Most of my attention has shifted over to following emerging research on large language models. Right now my key focus there is relating to alignment strategies. I’ve had a strong suspicion since GPT-4 released that the way in which the most recent models are being fine tuned throws away a lot of valuable skills outside what we measure for, and that instead an alignment strategy more similar to the interplay between intrusive thoughts and the prefrontal cortex would achieve more consistent alignment results without sacrificing capabilities. There’s been a few papers over the last year (and even just the last few weeks) that are starting to support similar findings.
In terms of history stuff, there’s still a few odd details I might circle back to, but it mostly feels like I hit diminishing returns unless we see significant new discoveries in materials (which I actually hope we will as LLMs become capable enough to translate into English the extensive bodies of untranslated but discovered works like the Oxy papyri).
One is to follow up on a line of inquiry I’d find relating to grammatical fingerprinting of Paul’s epistles. There was a 2017 psych study that found vulnerable narcissists have a greater degree of personal reference in their writing, and he’s always struck me a bit of that type (“I’m the least of the apostles” fluctuating with “I’m not less than the greatest of the apostles”). When I analyzed the letters in English, there’s significantly more personal reference in the undisputed Pauline letters than non-Pauline Epistles. But the really interesting part is the disputed letters. Only one falls within the range of the undisputed letters in its frequency of personal reference, and it’s one that most scholars have historically thought was forged (2 Timothy). At some point I’ll come back around to doing a similar analysis on the original Greek.
Another recent thread I may look more into would be the Mediterranean parallels for terms translating as “Great Lady” in the LBA and early Iron Age. There’s some weird nuances to a term like that being applied in the Bible to various women, particularly alleged around women connected to the Egyptian pharoh’s household - but when I cross referenced Egyptian records around the relevant time I only see a similar translated term being applied to a Hittite queen who was co-signing the world’s first extant treaty. So now I’m wondering if either (a) the association with Egypt in the OT was an anachronistic rationalization for a foreign concept that was actually originating from Anatolia (like the bees and potentially the tribe of Dan) or (b) if it really did relate to Egypt but because of one or more queens coming from Anatolia marrying into the Pharoh’s household. If the latter, it might help narrow down specifically which dynasties a few alleged events were supposed to have been occurring.
What’s on your research agenda now? Like, what is an unanswered question (with respect to your own knowledge) that you’re curious about?
Most of my attention has shifted over to following emerging research on large language models. Right now my key focus there is relating to alignment strategies. I’ve had a strong suspicion since GPT-4 released that the way in which the most recent models are being fine tuned throws away a lot of valuable skills outside what we measure for, and that instead an alignment strategy more similar to the interplay between intrusive thoughts and the prefrontal cortex would achieve more consistent alignment results without sacrificing capabilities. There’s been a few papers over the last year (and even just the last few weeks) that are starting to support similar findings.
In terms of history stuff, there’s still a few odd details I might circle back to, but it mostly feels like I hit diminishing returns unless we see significant new discoveries in materials (which I actually hope we will as LLMs become capable enough to translate into English the extensive bodies of untranslated but discovered works like the Oxy papyri).
One is to follow up on a line of inquiry I’d find relating to grammatical fingerprinting of Paul’s epistles. There was a 2017 psych study that found vulnerable narcissists have a greater degree of personal reference in their writing, and he’s always struck me a bit of that type (“I’m the least of the apostles” fluctuating with “I’m not less than the greatest of the apostles”). When I analyzed the letters in English, there’s significantly more personal reference in the undisputed Pauline letters than non-Pauline Epistles. But the really interesting part is the disputed letters. Only one falls within the range of the undisputed letters in its frequency of personal reference, and it’s one that most scholars have historically thought was forged (2 Timothy). At some point I’ll come back around to doing a similar analysis on the original Greek.
Another recent thread I may look more into would be the Mediterranean parallels for terms translating as “Great Lady” in the LBA and early Iron Age. There’s some weird nuances to a term like that being applied in the Bible to various women, particularly alleged around women connected to the Egyptian pharoh’s household - but when I cross referenced Egyptian records around the relevant time I only see a similar translated term being applied to a Hittite queen who was co-signing the world’s first extant treaty. So now I’m wondering if either (a) the association with Egypt in the OT was an anachronistic rationalization for a foreign concept that was actually originating from Anatolia (like the bees and potentially the tribe of Dan) or (b) if it really did relate to Egypt but because of one or more queens coming from Anatolia marrying into the Pharoh’s household. If the latter, it might help narrow down specifically which dynasties a few alleged events were supposed to have been occurring.