Peter Wang 🦋 @pwang

Wow. @Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose. It responds with nuance and judgement, & handles long context VERY well. I've never experienced an open weights model like this before.

theatlantic.com/newsletters/20…

Important to note @Zai_org train on @Huawei Ascend chips, no NVIDA (!) So you have frontier -3 months on a fully Chinese stack, 90% cheaper. Would estimate the total cost of this to be $25m, largely post training (80%) @Zai_org market cap now nearly $100b, $$s in open source!

Zixuan Li @ZixuanLi_

a week ago

Finally, Artificial Analysis Intelligence Index concludes the GLM-5.2 release.

57 65 1K 252K 105

I am very excited about this research: We show 2 things: 1. If you just do random sampling (i.e. you try to solve a problem k times independently, and keep the best) your ELO scaling will be linear in log(test-time-compute). Agents like Claude-Code and Codex scale like that after a few hours. 2. We compare human expert coders to coding agents on the same tasks (from AtCoder Heuristic Contest). The exciting finding is that humans scale super-linearly. This is evidence that humans do continual learning, while they are solving a problem! I.e. they learn more about the coding problem they are trying to solve and scale fundamentally better compared to randomly trying things in a memoryless fashion. This is empirical evidence that supports what many of us have felt for a while: unless we solve continual learning we will not be able to outperform humans in tasks that take many days. Current coding agents are not able to do this.

Qiuyang Mang @MangQiuyang

a week ago

(1/n) New blog from UC Berkeley, UW, and Princeton: Who scales better in long horizon: AI coding agents or top coders? We compared modern agents to top human contestants in an open-ended coding marathon. Agents sprinted early. Then they plateaued. Top humans kept improving. We

12 51 321 170K 210

As a long-time member of the OSS community, and as a founder of an OSS enterprise software company, I agree with Satya 100% about "the frontier is an ecosystem". However, a hugely important learning of the last 40+ yrs is: open ecosystems outcompete closed ones. Furthermore, ecosystems are ultimately a reflection of the values & cultural DNA of their participants. This is going to be 1000% more important to the AI innovation ecosystem than mere software. Open Source right now is going through an existential crisis and cultural transition, but what follows in its footsteps for AI must incorporate certain ethos of permissionless innovation, user empowerment, inclusive communities, and the defense of an "innovation commons". Unfortunately, most large corporate players cannot comprehend (or justify to Wall St) these particular elements, because each one is an existential rejection of gatekeeping and capture.

Satya Nadella @satyanadella

a week ago

x.com/i/article/2065…

3K 8K 41K 66.1M 57K

Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇

Even more data to support what I have been talking about. The combination of model intelligence (and this includes human expertise) has a compounding effect unlike anything I've seen. There are too many assumptions that a large general-purpose model will be a one-size-fits-all. I don't buy it. The reality, and the research supports this, is that these different models show different strengths and capabilities. Understanding how to tap into them in combination is a huge unlock. All engineering teams need to be thinking about this more carefully as a strategy going forward. Especially now, given the trends from frontier models in terms of selective access.

OpenRouter @OpenRouter

2 weeks ago

Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇

714 2K 15K 6.2M 14K

Prepare for takeoff. ✈️ Flight simulator is now available globally on web to all users. goo.gle/4fBYnWO We've recently added many our most powerful professional desktop features to web. Elevation profiles, new import types, but there's always been one other feature you've been asking us to add to the web version of Google Earth, just for fun... Where will you fly? Share your best maneuvers, views, and flyovers with us!

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

🏰⛏️ From cardboard to castles! Our Fort Minecraft campers built an epic giant fort, crafted their own armor and tools, and brought their creativity to life one block at a time. 🎮✨

keyboard_arrow_left Previous keyboard_arrow_right Next

NEW: malware developers added nuclear & biological weapons text to to their spyware. Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner. Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky. When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit. We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted. In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation. H/T to colleagues that shared this with me socket.dev/blog/mini-shai…

Additionally, all of this exists in a modernity awash in tokenized power (aka "currency"). Historical models of power and governance allowed for a liberal use of physical violence and lethal enforcement of transgressions of trust. Lots of dead kings, princes, and peasants later, we have a more "domesticated" approach to power. Violence against agents is not condoned (outside of drug cartels). Therefore, agents can defect against their (powerful) principals, and still yet live, under the protection of the law.

@jamescham Because any attempt to do so imposes requirements on the Principal to assume responsible, and to transform themselves and the world to be one of peer & trustful stakeholding. The whole reason for agents is to scale *action* beyond the scope of moral responsbility.

Big ideas are becoming real. From raw space to a lobby designed to inspire creativity, collaboration, and innovation — this vision is finally taking shape. 🚧✨

nobody wants to hear this but the classical NASA systems engineering is the perfect model for developing code with LLMs. people try to approximate this with planning modes, but if you’re explicit in your docs it’s never been easier to build, test, and verify complex codebases.

🌌 Observer of Suns @ObserverSuns

4 years ago

“oh, every morning at 1:00am our language model regenerates the whole codebase from scratch based on the current requirements document. it’s more reliable than trying to make incremental edits”

16 50 710 0 171

@misislavski There isn’t one 😊

I don’t post much on Twitter any more but just wanted to surface the kind of rampant, unabashed white nationalism that has become normalized under Trump. And Twitter is fine being a place for these idiots to thrive, build a following, and metastasize their un-American poison.

Andrew Torba @BasedTorba

a month ago

My home isn’t your career move. You are not entitled to the inheritance my forefathers left me. No, I will not stand by and watch you rob opportunities from my children and kinsman.

61 750 6K 78K 128