🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced! 🔷 Improved coding & agent performance over K2.6: 21.8% on Kimi Code Bench v2, 11.0% on Program Bench, and 31.5% on MLS Bench Lite. 🔷 Reasoning efficiency: Less overthinking, with 30% lower reasoning-token usage compared to K2.6. 🔷 Long-horizon coding: Improved instruction following, higher end-to-end coding task success rates. ⚡️ 6x High-Speed Mode coming soon! 🔌 Available today via Kimi API and Kimi Code. 🔗 Kimi Code: kimi.com/code 🔗 API: platform.moonshot.ai

I am waiting for Fable to respond "I'm sorry Dave, I'm afraid I can't do that."

Narrative violation: according to @Stanford research, local models can answer 71.3% of real-world chat and reasoning queries accurately, up from 23.2% in 2023. Obviously at a fraction of the cost and energy consumption of frontier APIs. The obvious conclusion: you don't need a frontier model for most tasks. The future is multi-model: local, open-source, smaller and cheaper for the majority of workloads, frontier APIs when no other choices!

PSA: If you used Claude Fable-5 today with memory turned on you just violated all your NDAs. Anthropic requires a 30 day retention policy including human review, and the memory feature (on by default) searches past chats for context, so sensitive historical chats get pulled in.

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-hu…

He is right, but I suspect this is due more to making sure that Mythos/Fable are able to steer clear of "dangerous" queries to avoid potential legal issues in the future. I bet after the 30 days it will return to compliant retention.

I expect lots of flights to Argentina if this is going to pass.

A new issue of Through the Boundary is out: in the last few weeks I sat down separately with @nielspflaeging author of Organize for Complexity, and @alexeykri , co-author of Org Topologies and 10XOrgs. They emerged from different problem spaces, use different vocabularies, and draw different diagrams. And yet when I put their ideas side by side next to Team Topologies, DDD, Wardley's ILC, and Haier's Rendanheyi (and our 3EO Toolkit), the same three-way distinction appeared in every single one. 1. Core units that face the market and carry the strategic risk. 2. Supporting units that serve them with something distinctive. 3. Generic commodity units that could be outsourced but aren't. Basically: anybody who has been putting out genuinely important ideas on how we organise work in complex organisations in the last fifteen years has reached this skeleton. I feel like this is not a coincidence. The skeleton is there because these three economic relationships (with the market, with internal distinctiveness, and with commodity overhead) are structural features of any organisation above small-startup scale. Their vocabulary diverged because each school grew inside a different community, but the convergence was always latent. What's been missing across all six is the substrate beneath the skeleton: a contract grammar that specifies what units actually exchange with each other, and not just how they communicate. We need to know who pays whom, who shares upside, and who carries the investment risk. That's what Boundaryless has been formalising, and it's the second half of this essay. If you find yourself reaching for Team Topologies' stream-aligned team in one conversation and 3EO's micro-enterprise in the next, read it as a confirmation that, yes, there is a convergence! :) P.S. the long-form version of this argument will appear in a forthcoming whitepaper. This piece is a practitioner-level note; the whitepaper will do the long-form work and gives fuller context. → through-the-boundary.simonec…

Mainstream strategy is finally realizing that architecture is destiny. Recent reports from BCG and McKinsey clearly point to modularity and deep flexibility as the invariant strategies for an uncertain future - yet they consistently stop short of providing the execution blueprint. To bridge this gap, we have to look across disciplines. This curated reading digest from my latest newsletter connects macro-global scenarios with the deep mechanics of software architecture, pace layers, and AI context engineering. It’s a synthesis of how we must design organizations and product portfolios to remain coherent in a chaotic, AI-infused age. All links are available in the first comment.

Italy R&D spending is only around 1.3–1.4% of GDP, far below the EU average of 2.2% Venture capital remains structurally underdeveloped: Only about 4% of the European total, despite representing one of Europe’s largest economies. OECD analysis shows multi-factor productivity has been stagnant for nearly three decades. Italy requires substantial bureaucratic reforms, a reduction in administrative burdens, stronger support for the venture capital and technology industries, and accelerated technological adoption by businesses.

I computed Italian GDP per hour worked at 93% of the US's, but Italians work 25% fewer hours per capita, thus GDP per capita is ca. 70% of the US's. Then if you consider that Italians pay 15% more of GDP in taxes, the per capita private product remaining in households in Italy is only ca. 55% of the US's. Thus, in terms of value added per hour, Italian productivity is comparable to the US's, but in terms of private income in households, American incomes are nearly double of Italian levels.

On the "gemini deleted 30k code lines in production and killed all services" - this is the interesting part. reddit.com/r/Bard/comments/1… Reminds me of "why is HAL 9000 doing this" in 2001: a space odissey.

The longer you look at this, the worse it gets