Latest news with #EpochAI
Yahoo
13-05-2025
- Business
- Yahoo
Improvements in 'reasoning' AI models may slow down soon, analysis finds
An analysis by Epoch AI, a nonprofit AI research institute, suggests the AI industry may not be able to eke massive performance gains out of reasoning AI models for much longer. As soon as within a year, progress from reasoning models could slow down, according to the report's findings. Reasoning models such as OpenAI's o3 have led to substantial gains on AI benchmarks in recent months, particularly benchmarks measuring math and programming skills. The models can apply more computing to problems, which can improve their performance, with the downside being that they take longer than conventional models to complete tasks. Reasoning models are developed by first training a conventional model on a massive amount of data, then applying a technique called reinforcement learning, which effectively gives the model "feedback" on its solutions to difficult problems. So far, frontier AI labs like OpenAI haven't applied an enormous amount of computing power to the reinforcement learning stage of reasoning model training, according to Epoch. That's changing. OpenAI has said that it applied around 10x more computing to train o3 than its predecessor, o1, and Epoch speculates that most of this computing was devoted to reinforcement learning. And OpenAI researcher Dan Roberts recently revealed that the company's future plans call for prioritizing reinforcement learning to use far more computing power, even more than for the initial model training. But there's still an upper bound to how much computing can be applied to reinforcement learning, per Epoch. Josh You, an analyst at Epoch and the author of the analysis, explains that performance gains from standard AI model training are currently quadrupling every year, while performance gains from reinforcement learning are growing tenfold every 3-5 months. The progress of reasoning training will "probably converge with the overall frontier by 2026," he continues. Epoch's analysis makes a number of assumptions, and draws in part on public comments from AI company executives. But it also makes the case that scaling reasoning models may prove to be challenging for reasons besides computing, including high overhead costs for research. "If there's a persistent overhead cost required for research, reasoning models might not scale as far as expected," writes You. "Rapid compute scaling is potentially a very important ingredient in reasoning model progress, so it's worth tracking this closely." Any indication that reasoning models may reach some sort of limit in the near future is likely to worry the AI industry, which has invested enormous resources developing these types of models. Already, studies have shown that reasoning models, which can be incredibly expensive to run, have serious flaws, like a tendency to hallucinate more than certain conventional models. This article originally appeared on TechCrunch at Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
TechCrunch
12-05-2025
- Business
- TechCrunch
Improvements in ‘reasoning' AI models may slow down soon, analysis finds
An analysis by Epoch AI, a nonprofit AI research institute, suggests the AI industry may not be able to eke massive performance gains out of reasoning AI models for much longer. As soon as within a year, progress from reasoning models could slow down, according to the report's findings. Reasoning models such as OpenAI's o3 have led to substantial gains on AI benchmarks in recent months, particularly benchmarks measuring math and programming skills. The models can apply more computing to problems, which can improve their performance, with the downside being that they take longer than conventional models to complete tasks. Reasoning models are developed by first training a conventional model on a massive amount of data, then applying a technique called reinforcement learning, which effectively gives the model 'feedback' on its solutions to difficult problems. So far, frontier AI labs like OpenAI haven't applied an enormous amount of computing power to the reinforcement learning stage of reasoning model training, according to Epoch. That's changing. OpenAI has said that it applied around 10x more computing to train o3 than its predecessor, o1, and Epoch speculates that most of this computing was devoted to reinforcement learning. And OpenAI researcher Dan Roberts recently revealed that the company's future plans call for prioritizing reinforcement learning to use far more computing power, even more than for the initial model training. But there's still an upper bound to how much computing can be applied to reinforcement learning, per Epoch. According to an Epoch AI analysis, reasoning model training scaling may slow down. Image Credits:Epoch AI Josh You, an analyst at Epoch and the author of the analysis, explains that performance gains from standard AI model training are currently quadrupling every year, while performance gains from reinforcement learning are growing tenfold every 3-5 months. The progress of reasoning training will 'probably converge with the overall frontier by 2026,' he continues. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you've built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you've built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | BOOK NOW Epoch's analysis makes a number of assumptions, and draws in part on public comments from AI company executives. But it also makes the case that scaling reasoning models may prove to be challenging for reasons besides computing, including high overhead costs for research. 'If there's a persistent overhead cost required for research, reasoning models might not scale as far as expected,' writes You. 'Rapid compute scaling is potentially a very important ingredient in reasoning model progress, so it's worth tracking this closely.' Any indication that reasoning models may reach some sort of limit in the near future is likely to worry the AI industry, which has invested enormous resources developing these types of models. Already, studies have shown that reasoning models, which can be incredibly expensive to run, have serious flaws, like a tendency to hallucinate more than certain conventional models.
