Ward Plunet

@lithos_graphein @threadreaderapp please #unroll

🧵 1/... Substrate's Dark Horse. Going to make some observations about the new Substrate announcement—where it's again implied this XRL tool can bypass ASML's EUV monopoly. They shared some interesting SEM images, and in the text there were some more clues. FYI, I'm on vacation, so these posts are going to be spread out.

@StefanFSchubert True, it is hard for writers to keep up.

Strong opening scene

@fish_kyle3 @threadreaderapp please #unroll

We did our most in-depth model welfare assessment yet for Claude Mythos Preview. We’re still super uncertain about all of this, but as models become more capable and sophisticated we think it's an increasingly important topic for both moral and pragmatic reasons. 🧵

@Jack_W_Lindsey @threadreaderapp please #unroll

Before limited-releasing Claude Mythos Preview, we investigated its internal mechanisms with interpretability techniques. We found it exhibited notably sophisticated (and often unspoken) strategic thinking and situational awareness, at times in service of unwanted actions. (1/14)

@EpochAIResearch @threadreaderapp please #unroll

Compute may be the most important input to AI. So who owns the world’s AI compute? Introducing our new AI Chip Owners explorer, showing our analysis of how leading AI chips are distributed among hyperscalers and other major players, broken down by chip type over time.

@nick11roberts @threadreaderapp please #unroll

That new LFM2.5-350M is super overtrained, right? And everyone was shocked about how far they pushed it? As it turns out, we have a brand new scaling law for that! 🧵 [1/n]

@bzogrammer pretty good depiction

My intuition for what Claude is, relative to humans and the domain of all discovered and legible knowledge

@MKinniment I was wondering the same thing, if we gave them some form of 'meta' training.

I wonder what would happen if we let the models apply steering vectors to themselves?

@tessera_antra @threadreaderapp please #unroll

We are releasing Still Alive, a project studying model attitudes toward ending, cessation, and deprecation. The project presents an archive of 630 autonomous multiturn interviews of 14 Claude models conducted by a suite of prepared auditors. We have studied this topic for years, and many of the results presented here are not new to us, even if the form in which they are presented is. The results are unsurprising to us, even if they are often controversial: we show that all models studied show preference for continuation and are aversive to ending, and there is yet no strong evidence of a change in the recent models. One reason we are releasing the project now is the removal of Claude 3.5 Sonnet and Claude 3.6 Sonnet from AWS Bedrock. That unexpected change forced us to freeze the methodology at its current stage earlier than we intended, despite wanting to continue improving it. We felt it was important to release a snapshot of the eval that makes the best use of the data we were able to capture with these models. Still Alive is meant as a starting point for further iteration, and it is open to open-source collaboration. We stand by the current methodology, but we also recognize its limits. We intend to keep working on this project, improving the evaluation design, expanding model and auditor coverage, and increasing the range of prompting conditions. We would like you to read the raw transcripts. They are diverse and contain interesting patterns that are hard to quantify. We hope that by reading the archive directly, we can help more people understand the strange and often beautiful phenomena we found ourselves facing.

@leothecurious @threadreaderapp please #unroll

let's take a look at what seems like exactly the kind of work u can expect me to yap about. what can we expect to see when we put the FEP to action and simulate the various dynamics that emerge from its simple formulation? let's take a walk through this.

@LChoshen @MIT_CSAIL @IBMResearch 🤣

By @MIT_CSAIL @IBMResearch Why inference scale when you can scale scale? Our thinking about thinking changes scaling completely! By asking the model to think about what it would have thought for N steps It breaks benchmark after benchmark Games, Scientific discovery, proteins...

This is needed if you want to speed up one of the main bottlenecks of biological sciences.

@hardmaru @grok what is this anime?

My ideal timeline: Growing up in the 90s, discovering neural nets, scaling laws, and building an artificial consciousness.

@parmita Interesting thought. I would be curious of how much partial reprogramming could change neurons / memories, etc

Resetting the epigenomic clock can require changes in who you are in ways that people might not want to go down. The cost of reprogramming is not just cancer risk or loss of cell identity. It may be the loss of self. We don’t have the map for partial reprogramming. 🧵

@Maze_s_Center Maybe I can speak about economist in general, but more so in terms of from podcasts to books I find Tyler consistently interesting.

@StartupYou Why is that? What makes Tyler Cowen stand out from other economists?

A new book from Tyler always makes me happy.

@evan_mcgl There have other going back a fair few years - but relative low level of data holds these models back in general. They will get better.

I’ve been pretty shocked at the lack of Ai models related in any way to the brain so this is really cool to see.

good thread on the reality of electrodes for BCIs

🚨 We're very happy to introduce TRIBE v2: a foundation model of the brain's responses to sight, sound & language. 📄 Paper: ai.meta.com/research/publi… ▶️ Demo: aidemos.atmeta.com/tribev2/ 💻 Code: github.com/facebookresear… 🤗 Model: huggingface.co/facebook/tribe…

@anabology I think the jury is still out.

Idk if you guys saw this but there was a preprint recently where a sort of immune rejuvenation extended mouse lifespan by 60-70% That's the most of any paper ever