Five days ago I wrote that the GLaDOS 3.0 corpus had become the project. At the time, one independent review was still running and the third had not finished. The numbers in that post were an in-flight snapshot. All …
Continue readingNearly sixty percent of complete three-way judgments split, but the more important result is how strongly the direction and vocabulary of disagreement depend on the judge.
Continue readingA moderation category fired on GLaDOS’s figurative register rather than on anything in the conversation, and most of a human review queue turned out to be one verbal habit.
Continue reading
Last month I wrote an update on the work going into SkyeEngine. Since then, I’ve spent an absolutely unreasonable amount of time working on the GLaDOS 3.0 corpus. Originally, the corpus was more or less a large folder of …
Continue reading
Ask GLaDOS-14B enough questions and she will eventually tell you that something is quaint. Actually, “eventually” is doing heroic work in that sentence. The archived 14B evaluation dump contains 84,826 generations from …
Continue reading
Before GLaDOS-14B, I trained three versioned builds of an 8B model based on a DeepSeek distill. Those smaller models established nearly everything the larger one would later amplify, both good and bad. The NAS archive …
Continue readingNo posts on this page match that filter.