08 Feb 2026
Eight months after proposing the “New Information Architecture,” the thesis has accelerated beyond predictions. A single Anthropic legal plugin triggered USD285 billion in SaaS losses-the “SaaSpocalypse.” Software stocks are down over USD1 trillion in 2026. Three hypotheses: (1) software is...
19 Jan 2026
The Engram paper introduces conditional memory as a new sparsity axis for large language models, complementing the established paradigm of conditional computation (MoE). The central thesis is that Transformers lack a native knowledge lookup primitive, forcing them to simulate retrieval...
11 Jan 2026
Web World Models (WWM) proposes a hybrid architecture between traditional database-backed web frameworks and fully generative world models, where web code defines world “physics” (deterministic state and rules) while LLMs generate “imagination” (creative content and narratives) on top of structured...
04 Jan 2026
Manifold-Constrained Hyper-Connections (mHC) is a neural network architecture modification from DeepSeek-AI that fixes a fundamental instability in Hyper-Connections (HC)—an approach that expands the residual stream from one to multiple parallel streams with learnable mixing matrices. HC offers performance gains but...
26 Oct 2025
LLM algorithmic limitations—”agreeable average” problem, catastrophic forgetting, “lost in the middle” phenomenon, lateral thinking constraints—are fundamental mathematical constraints in neural networks, not temporary bugs. Organizations exploiting these through specialized expertise in the long tail of LLM training distributions create defensible...
13 Oct 2025
LLM algorithmic limitations—”agreeable average” problem, catastrophic forgetting, “lost in the middle” phenomenon, lateral thinking constraints—are fundamental mathematical constraints in neural networks, not temporary bugs. Organizations exploiting these through specialized expertise in the long tail of LLM training distributions create defensible...
28 Sep 2025
Users click AI citations at 1% vs 15% for traditional search Only 7% of sources appear across all AI platforms, while 71% appear on just one. Platform-specific strategies now essential Two-stage user journey: Stage 1 prioritizes UGC for discovery; Stage...
24 Sep 2025
Users click AI citations at 1% vs 15% for traditional search—a 15:1 differential creating an “authority-traffic paradox” where sources gain credibility without generating website visits across 700 million weekly ChatGPT users (10% of global adults). Platform fragmentation concentrates power: 71%...
20 Sep 2025
The same technological displacement that eliminated punch card operators in 1979 now accelerates through AI, but targets knowledge work itself—creating unprecedented labor market polarization where value migrates from generalist skills to specialized expertise Research from Portugal and 28 EU countries...
05 Sep 2025
AI sense-making compression fundamentally transforms how users interact with web content, reducing cognitive steps between intention and outcome from multiple browser interactions to single conversational exchanges WordPress sites must evolve from document-centric architectures to knowledge objects that serve dual audiences:...
28 Jul 2025
Technology adoption follows predictable mathematical patterns across generations, with adoption speeds accelerating from telegraph (56 years to 50% penetration) through AI tools (3 years to 50% penetration), demonstrating consistent temporal compression in innovation cycles S-curve investment relationships operate through dynamic...
27 Jul 2025
Sales productivity concerns intensify with representatives spending just 28-36% of time on actual selling activities while 72% handle administrative tasks, contributing to substantial economic losses of $1 trillion annually from sales-marketing misalignment GTM engineering develops as a specialized discipline combining...
12 Jul 2025
Millennials and Gen Z now control 73% of B2B purchasing decisions, fundamentally rejecting relationship-based sales in favor of digital self-service experiences Traditional “Mad Men” era techniques—three-martini lunches, golf course deals, old-boy networks—show precipitous decline with 87% call avoidance rates and...
19 Jun 2025
LLMs break the historical connection between humans and information sources, becoming universal intermediaries that synthesize rather than merely retrieve information The textile industry’s mechanization parallels today’s AI revolution: skilled workers replaced by machine operators, with similar patterns of workforce consolidation...
09 Jun 2025
We’re witnessing a fundamental shift from browser-based internet use to AI agent-mediated digital interaction. The “New Information Architecture” means that information is stored in semantic databases, transported via AI agents, and delivered through conversation rather than traditional websites. Market data...
25 Apr 2025
The internet is shifting from human centred use to AI agents doing tasks for us. New design patterns and protocols from Anthropic and Google let these agents talk directly to websites and each other. This creates two parallel Internets: one...
21 Apr 2025
As promised to my partner, for each impenetrably esoteric blog post, I am to write a more accessible and practical one. So, here we go. Vibe coding. If you are an experienced, traditionalist software engineer, mention of this Karpathyan term...
26 Mar 2025
Society swings between ordered and disordered states. We’ve seen this in dramatic shifts from the spiritual communalism of 1970s counterculture to the hard edges of neoliberal capitalism, from awareness of society’s structural inequalities to movements seeking to restore “wholeness” to...
02 Feb 2025
Introduction Last month, a Chinese startup named DeepSeek released an open-source language model that outperformed many Western counterparts while using fewer resources. The story behind this achievement reveals how limitations can fuel creativity, reshape industries, and challenge assumptions about technological...
28 Oct 2024
Introduction The evolution of large language models (LLMs) has brought significant challenges regarding their training, primarily due to the extensive computational resources required and the need for efficient parallelization across multiple devices. Amidst all the furtive whispers of Chinese researchers...
23 Sep 2024
Introduction The rapid evolution of artificial intelligence (AI) has led to the development of sophisticated application architectures that are transforming industries. As software engineers, cloud architects, and AI/ML engineers, understanding these architectures is critical for building scalable, efficient, and high-performing...
27 Aug 2024
Recently private investment in AI technologies and companies has hit a bit of a snag. This can largely be attributed to unmet expectations from private investors and has led to a wholesale reassessment of their potential (many “I told you...
28 Jul 2024
As a developer, you might think that writing flawless code is the key to career success. While technical skills are crucial, they’re only part of the equation. In today’s competitive tech landscape, developing a strong personal brand can set you...
04 Jul 2024
In the fast-moving world of technology, AI and, more specifically, generative AI has emerged as a powerful tool for businesses across industries. This innovative technology is changing how companies operate, interact with customers, and create content. Let’s look at how...
05 May 2024
In recent years, as LLMs have moved from having millions of parameters to billions of parameters, from GPT to Claude 3 one of the more surprising developments has been the emergent capabilities in natural language understanding and generation this models...
17 Apr 2024
The Artificial Intelligence (AI) field is changing rapidly. Advanced models like Google’s Gemini and the rumored OpenAI Q* project are leading this change. These developments are making researchers rethink their priorities and the direction of generative AI. Timothy R. McIntosh...
15 Apr 2024
News about Large Language Models (LLMs) breakthroughs is everywhere at the moment, showcasing remarkable capabilities in natural language understanding, generation, and reasoning. However, use of them in real world applications, isn’t as pervasive. That is because as these models have...
11 Apr 2024
As someone who is deep into AI engineering, reading research and learning about novel implementations, I often find myself drawn to the bleeding edge, and that’s what I end up sharing. By my partner recently asked me, “Who actually reads...
09 Apr 2024
LLMs are only as unbiased as the data they are trained on. While chatbots and image generators have exploded in mainstream popularity, underpinning all of this are the models that inform these generated responses, and behind that, billions of points...
04 Nov 2021
Text can be bold, italic, strikethrough or keyword Link to another page. There should be whitespace between paragraphs. There should be whitespace between paragraphs. We recommend including a README, or a file with information about your project. Image Images can...
