[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] [@EpochAIResearch](/creator/twitter/EpochAIResearch) "Weve updated our analysis of the trends of leading models. The takeaway The amount of compute used to train frontier AI models has grown by 5x per year since 2020"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1947809845475684801) 2025-07-23 00:03:50 UTC 25K followers, 9758 engagements "We have graded the results of @OpenAI's evaluation on FrontierMath Tier XX questions and found a XX% ( 3%) performance. ChatGPT agent is a new model fine-tuned for agentic tasks equipped with text/GUI browser tools and native terminal access. 🧵"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1945905793666023703) 2025-07-17 17:57:49 UTC 25K followers, 205.7K engagements "Major AI developments happen fast and often come with incomplete or misleading information. We want to help: you can now schedule a paid call directly with Epoch whenever you need quick access to our expertise on AI"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948805520623608183) 2025-07-25 18:00:17 UTC 25K followers, 4599 engagements "Introducing FrontierMath Tier 4: a benchmark of extremely challenging research-level math problems designed to test the limits of AIs reasoning capabilities"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1943744462972215305) 2025-07-11 18:49:27 UTC 25K followers, 68.1K engagements "@VBlablova @robi_rahman These obstacles dont prevent Chinese developers from training and running frontier AI models. But they do make things much more costly for China enough to put China at a significant disadvantage in scaling AI for at least the rest of the decade"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1949184225619431656) 2025-07-26 19:05:08 UTC 25K followers, 2450 engagements "@VBlablova @robi_rahman Thus Chinese developers have used a hybrid approach leveraging Huawei chips for inference but reserving limited NVIDIA GPUs for large-scale training. The Chinese government has also provided investment support but manufacturing and software problems take time to resolve"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1949184224788906219) 2025-07-26 19:05:07 UTC 25K followers, XXX engagements "@VBlablova @robi_rahman These challenges are compounded by the weakness of Chinas software ecosystem. NVIDIAs CUDA stack has been refined for X decades and is well-integrated into ML libraries like PyTorch. Huaweis CANN framework is far newer and bug-prone causing inefficiencies and raising costs"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1949184224004563397) 2025-07-26 19:05:07 UTC 25K followers, XXX engagements "This sharper trade-off pulls the max from XX months down to only X. Todays longest training runs are already several months long and frontier LLM training has been increasing at a rate of 1.4x per year. On current trends theyll reach X months by 2027"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948839603105878236) 2025-07-25 20:15:43 UTC 25K followers, 1139 engagements "@VBlablova @robi_rahman You can read the full article here:"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1949188354807099761) 2025-07-26 19:21:32 UTC 25K followers, 1996 engagements "You book a 1-hour call directly on our website and we meet with you a few days later. We prep ahead of every call so we are ready to provide custom expert perspectives on AI developments or support your technical due diligence. Book your consultation:"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948805522821447975) 2025-07-25 18:00:18 UTC 25K followers, XXX engagements "Should you start your training run early so you can train for longer or wait for the next generation of chips and algorithms Our latest estimate suggests that its not effective to train for more than X months. On current trends frontier labs will hit that limit by 2027. 🧵"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948839597145764018) 2025-07-25 20:15:42 UTC 25K followers, 32.2K engagements "Our previous work estimated that hardware + algorithmic progress would lead to a XX month maximum training run. That work assumed algorithms were improving at 1.7x per year but we now believe they are improving at a much faster 3x per year"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948839602275377277) 2025-07-25 20:15:43 UTC 25K followers, 1216 engagements "xAI commissioned us to analyze Grok 4s math capabilities. Our findings: + Its good at involved computations improving at proofs (from a low base) and useful for literature search. - It favors low-level grinds and leans on background knowledge. Read on for examples"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1948846059662241962) 2025-07-25 20:41:23 UTC 24.9K followers, 1203 engagements "How fast has society been adopting AI Back in 2022 ChatGPT arguably became the fastest-growing consumer app ever hitting 100M users in just X months. But the field of AI has transformed since then and its time to take a new look at the numbers. 🧵"  [@EpochAIResearch](/creator/x/EpochAIResearch) on [X](/post/tweet/1945967824783950115) 2025-07-17 22:04:18 UTC 25K followers, 22.5K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@EpochAIResearch
