Cognitive Warmup. China’s Zhipu AI (Z.AI) has an open-weighted GLM-5.2 model, which some researchers insist matches Anthropic’s controversial mythos in some cybersecurity and vulnerability identification tasks. It should be noted that while GLM lags behind Anthropic and OpenAI’s other models in more general tasks, this is a representation that Chinese AI models have systematically narrowed the gap in average capabilities compared to other AI companies.
GLM-5.2 is one of the 10 most used AI models on OpenStreetMap’s LLM leaderboard, alongside models from Anthropic, DeepSeek, Xiaomi, and Tencent. According to cybersecurity company Semgrep, in some benchmarking tests, GLM-5.2 outperformed Anthropic’s Cloud Opus 4.8 model (it was released in May). According to the researchers, depending on the instructions and specification, Opus 4.8 and GLM-5.2 can match Mythos in vulnerability-discovery. Would this be worrying?
a semiconductor breaker
IBM created the world’s first sub-1 nanometer (nm) chip technology, working exclusively at the 0.7 nm or 7 angstrom node. This achievement represents a historic moment for an industry that is finding ways to work around the physical limitations of traditional chip scaling. As semiconductors play an increasingly important role in everything from computing and home appliances to transportation systems and critical infrastructure, the ability to continue reducing the size of transistors while improving performance had wide-ranging implications.
At its center is IBM’s entirely new transistor architecture known as the “Nanostack”, and architecture that becomes the first architecture to vertically stack components like a skyscraper. IBM says this is a significant advancement from their own previously developed nanosheets used for many 3nm and 2nm chips. Here are some title capabilities:
- 100 billion transistors on a chip the size of a fingernail – that’s twice the density of IBM’s 2nm node chip, which made a huge leap forward just a few years ago
- Up to 50% higher performance or 70% more energy efficiency, providing dual performance of powerful computing with low power requirements.
- IBM also says this approach enables 40% scaling in SRAM (they say this is the biggest jump in a decade) which is important for AI workloads, cloud infrastructure and next generation electronic devices.
“IBM’s latest chip breakthrough is a historic moment in computing, taking technology beyond the nanometer era to the scale of atoms. With our new Nanostack architecture, we’re not just making smaller transistors, we’re reinventing how chips are made to deliver dramatically greater power and energy efficiency,” says Jay Gambetta, director of IBM Research and IBM Fellow.
This technology means it is possible to expand scaling below the 1 nm node, propelling the semiconductor industry into an era of Angstrom-level scaling where dimensions approach the size of individual atoms. While transistor nodes now refer to a generation of manufacturing technology rather than an exact physical dimension, IBM’s 0.7nm technology is a definitive demonstration that sustained scaling is highly possible. According to IBM, this nanostack architecture establishes a semiconductor roadmap that projects at least a decade of future scaling.
Google’s practical approach
Google’s recently published white paper titled “A Practical Approach to AI Governance in the US” is a clear and concrete effort to get the artificial intelligence regulation transformation on track. Google makes two very clear things. First, a clear distinction between frontier models and widely used AI applications. Secondly, what they call a “practical, evidence-based approach” to the overlap between the two.
I find merit in Google’s call for that separation as a foundation for any regulation. AI is in both areas, i.e. everyday use of chatbots and tools, as well as extraordinary scientific discoveries. The two streams (and there are many sub-streams, which cannot be ignored) cannot be regulated in the same way, with the same intensity. Google’s call points to an ineffective duality that the AI ​​field is currently grappling with – it’s either draconian over-regulation that stifles progress, or inattentive regulation that inevitably puts users in danger. Google advocates a “middle way” – a practical, evidence-based approach tailored to the diverse realities of different AI systems.
The cornerstone of Google’s proposal is a bifurcated regulatory framework that actively differentiates between “frontier AI” (the most advanced, highly capable models) and “widely deployed AI applications” (everyday devices with fewer, narrower capabilities). By avoiding a one-size-fits-all legislative blanket, Google argues that regulators can successfully target specific, identifiable real-world harms without fundamentally disrupting the underlying computer science.
For Frontier AI, Google believes these are important:
- An independent regulatory organization that can keep pace with rapidly advancing AI research and development.
- Scientific benchmarks to identify leading capabilities in the cyber and chemical, biological, radiological and nuclear (CBRN) domains, complemented by clear safety and security standards for building, testing and deploying the most advanced AI systems.
- Annual audits to demonstrate compliance with procedural, and ultimately substantive, security standards, supported by model transparency and reporting requirements
For frontier AI, where the implications span systemic security and national security, Google proposes the creation of a Frontier Regulatory Organization (FARO) – an independent, federally overseen and industry-supported entity. FARO will be tasked with setting efficient security standards and verifying independent, voluntary audits of the most advanced AI models. This model reflects how other important areas are managed, providing a flexible framework that can keep pace with rapid algorithmic innovation.
“For AI applications enabled by models at lower levels of capability, the federal government does not need new regulatory regimes that duplicate or conflict with existing law. AI applications like chatbots raise social and consumer protection issues distinct from the national security issues raised by the most advanced frontier AI models. For these broader applications the federal government can amend existing laws and regulations to address real-world outputs and specific harms,” ​​Google writes in its white paper.
Beyond model governance, the white paper addresses the need for a broader ecosystem for sustainable AI stewardship. They discuss the topic of public-private initiatives to increase America’s energy production (this is a very uncomfortable topic at this point) and a transmission grid similar to the “Eisenhower Highway Program”, the importance of information integrity, urging regulators to mandate watermarking technologies like SynthID, and mandating tamper-resistant cryptographic provenance standards like C2PA for generic AI services.
Cost, value and sensitivity
uh oh! Is running AI proving more expensive than the humans it urged corporations to replace? Common sense hasn’t exactly been common as investors and boardrooms have become obsessed with AI over the past few years. Anyway, don’t believe me, but believe the numbers.
According to research firm Gartner, AI coding costs will exceed the salary of an average software developer by 2028. At its core will be the extremely high cost of consuming large language model (LLM) tokens and an industry that largely operates on a consumption-based licensing model. Gartner warns that organizations are rapidly shifting from early experimentation to large-scale deployment of AI coding agents, while significantly underestimating the financial impact of this growing token use. Tokens—the fundamental units of data processed by generative AI models—directly determine the cost of these software tools under new consumption-based pricing structures.
AI companies are undoubtedly smart. A subtle shift from more financially predictable seat-based licensing to a more volatile (and therefore expensive) token-based structure, they are designed to generate revenue. Don’t save corporate money by replacing humans with AI. I wouldn’t put it past anyone to compound the issue with a lack of transparency regarding token consumption calculations and an apparent inability to accurately plan budgets as well as track the cost of results.
It wouldn’t be strange to say that most organizations, apart from boardroom enthusiasm about being AI-first or whatever the terminology, lack the maturity and framework required to proactively measure the cost of AI with real business impact.
