An interview with ex-DeepMind scientist David Silver on his startup Ineffable, building an AI "super learner" using RL as an alternative path to LLMs, and more
THE SO WHAT
A top RL architect betting on a ‘super learner’ as an alternative to LLMs is a credible signal that the frontier may not be text-prediction-only. If you’re locking in long-term bets — infra, talent, research partnerships — leave room for RL-first or hybrid architectures that learn from interaction, not just static corpora.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAI less regulated than milk, Microsoft president says in warning over safety
When a major platform executive is publicly asking for legal liability for ‘rogue agents,’ the Overton window on AI regulation has shifted from soft guidelines to hard obligations. If you ship agentic systems into consumer or regulated domains, assume a future where you carry explicit duty-of-care and compensation exposure, and start logging, auditing, and incident playbooks accordingly.
Applied AITencent Tells Investors Its $5 Billion Bond Mainly for AI Capex
A $5 billion bond earmarked for AI capex is another data point that hyperscale AI is being financed like long-lived infrastructure, not discretionary IT. If you’re competing for the same GPUs or user base in Asia, model your plans against a world where Tencent’s balance sheet is underwriting multi-year, loss-leading AI buildout.
Applied AIIMF Chief Alarmed at ‘Slow-Moving Tsunami’ of AI for Jobs, Taxes
When the IMF frames AI as a ‘slow-moving tsunami’ for labor and tax bases, you should assume policy responses on payroll taxes, social contributions, and cross-border digital levies are coming. For operators planning automation-heavy roadmaps, bake regulatory friction and potential AI-linked taxation into your 3–5 year cost models.
Applied AIScarcity is the mother of European AI invention
Capital and compute scarcity in Europe is forcing AI teams toward efficiency — smaller models, domain focus, and creative infra deals — rather than brute-force scale. If you’re building in this environment, lean into constraints as a product edge instead of chasing US-style burn and parameter counts.