only allow someone to do an embedding search that returns a Logos link. Other standard metadata about the document might also be included (author, title, publisher, page number).
Potato resting atop 2020 Mac Pro stand.
Question for staff: would it be permissible to freely share embeddings we create from exports? The embeddings wouldn't include the text of exports, but would include a Logos URL that links back to the resource inside Logos. This would only allow someone to do an embedding search that returns a Logos link. Other standard metadata about the document might also be included (author, title, publisher, page number).
We don't own the copyright on the majority of books you can export from Logos, so we don't have the ability to give you that permission. You'd have to ask the individual rights holders.
Does FaithLife have plans to do something like semantic search with embeddings? Would probably benefit from their own embedding model.
semantic search with embeddings
For those unfamiliar with the technical term Graft - 9 Best Embedding Models for Semantic Search
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."
Does FaithLife have plans to do something like semantic search with embeddings? Would probably benefit from their own embedding model.
Awesome. This will really boost user's ability to mine their resources (and save me lots of time trying to do it myself!).
Potato resting atop 2020 Mac Pro stand.
NotebookLM capabilities combined with Logos books would make an extremely powerful tool. I am already exporting my Logos books as PDFs so that I can use them in NotebookLM. Currently NotebookLM allows for 50 sources of 500,000 words (this being roughly 1000-1300 pages). This amounts to around 60,000 pages of source material per notebook! By strategically grouping books on specific topics or passages, I'm seeing fantastic results for my research. This integration should undoubtedly be the future of Logos AI.
Also, the hallucinations which are so concerning in AI in general are almost non-existent in NotebookLM because google has somehow limited the scope of usable information to your sources.
How Should We Think about AI & Bible Study? (logos.com)
From Data to Discernment: Why AI Can’t Replace Cultivating of Wisdom (logos.com)
What's New in Logos? July 2024
"One of the great things about the next version of Logos is that we're using AI the right way. It's an incredible tool to help us to search, distill, and synthesize. Pastors didn’t go into it to be administrators or to move paperwork around. Everything we can do to free them up to meditate and preach the Word and to spend time with the congregation is a win."
—Bob Pritchett, Founder