https://epoch.ai/
Epoch AI
Epoch AI is a research institute investigating key trends and questions that will shape the trajectory and governance of Artificial Intelligence.
epoch ai
https://epoch.ai/gradient-updates/keeping-up-with-the-gpts
Keeping up with the GPTs | Epoch AI
Can Chinese and open model companies compete with the frontier through e.g. distillation and talent?
keeping upwith theepoch aigpts
https://epoch.ai/about/team/ben-cottier
Ben Cottier | Epoch AI
Ben Cottier is a senior researcher at Epoch AI. He leads the Frontier Data Centers project. Besides data centers, Ben is interested in AI cost trends and the di
epoch aiben
https://jobs.lever.co/epoch-ai/ab88ba6e-6a92-44cc-8830-a2dafca31f1a
Epoch AI - Data Scientist (Contract)
Epoch is seeking part-time data scientists to assist with our AI research efforts. This role involves reviewing technical literature, tracking benchmark data,...
ai data scientistepochcontract
https://epoch.ai/blog/how-fast-could-robot-production-scale-up
How Fast Could Robot Production Scale Up? | Epoch AI
We look at reference classes, factory buildout timelines, and upstream component supply to estimate plausible production rates for humanoids, quadrupeds,...
how fastproduction scaleepoch aicouldrobot
https://epoch.ai/trends
Trends in Artificial Intelligence | Epoch AI
Frontier AI systems are advancing rapidly from increases in compute, hardware performance, software efficiency, and investment. This dashboard explores those...
artificial intelligenceepoch aitrends
https://epoch.ai/data-insights/output-length
LLM responses to benchmark questions are getting longer over time | Epoch AI
Epoch AI is a research institute investigating key trends and questions that will shape the trajectory and governance of Artificial Intelligence.
over timeepoch ai
https://epoch.ai/gradient-updates/less-than-70-percent-of-frontiermath-is-within-reach-for-todays-models
Less than 70% of FrontierMath is within reach for today’s models | Epoch AI
57% of problems have been solved at least once.
less thanwithin reachepoch aifrontiermathmodels
https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use
How much energy does ChatGPT use? | Epoch AI
This Gradient Updates issue explores how much energy ChatGPT uses per query, revealing it's 10x less than common estimates.
how muchepoch aienergychatgptuse
https://epoch.ai/blog/deep-think-math
Evaluating Gemini 2.5 Deep Think's math capabilities | Epoch AI
Improved use of knowledge and precision, helpful for research, more conceptual in geometry, but limited creativity and citation issues.
deep thinkepoch aievaluatinggeminimath
https://epoch.ai/topics
Topics | Epoch AI
Browse Epoch AI's research and analysis by topic.
epoch aitopics
https://epoch.ai/data/data-centers-documentation
Frontier Data Centers Documentation | Epoch AI
Epoch’s Frontier Data Centers Hub is an independent database tracking the construction timelines of major AI data centers through high-resolution satellite...
frontier data centersepoch aidocumentation
https://epoch.ai/blog/predicting-gpu-performance
Predicting GPU performance | Epoch AI
We forecast field-effect transistor-based GPUs will plateau sometime between 2027 and 2035, offering a performance between 1e14 and 1e15 FLOP/s in FP32.
epoch aipredictinggpuperformance
https://epoch.ai/data-insights/b200-cost-breakdown
NVIDIA's B200 costs around $6,400 to produce | Epoch AI
We estimate that NVIDIA’s B200 costs around $6,400 to produce, with High Bandwidth Memory accounting for half of that.
epoch ainvidiab200costsaround
https://epoch.ai/data/gpu-clusters
Data on GPU clusters | Epoch AI
Our database of over 500 GPU clusters and supercomputers tracks large hardware facilities, including those used for AI training and inference.
gpu clustersepoch aidata
https://epoch.ai/blog/measure-FLOP-empirically
How to measure FLOP for neural networks empirically? | Epoch AI
Computing the utilization rate for multiple Neural Network architectures.
how to measureneural networksepoch aiflop
https://epoch.ai/blog/grok-4-math
Evaluating Grok 4’s math capabilities | Epoch AI
It's good at involved computations, improving at proofs, and useful for literature search. It still favors low-level grinds and leans on background knowledge.
epoch aievaluatinggrokmathcapabilities
https://epoch.ai/blog/the-direct-approach
The Direct Approach | Epoch AI
We propose a method using neural scaling laws to estimate the compute needed to train AI models to reach human-level performance on various tasks.
epoch aidirectapproach
https://epoch.ai/blog/algorithmic-progress-in-language-models
Algorithmic progress in language models | Epoch AI
Progress in pretrained language model performance outpaces expectations, occurring at a pace equivalent to doubling computational power every 5 to 14 months.
language modelsepoch aialgorithmicprogress
https://epoch.ai/latest
Latest | Epoch AI
Explore Epoch AI’s most recent work, including research papers, data insights, our newsletter, and podcast episodes in one comprehensive view.
epoch ailatest
https://epoch.ai/gradient-updates/the-changing-drivers-of-llm-adoption
The changing drivers of LLM adoption | Epoch AI
Public data as well as our original polling suggest LLM adoption is roughly on trend, but the underlying drivers are shifting.
