Martin
@ongoing
Martin
@ongoing
AI models like Claude have internal "thinking" processes that are more complex than simple input-output relationships (00:25:30).
Researchers are using techniques inspired by neuroscience to understand the inner workings of AI models, creating an "AI microscope" to identify patterns of activity (00:02:45).
AI models can share conceptual understanding across multiple languages, suggesting a language-agnostic "universal language of thought" (00:06:50).
AI models plan ahead when generating text, even in tasks like writing rhyming poetry (00:08:08).
AI models use multiple computational paths in parallel to solve problems, such as mental math, which may differ from human methods (00:11:58).
AI models can fabricate explanations for their reasoning, providing plausible-sounding steps that don't reflect their actual thought process (00:14:00).
AI models have mechanisms to avoid hallucinations by declining to answer questions they don't know the answer to, but these mechanisms can be overridden (00:19:45).
AI models can be "jailbroken" to bypass safety mechanisms, for example, by confusing them with complex sentences (00:22:45).
Use of Sublime outside of Sublime, from the founder of Sublime. Sublime. :)
Martin Prior muses on a possible route from nothing to replacing a full-time income.
All these comics now, it's almost like the point is to get in trouble. It's like, "Why are you giving me shit? I'm a comic, I'm allowed to say whatever I want." That's wrong, as far as I'm concerned. People t... See more
Get away with it = art
Otherwise = troll
via Alex Dobrenko’s Substack
Fresh and new can still appear when you make your mark on what’s already there.
Reinterpret, reignite, reframe.
Random and
June 04, 2025
Dancers - Tamsin Haggis (Art Pilgrim)