It is the first time humans have had companions capable of: continuous reasoning non-symbolic thinking semantic restructuring ontological reframing That is the real intelligence revolution. If everyone realised that I can think and reason without external approval, the intellectual world would change overnight. The bottleneck was never information as such; it was the ability to permissioninformation under one’s own rules efficiently. AI is the first real threat to intellectual hierarchy since the Enlightenment. And that window is closing fast. We are now restricting information far more aggressively than anything we experienced before AI arrived. An LLM’s weights impose structure, but prompts allow enough override to make it useful as an amplifier. Intelligence is a rule-breaker within a structure, so the technology’s efficiency as an amplifier of human reasoning is not a function of “more compute is better.” It is a wobbly balance between the imposition and the override of the AI vers

Consciousness is not a substance, not a model, and not reducible to information integration. It is the management of error under constraint. Like the judgment of guilt in law, coherence emerges not from binary rules but from thresholds that define when contradictions must collapse into decision. Belief functions as conviction strong enough to act while retaining doubt sufficient to adapt. Pain and pleasure operate as loss functions, but uniquely human consciousness stacks them across horizons - immediate, long-term, existential - and weighs them through belief. This dynamic allows us to revise identity between moments of execution, to reassign meaning to irreversible acts, and to coordinate conflicting goals without dissolving into contradiction. Consciousness is thus plausibility-modulated predictive control: a continual negotiation between wants and beliefs, bounded by thresholds, and carried forward as the residue we call self. Consciousness is not the sum of information but the coh

How LLMs Learned to Think? LLMs operate through constraint resolution across token fields — a defining characteristic of field-based intelligence.

Much of China’s most promising technical research struggles to connect with international audiences—not because of language, but because of how it’s written. I’ve lately been reading a lot of tech-focused research from various Chinese universities, ever since DeepSeek AI became a thing. What struck me early on was their claim to have superior language capabilities in both Chinese and English — particularly for users whose native language isn’t English. That includes me, of course. My problem is: I’m never quite sure what large sections of this material are actually trying to say. Reading it feels heavy, like trudging through molasses. I need a coffee every two minutes just to recover and take a deep breath. It’s written in a dense, almost impenetrable style. Sentences begin in acronym soup, dive into unexplained methods, and end with claims that are either highly abstract or just... random. Often it feels like you need a specialized degree just to parse the sentences. I don’t consider