Library
Library
Every piece, with links to the source papers.
OpenAI's Astra Solves 10 Decade-Old Math ProblemsBreaking · AI researchThe Frontier Price Collapse: Near-Best AI Gets CheapAnalysis · economicsClaude Opus 5: Frontier-Class, at Half the Price of FableBreaking · model releaseGPT-5.6's Three Tiers and the 'ChatGPT Work' AgentBreaking · model releaseGemini Crosses 1 Billion Users as Google Starts Gemini 4Breaking · industryxAI Launches Grok 4.6, Matching Top-Tier Intelligence at Low CostBreaking · model releaseMoonshot's Kimi K3 Is Reported as the Largest Open-Weight Model YetBreaking · open modelsGoogle DeepMind's New Leadership and the Race to Catch UpBreaking · industryAnthropic Reports First Profit Amid IPO SpeculationBreaking · businessOpenAI's S-1 Expected as a Fall IPO Comes Into ViewBreaking · businessLong-Horizon Agents: AI That Works for HoursAnalysisMulti-Agent Systems for Genuinely Hard ProblemsTechnique explainerMachine-Verified Proofs: Why They Change EverythingAnalysisModel Tiers: Picking the Right One for the JobPractical guideWhy Frontier AI Suddenly Got CheapAnalysis · economicsWhat a 2-Million-Token Context Actually ChangesAnalysisNative Multimodal Generation: One Model, Many MediaExplainerOpen-Weight vs Closed Models in 2026AnalysisThe Economics of Test-Time ComputeAnalysis · economicsIs AI Doing Real Research Now?AnalysisReasoning Models vs Agents: Not the Same ThingExplainerWhy Benchmarks Like IMO and SWE-bench Matter (and Don't)ExplainerHow to Pick an AI Model in 2026Practical guideWhat Comes After the Current Frontier?AnalysisThe Router Pattern: Triage Cheap, Escalate HardPractical guideSK Hynix's Historic ~$28B US Listing and the Memory BoomAI Funding BriefNeura Robotics Raises a Record ~$1.4B for Full-Stack RobotsFunding newsOpenAI Preps a Confidential IPO at a Reported ~$730B ValuationBusiness newsEnterprise AI Agents Go MainstreamIndustry newsHBM: The Memory That AI Actually Runs Out OfAnalysisApple vs OpenAI: The Talent War for the Next DeviceBusiness newsGitHub Copilot Adds Its First Open-Weight Coding ModelProduct newsQualcomm Reportedly Eyes Tenstorrent in an AI-Chip PlayBusiness newsSouth Korea Unveils a Reported $880B AI and Chip PlanPolicy newsAccess to Top AI Tightens: ID Checks and Vetted PreviewsAnalysisAre We Hitting a Scaling Wall?AnalysisAGI: What the Term Actually MeansAnalysisThe Bitter Lesson, RevisitedAnalysisThe July 2026 Model Wave: Everyone Shipped at OnceNews roundupGemini 3.6 Flash Is HereGoogle · GeminiEmergent Abilities: Real or Measurement Artifact?AnalysisGemini 3.5 Pro: What's Confirmed vs What's RumoredAnalysisGPT-5.6: One Family, Three SizesOpenAI · GPT-5.6Chain-of-Thought Prompting, ExplainedTechnique explainerGrok 4.5 LandsxAI · GrokThe Reversal Curse: A Strange LLM Blind SpotAnalysisMeta's Muse Spark 1.1 and Its First Paid APIMeta · Muse SparkFew-Shot vs Zero-Shot PromptingTechnique explainerWhy AI Models Sometimes Get 'Lazy'AnalysisTemperature and Sampling: Controlling AI RandomnessTechnique explainerLost in the Middle: Long Context's Weak SpotAnalysisSelf-Consistency: Voting for Better AnswersTechnique explainerLLM Routers: The Right Model for Each QueryPractical guideReward Models: The Judge Behind Aligned AIExplainerMixture-of-Agents: When Many Models Beat OneTechnique explainerHow to Evaluate a RAG SystemPractical guideSemantic Caching: Reuse Answers, Cut CostsPractical guideVector Search at Scale: HNSW and ANN, ExplainedSystems explainerLLMOps: Running AI in ProductionPractical guideFine-Tuning on a Budget: QLoRA and BeyondMethod explainerObservability for AI ApplicationsPractical guideInference Servers: vLLM, TGI, and the Serving StackSystems explainerHow to Actually Measure HallucinationPractical guideAI Gateways: Running Many Models in ProductionPractical guideSparse vs Dense Retrieval, ExplainedExplainerText-to-Video: The State of the ArtAnalysisMultimodal Embeddings: Text and Images in One SpaceExplainerAI Music Generation in 2026AnalysisKnowledge Graphs Meet LLMsExplainerText-to-3D: Generating Objects and ScenesAnalysisAI Agent Frameworks: The LandscapeAnalysisRobotics and Embodied AI: Models That Act in the WorldAnalysisFunction Calling vs MCP vs Full AgentsExplainerSpeech-to-Speech: Skipping Text EntirelyExplainerToken Pricing, Explained: Why AI Costs What It CostsPractical guideDeepfakes and Detection: The Arms RaceAnalysisGuardrails vs Fine-Tuning for Safe AIPractical guideInterpretability: Looking Inside the Black BoxAnalysisAI and Jobs: A Level-Headed LookAnalysisThe Risks of Open-Source AIAnalysisC2PA: Proving Where Content Came FromExplainerAI in Healthcare: Real Uses vs HypeAnalysisCustomer-Support Agents That Actually ResolvePractical guideAI Tutors: Personalized Learning, ExplainedAnalysisJailbreaks and Prompt Injection, ExplainedSecurity explainerDPO vs PPO: How Preference Tuning WorksMethod explainerAI for Legal Work: Contracts and ResearchAnalysisKimi K3: The 2.8-Trillion-Parameter Bid for the TopMoonshot AI · Kimi K3Why Self-Correction Beats Longer ChainsAnalysis · reasoning modelsAlignment: What It Means and Why It's HardExplainerGRPO and the Rise of Group-Relative RLMethod explainerAI in Finance: Fraud, Forecasting, and RiskAnalysisChina's Open Models Are Winning on CostModel landscape 2026Answer Engines: The New Front Door to the WebAnalysisWatermarking AI Content: Can We Tell What's Generated?AnalysisRLHF, Explained SimplyExplainerThe Real Reason Models HallucinateAllen Institute for AIVibe Coding: Building Software by ConversationAnalysisThe Data Contamination ProblemAnalysisGLM-5 Is the Open Coding FrontierZhipu AI · GLM-5