Yahoo
24-04-2025
- Business
- Yahoo
AI supercomputers are getting bigger each year — and one study projects they could need as much power as a city by 2030
The future of AI depends on building more powerful supercomputers. New research suggests how gargantuan these supercomputers could be by the end of the decade. Epoch AI projects that a top supercomputer in 2030 could need as much power needed for up to 9 million homes. Silicon Valley needs bigger and better supercomputers to get bigger and better AI. We just got an idea of what they might look like by the end of the decade. A new study published this week by researchers at the San Francisco-based institute Epoch AI said supercomputers — massive systems stacked with chips to train and run AI models — could need the equivalent of nine nuclear reactors by 2030 to keep them chugging along. Epoch AI estimates that if supercomputer power requirements continue to roughly double each year, as they have done since 2019, the top machines would need about 9GW of power. That's about the amount needed to keep the lights on in a city of roughly 7 to 9 million homes. Today's most powerful supercomputer needs around 300MW of power, which is "equivalent to about 250,000 households." This makes the potential power needs of future supercomputers look extraordinary. There are a few reasons the next generation of computing seems so demanding. One explanation is that they will simply be bigger. According to Epoch AI's paper, the leading AI supercomputer in 2030 could require 2 million AI chips and cost $200 billion to build — again, assuming that current growth trends continue. For context, today's largest supercomputer — the Colossus system, built to full scale within 214 days by Elon Musk's xAI — is estimated to have cost $7 billion to make, and, per the company's website, is stacked with 200,000 chips. Companies have been looking to secure more chips to provide the computing power needed for increasingly powerful models as they race toward developing AI that surpasses human intelligence. OpenAI, for instance, started the year with a huge supercomputer announcement as it unveiled Stargate, a project worth over $500 billion in investment over four years aimed at building critical AI infrastructure that includes a "computing system." Epoch AI explains this growth by stating that where once supercomputers were used just as research tools, they're now being used as "industrial machines delivering economic value." Having AI and supercomputers deliver economic value isn't just a priority for CEOs trying to justify exorbitant capital expenditures, either. Earlier this month, President Donald Trump took to Truth Social to celebrate a $500 billion investment from Nvidia to build AI supercomputers in the US. It's "big and exciting news," he said, branding the announcement a commitment to "the Golden Age of America." However, as Epoch AI's research suggests — research based on a dataset that covers "about 10% of all relevant AI chips produced in 2023 and 2024 and about 15% of the chip stocks of the largest companies at the start of 2025" — this would all come with greater power demands. Epoch AI did note that "AI supercomputers are improving in energy efficiency, but the shift is not quickly enough to offset overall power growth." It also explains why companies like Microsoft, Google, and others have been looking to nuclear power as an alternative to their energy needs. If the AI trend continues to grow, expect supercomputers to keep growing with it. Read the original article on Business Insider Sign in to access your portfolio
Yahoo
24-04-2025
- Business
- Yahoo
AI supercomputers are getting bigger each year — and one study projects they could need as much power as a city by 2030
The future of AI depends on building more powerful supercomputers. New research suggests how gargantuan these supercomputers could be by the end of the decade. Epoch AI projects that a top supercomputer in 2030 could need as much power needed for up to 9 million homes. Silicon Valley needs bigger and better supercomputers to get bigger and better AI. We just got an idea of what they might look like by the end of the decade. A new study published this week by researchers at the San Francisco-based institute Epoch AI said supercomputers — massive systems stacked with chips to train and run AI models — could need the equivalent of nine nuclear reactors by 2030 to keep them chugging along. Epoch AI estimates that if supercomputer power requirements continue to roughly double each year, as they have done since 2019, the top machines would need about 9GW of power. That's about the amount needed to keep the lights on in a city of roughly 7 to 9 million homes. Today's most powerful supercomputer needs around 300MW of power, which is "equivalent to about 250,000 households." This makes the potential power needs of future supercomputers look extraordinary. There are a few reasons the next generation of computing seems so demanding. One explanation is that they will simply be bigger. According to Epoch AI's paper, the leading AI supercomputer in 2030 could require 2 million AI chips and cost $200 billion to build — again, assuming that current growth trends continue. For context, today's largest supercomputer — the Colossus system, built to full scale within 214 days by Elon Musk's xAI — is estimated to have cost $7 billion to make, and, per the company's website, is stacked with 200,000 chips. Companies have been looking to secure more chips to provide the computing power needed for increasingly powerful models as they race toward developing AI that surpasses human intelligence. OpenAI, for instance, started the year with a huge supercomputer announcement as it unveiled Stargate, a project worth over $500 billion in investment over four years aimed at building critical AI infrastructure that includes a "computing system." Epoch AI explains this growth by stating that where once supercomputers were used just as research tools, they're now being used as "industrial machines delivering economic value." Having AI and supercomputers deliver economic value isn't just a priority for CEOs trying to justify exorbitant capital expenditures, either. Earlier this month, President Donald Trump took to Truth Social to celebrate a $500 billion investment from Nvidia to build AI supercomputers in the US. It's "big and exciting news," he said, branding the announcement a commitment to "the Golden Age of America." However, as Epoch AI's research suggests — research based on a dataset that covers "about 10% of all relevant AI chips produced in 2023 and 2024 and about 15% of the chip stocks of the largest companies at the start of 2025" — this would all come with greater power demands. Epoch AI did note that "AI supercomputers are improving in energy efficiency, but the shift is not quickly enough to offset overall power growth." It also explains why companies like Microsoft, Google, and others have been looking to nuclear power as an alternative to their energy needs. If the AI trend continues to grow, expect supercomputers to keep growing with it. Read the original article on Business Insider Sign in to access your portfolio
Business Insider
24-04-2025
- Business
- Business Insider
AI supercomputers could need as much power as a city by 2030
Silicon Valley needs bigger and better supercomputers to get bigger and better AI. We just got an idea of what they might look like by the end of the decade. A new study published this week by researchers at the San Francisco-based institute Epoch AI said supercomputers — massive systems stacked with chips to train and run AI models — could need the equivalent of nine nuclear reactors by 2030 to keep them chugging along. Epoch AI estimates that if supercomputer power requirements continue to roughly double each year, as they have done since 2019, the top machines would need about 9GW of power. That's about the amount needed to keep the lights on in a city of roughly 7 to 9 million homes. Today's most powerful supercomputer needs around 300MW of power, which is "equivalent to about 250,000 households." This makes the potential power needs of future supercomputers look extraordinary. There are a few reasons the next generation of computing seems so demanding. One explanation is that they will simply be bigger. According to Epoch AI's paper, the leading AI supercomputer in 2030 could require 2 million AI chips and cost $200 billion to build — again, assuming that current growth trends continue. For context, today's largest supercomputer — the Colossus system, built to full scale within 214 days by Elon Musk's xAI — is estimated to have cost $7 billion to make, and, per the company's website, is stacked with 200,000 chips. As they grew in performance, AI supercomputers got exponentially more expensive. The upfront hardware cost of leading AI supercomputers doubled roughly every year (1.9x/year). We estimate the hardware for xAI's Colossus cost about $7 billion. — Epoch AI (@EpochAIResearch) April 23, 2025 Companies have been looking to secure more chips to provide the computing power needed for increasingly powerful models as they race toward developing AI that surpasses human intelligence. OpenAI, for instance, started the year with a huge supercomputer announcement as it unveiled Stargate, a project worth over $500 billion in investment over four years aimed at building critical AI infrastructure that includes a "computing system." Epoch AI explains this growth by stating that where once supercomputers were used just as research tools, they're now being used as "industrial machines delivering economic value." Having AI and supercomputers deliver economic value isn't just a priority for CEOs trying to justify exorbitant capital expenditures, either. Earlier this month, President Donald Trump took to Truth Social to celebrate a $500 billion investment from Nvidia to build AI supercomputers in the US. It's "big and exciting news," he said, branding the announcement a commitment to "the Golden Age of America." However, as Epoch AI's research suggests — research based on a dataset that covers "about 10% of all relevant AI chips produced in 2023 and 2024 and about 15% of the chip stocks of the largest companies at the start of 2025" — this would all come with greater power demands. Epoch AI did note that "AI supercomputers are improving in energy efficiency, but the shift is not quickly enough to offset overall power growth." It also explains why companies like Microsoft, Google, and others have been looking to nuclear power as an alternative to their energy needs. If the AI trend continues to grow, expect supercomputers to keep growing with it.
