Last night I was given access to Bing Search’s GPT-4
A couple months ago, when Chat-GPT3 first came out I posted about how it might change the Church landscape, and presented a sacrament meeting talk that GPT-3 had written. Several months before that I speculated about how natural language processing would lead to researchers in essence having their own on-hand research assistant.
With Chat-GPT4 it looks like we’re kind-of-sort-of there, if your research assistant is a sloppy undergraduate.
I asked it to “Create a table of primary sources during Joseph Smith-era Nauvoo that were authored by women. Show 1) the author, 2) what the primary source is, 3) example text from the primary source, and 4) something interesting about the source.”
This is what it gave me (it’s hard to get the footnotes to play nice with the T&S website, so those are omitted and you’ll have to trust what I have to say about them):
Here is a table of some primary sources during Joseph Smith-era Nauvoo that were authored by women:
|Author||Source||Example Text||Interesting Fact|
|Emma Smith||Minutes of the Nauvoo Relief Society12||“We are going to do something extraordinary. . . . We expect extraordinary occasions and pressing calls.”2||She was the first president of the Relief Society and Joseph Smith’s wife.|
|Eliza R. Snow||Poems, Religious, Historical, and Political12||“The voice of innocence is hush’d / The sobs of sorrow dim; / And martyrd virtue sleeps beneath / The ruthless hand of sin.”2 (On the death of Joseph and Hyrum Smith)||She was a poet, a leader in the Relief Society, and a plural wife of Joseph Smith.|
|Lucy Mack Smith||History of Joseph Smith by His Mother13||“I have never written one sentence which I could not substantiate by undoubted testimony.”3 (Preface)||She was Joseph Smith’s mother and wrote a biography of his life based on her memories and documents.|
Do you want more examples or more details about these sources?
Yes, show me more examples
Here are some more examples of primary sources during Joseph Smith-era Nauvoo that were authored by women:
|Sarah M. Cleveland||Minutes of the Nauvoo Relief Society123||“I rejoice in this privilege. I have no doubt but that this Society is of divine origin.”2 (March 17, 1842)||She was one of the counselors to Emma Smith in the Relief Society presidency and a plural wife of Joseph Smith.|
|Elizabeth Ann Whitney||History of Joseph Smith by His Mother13||“I was present at a meeting where he [Joseph Smith] spoke in an unknown tongue; and David W. Patten interpreted, and told what he had said afterwards; it was according to our faith.”3 (Chapter 44)||She was one of the counselors to Emma Smith in the Relief Society presidency and a plural wife of Joseph Smith.|
|Mary Elizabeth Rollins Lightner||Autobiography1||“When I heard that God had revealed the law of Celestial marriage that we would have the privilege of associating in family relationships in the worlds to come, I searched the scriptures and by humble prayer to my Heavenly Father I obtained a testimony for myself that God had required that order to be established in his Church.” (Page 7)||She was a pioneer woman who saved some pages of the Book of Commandments from being destroyed and a plural wife of Joseph Smith.|
That’s actually better than I would have expected. Keep in mind all these language models have no idea what a “primary source” is, they just know that when people use those words certain other words go with them. Success very much depends on having lots of text in their training data that talks about what you’re asking. Almost all of the progress we’re seeing in this field in recent years is the result of having much more training data available (i.e. much more text) and much bigger compute clusters with lots more GPUs. There haven’t been any fundamental breakthroughs.
So don’t expect a language model to be able to tell you anything new, or even anything obscure, any time soon. But there’s a middle ground of knowledge that may be well-known to people who specialize in a field but unknown outside it that a good chatbot could make available to everyone. The question asked here seems like a pretty good example.
So AI doesn’t really know what it’s doing it’s just has a sophisticated language model that associates words and various bits of content. That’s obviously a bad explanation, but it sparked the thought.
At some point in the future, if most content is AI generated, and AI content depends on an underlying source of content, all AI content will start to become self referential and real knowledge will diminish. You can imagine some dystopian scenarios resulting from that, even if they don’t have the whizbang appeal of terminator robots.
I guess you might say, trusting an AI to get it right is no different than trusting a researcher to get it right, at some level. But someday, an AI will read this post you made, with content generated by an AI, which you already stated has errors, and the AI will view this content as factual, because it’s put together with the correct phrasing.
Just as people have figured out how to undermine google’s algorithms to get to the top of the page to sell their useless websites/services/viruses, it seems a matter of time before the AI algorithms are exploited to shape future “knowledge” on an issue. Indeed, the fact that the AIs all seem to be trained to be more progressive rather than dispassionate suggests it’s already happening.