Talkie Is a ‘Vintage LLM’ Trained on Pre-1930 Data to Help Facilitate ‘Time Travel’
Talkie is a vintage language model trained on data published before 1930, allowing it to navigate copyright issues. Its creators aim to explore historical communication and predict future events based on a solid grounding in history. The project faces challenges such as reliable training data and the risk of contamination from post-1930 material.
- ▪Talkie, also known as 13B 1930 LM, is trained on material published before 1930.
- ▪The choice of 1930 as a cutoff date allows the model to utilize public domain material.
- ▪One of Talkie's initial uses was to rate the surprisingness of events occurring after 1930.
Opening excerpt (first ~120 words) tap to expand
If you’ve ever heard the term “vintage LLM”, you might have found yourself wondering if the AI-pocalypse has really been going on for long enough that early chatbots are worthy of nostalgia. Happily, though, that’s not what the term means; instead, it applies to an LLM that seeks to emulate the perspective of a certain point in the past.cnx.cmd.push(function(){cnx({"playerId":"92b7b46b-43ed-4e0e-b21b-2c999302d9d7","settings":{"advertising":{"macros":{"AD_UNIT":"/23178111854/od.gizmodo.com/article","CHILD_UNIT":"article","POST_ID":"2000751758","POST_TYPE":"post","CHANNEL":"tech","SECTION":"artificial-intelligence","SUBSECTION":"","CATEGORIES":"artificial-intelligence","TAGS":"talkie,vintage-llms","NOP":"0"},"timeBeforeFirstAd":0}}}).render("cnx-player-main")}); The idea is that you…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Gizmodo.