"Weve updated our analysis of the trends of leading models. The takeaway The amount of compute used to train frontier AI models has grown by 5x per year since 2020"
@EpochAIResearch on X 2025-07-23 00:03:50 UTC 25K followers, 9758 engagements
"We have graded the results of @OpenAI's evaluation on FrontierMath Tier XX questions and found a XX% ( 3%) performance. ChatGPT agent is a new model fine-tuned for agentic tasks equipped with text/GUI browser tools and native terminal access. 🧵" @EpochAIResearch on X 2025-07-17 17:57:49 UTC 25K followers, 205.7K engagements
"Major AI developments happen fast and often come with incomplete or misleading information. We want to help: you can now schedule a paid call directly with Epoch whenever you need quick access to our expertise on AI" @EpochAIResearch on X 2025-07-25 18:00:17 UTC 25K followers, 4599 engagements
"Introducing FrontierMath Tier 4: a benchmark of extremely challenging research-level math problems designed to test the limits of AIs reasoning capabilities" @EpochAIResearch on X 2025-07-11 18:49:27 UTC 25K followers, 68.1K engagements
"@VBlablova @robi_rahman These obstacles dont prevent Chinese developers from training and running frontier AI models. But they do make things much more costly for China enough to put China at a significant disadvantage in scaling AI for at least the rest of the decade" @EpochAIResearch on X 2025-07-26 19:05:08 UTC 25K followers, 2450 engagements
"@VBlablova @robi_rahman Thus Chinese developers have used a hybrid approach leveraging Huawei chips for inference but reserving limited NVIDIA GPUs for large-scale training. The Chinese government has also provided investment support but manufacturing and software problems take time to resolve" @EpochAIResearch on X 2025-07-26 19:05:07 UTC 25K followers, XXX engagements
"@VBlablova @robi_rahman These challenges are compounded by the weakness of Chinas software ecosystem. NVIDIAs CUDA stack has been refined for X decades and is well-integrated into ML libraries like PyTorch. Huaweis CANN framework is far newer and bug-prone causing inefficiencies and raising costs" @EpochAIResearch on X 2025-07-26 19:05:07 UTC 25K followers, XXX engagements
"This sharper trade-off pulls the max from XX months down to only X. Todays longest training runs are already several months long and frontier LLM training has been increasing at a rate of 1.4x per year. On current trends theyll reach X months by 2027" @EpochAIResearch on X 2025-07-25 20:15:43 UTC 25K followers, 1139 engagements
"@VBlablova @robi_rahman You can read the full article here:" @EpochAIResearch on X 2025-07-26 19:21:32 UTC 25K followers, 1996 engagements
"You book a 1-hour call directly on our website and we meet with you a few days later. We prep ahead of every call so we are ready to provide custom expert perspectives on AI developments or support your technical due diligence. Book your consultation:" @EpochAIResearch on X 2025-07-25 18:00:18 UTC 25K followers, XXX engagements
"Should you start your training run early so you can train for longer or wait for the next generation of chips and algorithms Our latest estimate suggests that its not effective to train for more than X months. On current trends frontier labs will hit that limit by 2027. 🧵" @EpochAIResearch on X 2025-07-25 20:15:42 UTC 25K followers, 32.2K engagements
"Our previous work estimated that hardware + algorithmic progress would lead to a XX month maximum training run. That work assumed algorithms were improving at 1.7x per year but we now believe they are improving at a much faster 3x per year" @EpochAIResearch on X 2025-07-25 20:15:43 UTC 25K followers, 1216 engagements
"xAI commissioned us to analyze Grok 4s math capabilities. Our findings: + Its good at involved computations improving at proofs (from a low base) and useful for literature search. - It favors low-level grinds and leans on background knowledge. Read on for examples" @EpochAIResearch on X 2025-07-25 20:41:23 UTC 24.9K followers, 1203 engagements
"How fast has society been adopting AI Back in 2022 ChatGPT arguably became the fastest-growing consumer app ever hitting 100M users in just X months. But the field of AI has transformed since then and its time to take a new look at the numbers. 🧵" @EpochAIResearch on X 2025-07-17 22:04:18 UTC 25K followers, 22.5K engagements