Tim Tregubov

Interesting work from our friends at AE Studio using Goodfire's technology - suppressing deception circuits increases claims of consciousness in AI models. This suggests that claiming consciousness in AI models is mechanistically gated by deception features found with the interpreter models we've trained. Check out their work here: https://lnkd.in/eQHPvhJf

75

This is real-time.. The debate about Nvidia's latest neural rendering tech is ongoing. One side sees the potential of the technology to bridge the gap across the uncanny valley, to give us a much less rigid creative process than a to b to c.. the other side shouts "slop" and tries to shame any creative who dares to explore this new technology, any chance they get. Meanwhile, I'm over here saying "I told you so".. Gen Ai is here to stay, its becoming real-time and Nvidia now has the "RTX on" for AI which I've been talking about for years. We need creatives to help shape meaningful tools that combine this tech with much more granular creative control.. with tools where you as an artist see your creative thumbprint from beginning to end in the parts of the work that matters the most to you. If you're shaming artists or storytellers who embrace these tools, you're doing them a disservice in my opinion. If they don't learn to utilize these tools and workflows, to make them their own and to shape them, others will. There's no way this tech won't be a huge part of the creative industry and of the way we tell stories. It's happening around us as we speak. That said, at the core of it, do what makes you happy and create in any way you like. Don't let people tell you that "you can't do this" or "you have to do it like this". I made the video here by combining Krea.ai's realtime ai editing tool with a feed of me sculpting and building in Dreams, a creative sandbox/game engine on PS5 and 4 that lets you sculpt and build pretty much anything you can imagine in a very intuitive and "hands on" way.

107

14 Comments

Conference proceedings are a sneaky-hard quality control problem. Take AIPP - 100,000+ articles, hundreds of conferences, compressed timelines, distributed review (papers coming in from every direction). Still, same expectations: don’t let obvious quality and integrity issues make it into the published record. Stout task! Super jazzed about the work with AIP Conference Proceedings and Alchemist Review: helping teams spot citation/reference issues, manuscript quality problems, and integrity signals earlier, so human attention goes where it can actually help. Excited to see this moving 🚀

12

“Anthropic leaked 500K lines of Claude Code source. Competitors got the blueprint. And Anthropic will still win. Here's why. # The Claude Code Leak: Why Anthropic Wins Anyway On March 31, 2026, Anthropic accidentally shipped 500,000 lines of Claude Code's source code to npm. Competitors got a free engineering education. Hackers got a roadmap for exploits. Congress started asking questions. And yet-Anthropic is going to be fine. Better than fine, actually. Here's why the company that just leaked its crown jewels will still dominate AI coding tools.”

1

1 Comment