https://epoch.ai/publicationshttps://epoch.ai/publications/a-rosetta-stone-for-ai-benchmarkshttps://epoch.ai/publications/a-time-invariant-version-of-laplace-s-rulehttps://epoch.ai/publications/ai-capabilities-can-be-significantly-improved-without-expensive-retraininghttps://epoch.ai/publications/algorithmic-progress-in-language-modelshttps://epoch.ai/publications/announcing-gatehttps://epoch.ai/publications/backward-forward-FLOP-ratiohttps://epoch.ai/publications/biological-sequence-models-in-the-context-of-the-ai-directiveshttps://epoch.ai/publications/can-ai-scaling-continue-through-2030https://epoch.ai/publications/challenges-in-predicting-ai-automationhttps://epoch.ai/publications/chinchilla-scaling-a-replication-attempthttps://epoch.ai/publications/chip-smugglinghttps://epoch.ai/publications/chips-topic-overviewhttps://epoch.ai/publications/compute-trendshttps://epoch.ai/publications/could-decentralized-training-solve-ais-power-problemhttps://epoch.ai/publications/data-movement-bottlenecks-scaling-past-1e28-flophttps://epoch.ai/publications/deep-think-mathhttps://epoch.ai/publications/direct-approach-interactive-modelhttps://epoch.ai/publications/do-the-returns-to-software-rnd-point-towards-a-singularityhttps://epoch.ai/publications/estimating-training-computehttps://epoch.ai/publications/expanding-our-analysis-of-biological-ai-modelshttps://epoch.ai/publications/explosive-growth-from-ai-a-review-of-the-argumentshttps://epoch.ai/publications/grok-4-mathhttps://epoch.ai/publications/grokking-bioanchorshttps://epoch.ai/publications/grokking-semi-informative-priorshttps://epoch.ai/publications/half-of-employed-ai-users-now-use-it-for-workhttps://epoch.ai/publications/hardware-failures-wont-limit-ai-scalinghttps://epoch.ai/publications/have-ai-capabilities-acceleratedhttps://epoch.ai/publications/how-fast-could-robot-production-scale-uphttps://epoch.ai/publications/how-much-does-it-cost-to-train-frontier-ai-modelshttps://epoch.ai/publications/how-predictable-is-language-model-benchmark-performancehttps://epoch.ai/publications/inference-economics-of-language-modelshttps://epoch.ai/publications/interviewing-ai-researchers-on-automation-of-ai-rndhttps://epoch.ai/publications/limits-to-the-energy-efficiency-of-cmos-microprocessorshttps://epoch.ai/publications/literature-review-of-transformative-artificial-intelligence-timelineshttps://epoch.ai/publications/machine-learning-model-sizes-and-the-parameter-gaphttps://epoch.ai/publications/measure-FLOP-empiricallyhttps://epoch.ai/publications/mirrorcode-preliminary-resultshttps://epoch.ai/publications/model-counts-compute-thresholdshttps://epoch.ai/publications/open-models-reporthttps://epoch.ai/publications/openai-stargate-where-the-us-sites-standhttps://epoch.ai/publications/optimally-allocating-compute-between-inference-and-traininghttps://epoch.ai/publications/parameter-countshttps://epoch.ai/publications/power-demands-of-frontier-ai-traininghttps://epoch.ai/publications/power-laws-in-speedrunning-and-machine-learninghttps://epoch.ai/publications/predicting-gpu-performancehttps://epoch.ai/publications/projecting-compute-trendshttps://epoch.ai/publications/revisiting-algorithmic-progresshttps://epoch.ai/publications/scaling-laws-literature-reviewhttps://epoch.ai/publications/the-direct-approachhttps://epoch.ai/publications/the-limited-benefit-of-recycling-foundation-modelshttps://epoch.ai/publications/the-longest-training-runhttps://epoch.ai/publications/tracking-large-scale-ai-modelshttps://epoch.ai/publications/trading-off-compute-in-training-and-inferencehttps://epoch.ai/publications/train-once-deploy-many-ai-and-increasing-returnshttps://epoch.ai/publications/training-compute-of-frontier-ai-models-grows-by-4-5x-per-yearhttps://epoch.ai/publications/trends-in-ai-supercomputershttps://epoch.ai/publications/trends-in-gpu-price-performancehttps://epoch.ai/publications/trends-in-machine-learning-hardwarehttps://epoch.ai/publications/trends-in-the-dollar-training-cost-of-machine-learning-systemshttps://epoch.ai/publications/trends-in-training-dataset-sizeshttps://epoch.ai/publications/what-do-economic-value-benchmarks-tell-ushttps://epoch.ai/publications/what-does-osworld-tell-us-about-ais-ability-to-use-computershttps://epoch.ai/publications/what-skills-does-swe-bench-verified-evaluatehttps://epoch.ai/publications/what-will-ai-look-like-in-2030https://epoch.ai/publications/what-you-need-to-know-about-ai-data-centershttps://epoch.ai/publications/where-autonomy-works-evaluating-robot-capabilities-in-2026https://epoch.ai/publications/who-is-leading-in-ai-an-analysis-of-industry-ai-researchhttps://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-datahttps://epoch.ai/publications/will-we-run-out-of-ml-data-evidence-from-projecting-dataset