changing driversepoch aillmadoption
https://epoch.ai/blog/chinchilla-scaling-a-replication-attempt
Chinchilla scaling: A replication attempt | Epoch AI
We replicate Hoffmann et al.’s parametric scaling law estimates, finding issues and providing better-fitting estimates that align with their other methods.
epoch aichinchillascalingreplicationattempt
https://epoch.ai/cookies
Cookie Policy | Epoch AI
Learn how Epoch AI uses cookies and similar technologies on this website.
cookie policyepoch ai
https://epoch.ai/blog/epoch-and-fri-mentorship-program-summer-2023
Epoch AI and FRI mentorship program summer 2023 | Epoch AI
We’re launching the Epoch and FRI mentorship program for women, non-binary, and transgender people interested in AI forecasting.
epoch aimentorship programfrisummer
https://epoch.ai/blog/parameter-counts
Parameter counts in machine learning | Epoch AI
Compiling a large dataset of machine learning models to determine changes in the parameters counts of systems since 1952.
machine learningepoch aiparametercounts
https://epoch.ai/gradient-updates/beyond-benchmark-scores-analysing-o3-mini-math-reasoning
Beyond benchmark scores: Analyzing o3-mini’s mathematical reasoning | Epoch AI
Examining o3-mini's math reasoning: an erudite, vibes-based solver that excels in knowledge but lacks precision, creativity, and formal human rigor.
mathematical reasoningepoch aibeyondbenchmarkscores
https://jobs.lever.co/epoch-ai/de7b4c71-ece2-454a-be70-e7b75c5f3b23
Epoch AI - Researcher / Senior Researcher
Epoch AI is looking for experienced researchers to lead new projects on one of multiple expanding teams. About the role We’re seeking Researchers and Senior...
epoch airesearchersenior
https://epoch.ai/about/team
Our Team | Epoch AI
Meet the team behind Epoch AI and learn about the values we are committed to.
our teamepoch ai
https://epoch.ai/about/transparency
Transparency | Epoch AI
Support Epoch AI’s research on the future of AI through a donation.
epoch aitransparency
https://epoch.ai/blog/please-report-your-compute
Please report your compute | Epoch AI
Compute is essential for AI performance, yet often underreported. Adopting reporting norms would improve research, forecasts, and policy decisions.
please reportepoch aicompute
https://epoch.ai/blog/estimating-training-compute
Estimating training compute of deep learning models | Epoch AI
We describe two approaches for estimating the training compute of Deep Learning systems, by counting operations and looking at GPU time.
deep learning modelsepoch aiestimatingtrainingcompute
https://epoch.ai/about/careers
Careers | Epoch AI
Explore Epoch AI’s career opportunities, apply to open positions, and help shape the future of AI.
epoch aicareers
https://epoch.ai/blog/epoch-impact-report-2023
Epoch AI 2023 impact report | Epoch AI
In 2023, Epoch published nearly 20 reports on AI, added hundreds of models to our database, helped with government policies, and raised over $7 million.
epoch aiimpact report
https://epoch.ai/data-insights/openai-compute-spend
Most of OpenAI’s 2024 compute went to experiments | Epoch AI
epoch aicomputewentexperiments
https://epoch.ai/topics/open-models
Open-Weight Models: Data & Research | Epoch AI
Some of the most powerful AI systems in the world are available for anyone to download, run, and build on. Others remain fully proprietary. These open models,...
open weight modelsdata researchepoch ai
https://epoch.ai/about/team/ricardo-pimentel
Ricardo Pimentel | Epoch AI
Ricardo Pimentel is an operations associate at Epoch AI, focusing on business operations. He has a background in finance, venture capital and venture philanthro
epoch airicardopimentel
https://epoch.ai/data-insights/llm-inference-price-trends
LLM inference prices have fallen rapidly but unequally across tasks | Epoch AI
Epoch AI is a research institute investigating key trends and questions that will shape the trajectory and governance of Artificial Intelligence.
llm inferenceepoch ai
https://epoch.ai/gradient-updates/the-promise-of-reasoning-models
The promise of reasoning models | Epoch AI
AI reasoning models will achieve superhuman performance in math and coding, yet their economic applications will lag behind, limiting real-world impact.
the promisereasoning modelsepoch ai
https://epoch.ai/contact
Contact | Epoch AI
Get in touch with Epoch AI. We will respond if we can accommodate your request.
contact epochai
https://epoch.ai/frontiermath/open-problems
FrontierMath: Open Problems - Unsolved Mathematical Challenges | Epoch AI
A collection of unsolved mathematical problems designed to test AI systems' ability to advance human mathematical knowledge.
open problemsepoch aifrontiermathunsolvedmathematical
https://epoch.ai/blog/backward-forward-FLOP-ratio
What’s the backward-forward FLOP ratio for neural networks? | Epoch AI
Determining the backward-forward FLOP ratio for neural networks, to help calculate their total training compute.
