One thing I've noticed working with @poolsideai's Laguna over the weekend is that it seems much less eager to immediately appease the user. Specifically, if a job is running and I insert some tangentially related context, it has a tendency to stay on the original task and respond to the new context whenever it's at a good stopping point. Much less easily derailed than other models.

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models.

Introducing BenchBench One of the most important questions in AI and society is how well different models can sit on shared elongated stools. Until now, we haven’t had a rigorous way to evaluate this.

Meet @ZaraJavar, Senior Cryptography Engineer at the Interfold. From pure math and formal security verification to threshold FHE and ZKPs, Zara is building the cryptographic core of E3s so private inputs can produce publicly verifiable outcomes.

The "CR" in CROPS stands for "Censorship Resistance". @ethereum's censorship resistance guarantees focus on the typical framing of censorship, where one has something one wants to publish, and another has power to allow/disallow it to be published. If there is a sufficient threshold of honest and reachable nodes and validators, Ethereum users have strong guarantees that if they are willing to pay the current market price for gas their transaction will be included in a block in a reasonable amount of time. But there is another important variant of censorship resistance that is beyond the scope of Ethereum's base guarantees: Forced Abstention. Forced Abstention attacks are those where a malicious party bribes, colludes with, or coerces another party to prevent them from publishing their input to a system. The number of honest Ethereum nodes and validators has no impact on this kind of censorship. The "P" in CROPS stands for "Privacy". The mitigation for forced abstention is a combination of privacy and receipt-freeness (the inability for one to prove they did or did not author a given input) that allows one to plausibly deny having participated, and thus resist a forced abstention attack. This latter form of censorship resistance is a large part of why we built @theInterfold. Similar to how an application building on top of Ethereum can inherit its censorship resistance guarantees, applications built on The Interfold can inherit resistance to forced abstention attacks. This is a critical quality for things like secure voting systems, efficient auctions, and other mechanisms involving adversarial games with private inputs. The O (open source) and S (secure) in CROPS round out the picture, and are likewise core values for The Interfold.