ai race

Bilingual

The Actual Reason Why Google "Fell Out" of the AI Race Changes Everything
2 months ago
- Google has withdrawn from the AI race focused on recursive self-improvement (RSI) that OpenAI and Anthropic are pursuing.
- Google DeepMind CEO Demis Hassabis believes world models (understanding/simulating real world) are the correct path to AGI, not automating AI research with coding agents.
- OpenAI and Anthropic are heavily investing in RSI, using coding agents to improve models autonomously, aiming for AGI via scaling and brute force.
- Google's model releases have lagged behind, and its latest Gemini Flash 3.6 scores poorly.
- The article suggests Google's strategy could either make it a leader or a laggard.
- Google is betting on a different theory of intelligence, and its existing business can subsidize AI development.
Mad 2.0: Mutually Assured Disruption
2 months ago
- The 'bitter lesson' states that raw computation always outperforms human cleverness.
- Those with the most compute—individuals, companies, or governments—will dominate.
- The race to hoard compute is likened to an AI Manhattan Project, making related resources irrationally valuable.
- When anything can be faked, trust in information collapses.
- Chains of human vouching, akin to 'SSL for people', can restore trust through shared history, reputation, and mutual risk.
- Relationship history is harder to fake than digital verification systems.
Wax On, Weights Off: Knowledge Distillation Explained
2 months ago
- Knowledge distillation is a technique where a small student model learns from a large teacher model's outputs, compressing years of R&D into weeks.
- Three types exist: response-based (matching outputs), feature-based (matching internal representations), and relation-based (matching relationships between examples); only response-based is feasible via API for closed models.
- Distillation against a closed model is a distributed attack using thousands of API accounts, query diversification (self-instruct loops, Evol-Instruct), and data engineering to harvest and clean training data.
- The pipeline is primarily a data engineering system: prompt generation, collection (with evasion of defenses), curation (filtering/quality scoring), and training (SFT, LoRA, preference optimization).
- Economic advantage: distillation pays only for compression, skipping costly discovery (data curation, architecture search, alignment), enabling fast-follower catch-up at lower cost.
- Frontier developer tech stack includes pretraining on massive data, alignment (RLHF, reward modeling), evaluation, and inference defense—none of which a distiller invests in unless selling a model.
- Distillation commoditizes the teacher's moat but forces the student into a perpetual catch-up position, inheriting the teacher's biases and capability ceiling.
- The US-China AI race is reframed by distillation: Chinese labs like DeepSeek, Moonshot, and MiniMax have been caught industrial-scale harvesting of Claude, leading to access restrictions and geopolitical tensions.

ai race

The Actual Reason Why Google "Fell Out" of the AI Race Changes Everything

Mad 2.0: Mutually Assured Disruption

Wax On, Weights Off: Knowledge Distillation Explained