neural networksepoch aibackwardforwardflop
https://epoch.ai/benchmarks/about
About | Benchmarking | Epoch AI
Epoch's AI Benchmarking Hub brings together results from many of the most informative AI benchmarks into one consistent, searchable place.
about benchmarkingepoch ai
https://epoch.ai/gradient-updates/the-software-intelligence-explosion-debate-needs-experiments
The software intelligence explosion debate needs experiments | Epoch AI
The existing debate rests on data and assumptions that are shakier than most people realize. To make progress, we need better evidence, and experiments are the...
the softwareintelligence explosionepoch aidebateneeds
https://epoch.ai/blog/the-limited-benefit-of-recycling-foundation-models
The limited benefit of recycling foundation models | Epoch AI
Reusing pretrained models can save on training costs, but it's unlikely to significantly boost AI capabilities beyond modest improvements.
the limitedfoundation modelsepoch aibenefitrecycling
https://epoch.ai/data/data-centers
Frontier Data Centers | Epoch AI
Open database of AI data centers using satellite and permit data to show compute, power use, and construction timelines.
frontier data centersepoch ai
https://epoch.ai/blog/power-laws-in-speedrunning-and-machine-learning
Power laws in speedrunning and machine learning | Epoch AI
Our model suggests ML benchmarks aren’t near saturation. While large improvements are rare, we find 1OOM gains happen roughly once in every 50 instances.
power lawsmachine learningepoch aispeedrunning
https://epoch.ai/data-insights
Data Insights | Epoch AI
Epoch AI’s data insights break down complex AI trends into focused, digestible snapshots.
data insightsepoch ai
https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use/
How much energy does ChatGPT use? | Epoch AI
This Gradient Updates issue explores how much energy ChatGPT uses per query, revealing it's 10x less than common estimates.
how muchepoch aienergychatgptuse
https://epoch.ai/data/machine-learning-hardware
Data on Machine Learning Hardware | Epoch AI
We present key data on over 170 AI accelerators, such as graphics processing units (GPUs) and tensor processing units (TPUs), used to develop and deploy...
machine learningepoch aidatahardware
https://epoch.ai/gradient-updates/quantifying-the-algorithmic-improvement-from-reasoning-models
Quantifying the algorithmic improvement from reasoning models | Epoch AI
Reasoning models were as big of an improvement as the Transformer, at least on some benchmarks
reasoning modelsepoch aialgorithmicimprovement
https://epoch.ai/data-insights/benchmark-correlations
Benchmark scores are well correlated, even across domains | Epoch AI
Model rankings are remarkably consistent across most AI benchmarks. Across 16 benchmarks with at least 10 models overlapping, the median pairwise correlation...
epoch aibenchmarkscoreswellcorrelated
https://epoch.ai/gradient-updates/how-fast-can-algorithms-advance-capabilities
How fast can algorithms advance capabilities? | Epoch AI
This week's issue is a guest post by Henry Josephson, who is a research manager at UChicago's XLab and an AI governance intern at Google DeepMind.
how fastepoch aialgorithmsadvancecapabilities
https://epoch.ai/blog/trading-off-compute-in-training-and-inference
Trading off compute in training and inference | Epoch AI
We characterize techniques that induce a tradeoff between spending resources on training and inference, outlining their implications for AI governance.
trading offepoch aicomputetraininginference
https://epoch.ai/blog/compute-trends
Compute trends across three eras of machine learning | Epoch AI
We’ve compiled a comprehensive dataset of the training compute of AI models, providing key insights into AI development.
machine learningepoch aicomputetrendsacross
https://epoch.ai/blog/epoch-impact-report-2022
Epoch AI 2022 impact report | Epoch AI
Our impact report for 2022.
epoch aiimpact report
https://epoch.ai/blog
Papers & Reports | Epoch AI
Explore Epoch AI's latest insights on the trajectory of AI, including topics like compute, data, algorithmic advances, economics, and forecasting.
epoch aipapersreports
https://epoch.ai/gradient-updates/frontier-language-models-have-become-much-smaller
Frontier language models have become much smaller | Epoch AI
In this Gradient Updates weekly issue, Ege discusses how frontier language models have unexpectedly reversed course on scaling, with current models an order of...
language modelshave becomeepoch aifrontiermuch
https://epoch.ai/about
About Us | Epoch AI
Epoch AI is a multidisciplinary research institute investigating the trajectory of AI and forecasting its economic and societal impact.
epoch aius
https://epoch.ai/about/donate
Donate | Epoch AI
Support our mission to improve the shared understanding of AI. Your contribution helps fund independent research that informs better policy and decision-making.
epoch aidonate
https://epoch.ai/gradient-updates/the-real-reason-ai-benchmarks-havent-reflected-economic-impacts
The real reason AI benchmarks haven’t reflected economic impacts | Epoch AI
The real reason that AI benchmarks haven’t reflected real-world impacts historically is that they weren’t optimized for this, not because of fundamental...
the real reasonai benchmarkseconomic impactsreflectedepoch
https://epoch.ai/topics/scaling
AI Scaling: Data & Research | Epoch AI
The story of AI progress is dominated by scale. Training AI systems with more compute, power and data has consistently led to better performance. Epoch tracks...
ai scalingdata researchepoch