Aborigen

The two primary concerns with AI use are energy consumption and violation of copyright. The first issue is maddening: we have an entire grid of sustainable energy that could power the nation, just sitting there because of politics. Popular thought holds that one OpenAI query takes ten times the energy of a Google search. But if someone wants to bring up this point with me, I’d like to learn how they manage to get along without Netflix, Hulu, Disney, Max, Spotify, or any other streaming service. Doing one without the other is like condemning Amazon while continuing to shop at Walmart and dine at Chick-fil-A.

Unfortunately, Google has intentionally downgraded itself to produce worse results, to compel users to interact with it longer and be exposed to more ads, which may push users to consider ChatGPT (now able to research online and provide accurate citations). My first response would instead be to explore other search engines, many of which protect your privacy and data: Kagi, Mojeek, Startpage, monocles, Ecosia, and even DuckDuckGo and Brave, if you can suspend your convictions long enough to use them.

Putting all that aside, there’s the copyright issue: large language models (LLMs) trained on stolen work that both OpenAI and Meta/Facebook are working frantically to justify stealing. But what if you could build your own model with legally acquired literature?

NotebookLM is a limited-language model where you upload your own material, and it responds only pertaining to that material. It’s a fun tool to play with: relatively quickly, you can get it to summarize all provided material, have it break out a mind-map of topics for specific research, and … even generate a podcast episode.

Well, it sounds like one, anyway. It generates an audio file, 10 to 20 minutes long, where a man and a woman banter as they discuss the material you provide. They call their show the Deep Dive, they refer to you (the listener) when you custom-request the topic to be discussed, and there’s even a feature to “speak to” them, like ask them a question in the middle of their show, but I don’t have a microphone and haven’t played around with this. It’s probably clunky but points toward an amazing and alarming future of seamless computer/human interaction. For the most part it sounds pretty slick: they pause for breath, their tone drops as they conclude a thought, they stumble over words and make little jokes.

This clip is on the nature of giantesses—where they come from, their roles within Scandinavian myth, what they looked like and where they lived, etc. Obviously the material I provided is tougher for them to chew on, as they struggle to pronounce Old Icelandic and Old Norse names like Skaði and Þorgerðr. I’m unaware of any way to insert instructions on how to manage these names or Americanize their pronunciation.

If you can bear to listen to the file, you’ll indeed receive a deep dive into Old Norse myths and suppositions at what came before, how the giantesses were formed and conceived before Óðinn and Þorr rolled into town. The information is solid: These are publicly available academic articles from Lotte Motz, Gunnhild Røthe, Michael A. Linton, John McKinnell, Aðalheiður Guðmundsdóttir, and many others, people reading between the lines of the Icelandic texts and piercing the haze of Christianization to discern what heathen giantess faith looked like.

It’s not intended to replace actually reading these books and documents and making my own notes. That’s the process I’m undertaking currently, setting aside three hours a night to fill out a notebook with the points I think are important from a stack of academic journals, to embed them in my memory and inform everything else I’m reading. (Now, honestly, I’m learning as much about the professional rivalry between these researchers as I am about medieval Old Norse society and earlier.) But a service like NotebookLM is handy for summarizing a large mass of information, breaking it down for further research, and I enjoy listening to this ersatz podcast while washing dishes and cooking.

This longer episode has to do with the Futhark runes and how they were used for magic. It cites ancient texts and contemporary analysis of these to sum it all up in another glib conversation. The “hosts” move methodically from topic to topic, calling back to earlier points made, laying dense educational material out in an easily digestible format. You can’t find this anywhere else, but I can make it for myself and consider myself edified until I find an evening to attack the source material myself.

I dunno. I’m excited about the technological tools at my disposal. I’m as thrilled to guide a 15-minute episode on how giantesses had sex with gods and heroes, as I am to plunder the university library system and discover all these people who made careers of researching material I never could. I mean, I could, but it would require a lifetime of mastering Old Norse and deciphering skaldic poetry and kennings. I would rather benefit from their labor and begin forming my own connections and systems, and the limited-language model tool enhances this process.