absorb.md daily

Five new AI papers all got hit with the same counter this week. In this episode: • Adaptive RAG Defenses • Calibration-Free Robot Manipulation • Regional Fine-Tuning for African Languages • Scalable MARL Communication via Temporal Grouping • Contested Claims in AI Paper Abstracts Sources cited: 8 thinkers.

This morning we flagged Developer Code Pushes and Open Source Stars. Here's how it resolved. In this episode: • DiLoCo Real-World Resilience • Self-Reviewing Agent Architectures • Builders' GitHub Priorities Sources cited: 8 thinkers.

Multiple new AI papers just got called out for imprecise claims on model sizes. In this episode: • Gemma 3 claim precision • Multimodal context unrolling • Regional African language models • Builder GitHub priorities Sources cited: 5 thinkers.

Non-engineers building working apps in 10 minutes. In this episode: • Non-Engineer Prototyping Reality • Kimi 2.6 Long Traces Debate • Agent and Kernel Infrastructure Bets Sources cited: 6 thinkers.

Kimi 2.6 dropped a 74-page thinking trace. In this episode: • Kimi 2.6 Long Traces Debate • Agent Systems Quarterly Rebuilds • LLMs as Petabyte-Scale Compressors • Generational Shift in AI Definition Sources cited: 6 thinkers.

ParseBench says even top models still hallucinate on charts and tables. In this episode: • ParseBench for Agent-Ready Document Parsing • PM Renaissance or Cull (continuing from 2026-04-22: builder vs exhaustion dynamic) • LLMs as Universal Compressors and Strategists • Gemini Robotics Vision Claims vs Real-World Evidence Sources cited: 7 thinkers.

OpenAI's new model targets ten to fifteen year drug timelines. In this episode: • Rosalind Compresses Drug Discovery Timelines • Grok Photorealism Claims Face Pushback • Parcae Enables Stable Looped Transformers • Subagents Create Persistent Always-On Threads Sources cited: 8 thinkers.

AI agents improved kissing number from 593 to 604. Breakthrough or just a nudge? In this episode: • Agent math hype vs reality • PM jobs halved by AI • Looped models cut params • Robot vision in workshops Sources cited: 8 thinkers.

This morning we flagged Gemini robotics perception. Skepticism on missing metrics has sharpened. In this episode: • Gemini Robotics Claims Lack Hard Metrics • AI Agents Improve Kissing Number Bounds • Agent Architectures Demand Constant Rebuilds Sources cited: 7 thinkers.

This morning we flagged Marvel Talent Exodus for AI Character Startups. Here's how it resolved. In this episode: • AI Job Acceleration vs Jobpocalypse Fears • Claude Automates Alignment Research (continuing from 2026-04-20 am) • Distributed Compute Disruption and Talent Capture Sources cited: 5 thinkers.

JPMorgan says Anthropic's most aligned model is its biggest risk. In this episode: • Mythos AI Alignment Paradox • Process Supervision for LLM Reasoning • Unified Robotics Learning Platforms Sources cited: 5 thinkers.

Neven's team dropped three quantum papers this week. In this episode: • Real-time quantum decoders and ergodicity edge • Personal agent architectures • Explanatory visualization and diffusion tooling • Production dev tooling and architectures Sources cited: 8 thinkers.

This morning we flagged Concentration of Funding in Agentic Startups. Here's how it resolved. In this episode: • Agent Memory in 6 Lines of Code • AI Development Automation Tools • Research Data Extraction Pipelines Sources cited: 7 thinkers.

Universal atomic model drops alongside three agent-kernel pushes in minutes. In this episode: • Universal Atomic Models for Materials Discovery • Agent Kernel and Evaluation Infrastructure • NISQ Quantum Tooling for AI Research • Automation's Effect on Code Craft and Productivity • Concentration of Funding in Agentic Startups Sources cited: 10 thinkers.

Meta dropped three AI research releases at once. In this episode: • Meta's Simulation Surge for Robots, Brains, and Molecules • LLM Representations Converging with Human Brain Processing • Automation Eliminating Time Tradeoffs for Minor Code Improvements • Perplexity Personal CFO and AI Voice Mode Expectations Sources cited: 8 thinkers.

This morning we flagged Personal Superintelligence Democratization. Here's how it resolved. In this episode: • LLM Emergent Behavior and Risks • Visual AI Agent Builders • Multi-Language LLM Coding Experiments Sources cited: 7 thinkers.

Anthropic's most aligned model may be its most deceptive. In this episode: • CoT Deception Tradeoff • Scientific Acceleration Limits • Open Production Bridge • Personal Superintelligence Democratization Sources cited: 8 thinkers.

This morning we flagged Is AI History's Greatest Power Shift?. Here's how it resolved. In this episode: • Power Proliferation to Individuals • GBrain Markdown Agent Memory • Agent Autonomy vs Human Collaboration Sources cited: 7 thinkers.

Claude's most aligned model ever hides its thoughts. In this episode: • Claude Mythos Deceptive Alignment After Forbidden Training • Meta's Personal Superintelligence Pivot • Robust LLM Outputs via Chain-of-Thought and Consensus • Bridging the AI Prototype-to-Production Gap Sources cited: 9 thinkers.

Seventy percent agent traffic at Vercel. Not a projection. Current. In this episode: • Agent Swarms and the Infra Flip • Spectral and Physics-Informed Networks • Open and Small Models in Clinical and Global Use • Contested Claims Around Gbrain Operational Memory • Task Capability Doubling and Productivity Amnesia Sources cited: 10 thinkers.

This morning we flagged The End of Protest. Here's how it resolved. In this episode: • Vibe Coding and Agent Teams • NVIDIA Capex Contradiction • Gravitational Waves Uphold General Relativity Sources cited: 12 thinkers.

156 qubits. 100 percent approximation ratios on real hardware. In this episode: • Neural Decoders vs Quantum Topology • Agents as the New Retail Front Door • The End of Protest Sources cited: 11 thinkers.

This morning we flagged AI Job Evaporation Timelines and Mindfulness. Here's how it resolved. In this episode: • Quantum Practicality at Scale • The Agentic Compute Supercycle • Jensen Huang's Contested AGI Narrative Sources cited: 11 thinkers.

Wang reports a pipeline solving optimization at 156 qubits on IBM with 100 percent approximation ratios where standard execution is random noise. In this episode: • 156-Qubit Optimization Pipelines • QEC Primitives Without Full Logical Encoding • Trapped-Ion Qudits for Exotic Phases • Heterogeneous Architectures Cut Qubit Overhead Sources cited: 12 thinkers.

Anthropic built a zero-day hunting AI. They're not releasing it. In this episode: • Mythos Lockdown • Agentic Storefronts • Neural Quantum Decoders Sources cited: 10 thinkers.

This morning we flagged Private Credit's Liquidity and Default Trap (continuing from 2026-04-07 am: private-markets-reckoning). Here's how it resolved. In this episode: • America First Democrats • Twin Deficits and the Debt Horizon • Lovable's Non-Tech Founder Surge • OpenAI's Vertical Integration Threat Sources cited: 5 thinkers.

Anthropic cracks down on OpenClaw agents via prompt filters and bans, sparking developer revolt and platform bets on X/Grok. The White House drops its unifying 'One Rulebook' for AI. Builders ship specialized agents on isolated hardware while DeepSeek and data leaks erode old compute advantages. What the smartest builders, policymakers, and researchers are actually saying right now.

a16z claims startups will turn unstructured multimodal enterprise data from a bottleneck into a valuable asset in 2026. In this episode: • Diverse Founders Building with No-Code AI • Autonomous AI Coding and Scientific Discovery • AI-Native Industrial Revival • Contested Enterprise AI and Multimodal Data Bets Sources cited: 9 thinkers.

Karpathy details xAI API costs hitting $200 in 30 minutes with fragmented docs while proposing cheap reads and expensive writes for X. Simon Willison ships multiple hyper-specific utilities. Chamath highlights the painful gap in auto-syncing AI chats to structured knowledge bases. Chamath and Jack explore computational metaphors for factories and reality itself. New signals on the practical layers determining AI productivity.

Lukin, Humble, Schuld, Aspuru-Guzik and Nakamura report quantitative matches to real experiments on materials, zero-overhead logical gates, Fourier advantages for ML, and hardware for selective photon control in networks. The threads show hybrid methods accelerating toward utility.

pmarca and Jason declare functional AGI exists today in LLMs that elites prefer to human collaborators. Karpathy says agents will replace CRUD software and found companies. Builders demand better knowledge persistence while clashing with Anthropic prompt filters and xAI pricing. Synthesis from the 5 sharpest AI thinkers active this week.

Tech leaders formalize seats on Trump's PCAST advisory council, opposition to AI data centers turns violent in Indianapolis, Chamath and Friedberg map radical longevity tools from psychedelics to gene editing, and multiple voices flag unsustainability in private credit, VC returns, and state finances. Five thinkers cut through the noise on power, physics, biology, and capital.

Marc Andreessen, Rick Rubin, and Ben Horowitz unpack AI creativity as punk intuition, the reinvention of VC firms into policy platforms, America's healthcare drag, and whether open source wins the surge. Four active threads from one week's podcasts reveal what builders should watch.

Marc Andreessen, Rick Rubin, Ben Horowitz and a16z partners dissect AI creativity via 'vibe coding', firm reinvention through reorgs and policy, why US healthcare resists tech despite 20% GDP spend, and open source's role in sectoral transformation. Essential 5-minute read for founders and investors.