OpenAI on Tuesday rolled out its o3-Professional mannequin for ChatGPT Professional and Groups subscribers, slashed o3 pricing by 80 p.c, and dropped a weblog put up from CEO Sam Altman teasing “intelligence too low-cost to meter.”
“The typical question makes use of about 0.34 watt-hours, about what an oven would use in a bit over one second, or a high-efficiency gentle bulb would use in a few minutes. It additionally makes use of about 0.000085 gallons of water; roughly one fifteenth of a teaspoon,” Altman stated in a put up.
That is in step with prior outdoors estimates. Epoch AI printed the same determine in February. The agency stated, “a GPT-4o question consumes round 0.3 watt-hours for a typical text-based query, although this will increase considerably to 2.5 to 40 watt-hours for queries with very lengthy inputs.”
However taking a look at AI vitality utilization on a mean question foundation grossly oversimplifies issues concerning the expertise’s environmental influence, given the huge variety of queries customers are getting into – over a billion a day as of final December, in keeping with the corporate.
When the MIT Know-how Evaluate explored AI vitality utilization just lately, the conclusion didn’t align with Altman’s declare that “Intelligence too low-cost to meter is nicely inside grasp.” Quite, the publication cited analysis from the Lawrence Berkeley Nationwide Laboratory estimating AI-specific functions in knowledge facilities will eat between 165 and 326 terawatt-hours of vitality in 2028 – sufficient to energy 22% of all US households.
OpenAI’s o3 mannequin is not too low-cost to meter, however attributable to an optimized inference stack, it is 80 p.c inexpensive than it was once: Enter: $2 per 1M tokens; Output: $8 per 1M tokens. However there are nonetheless many cheaper fashions.
General, Altman’s musings skew towards techno-optimism – shock! He posits a flood of wondrous discoveries a decade therefore arising from AI superintelligence, no matter that’s.
“Perhaps we’ll go from fixing high-energy physics one 12 months to starting house colonization the subsequent 12 months; or from a significant supplies science breakthrough one 12 months to true high-bandwidth brain-computer interfaces the subsequent 12 months,” he stated.
Perhaps. Or perhaps not. We word that fellow futurist Elon Musk, who as soon as cautioned about releasing the AI demon, predicted in 2016 that people would land on Mars by 2025. Tech leaders merely pay no price for misprediction.
However Altman has extra ideas to share.
Within the 2030s, intelligence and vitality – concepts, and the flexibility to make concepts occur – are going to grow to be wildly plentiful
“Within the 2030s, intelligence and vitality – concepts, and the flexibility to make concepts occur – are going to grow to be wildly plentiful,” he opined. “These two have been the basic limiters on human progress for a very long time; with plentiful intelligence and vitality (and good governance), we will theoretically have the rest.”
Altman’s put up additionally exemplifies the slap-and-kiss that has characterised current AI evangelism, citing dangers however insisting all will likely be nicely ultimately.
“There are critical challenges to confront together with the large upsides,” Altman wrote. “We do want to resolve the security points, technically and societally, however then it’s critically essential to broadly distribute entry to superintelligence given the financial implications.”
Gary Marcus, an AI knowledgeable, writer, and critic, took the put up as a chance to match Altman to discredited Theranos CEO Elizabeth Holmes, now serving time for fraud.
Had Altman consulted his personal firm’s ChatGPT on the basic limiters of human progress, he’d have been introduced with a much more intensive record that features: Cognitive and Psychological Constraints, Sociopolitical Methods, Financial and Useful resource Constraints, Technological and Scientific Limits, Ecological and Environmental Boundaries, Cultural and Moral Constraints, and Temporal and Bodily Limits. And every of those classes comes with a number of bullet factors.
But the most important information to emerge amid Altman’s pollyannaish prognostication could also be that OpenAI, nurtured by billions from Microsoft, reportedly plans to increase mannequin availability via a partnership with Google Cloud. And it was solely a 12 months in the past that Microsoft described longtime companion OpenAI as a competitor.
OpenAI didn’t instantly reply to a request for remark. ®
OpenAI on Tuesday rolled out its o3-Professional mannequin for ChatGPT Professional and Groups subscribers, slashed o3 pricing by 80 p.c, and dropped a weblog put up from CEO Sam Altman teasing “intelligence too low-cost to meter.”
“The typical question makes use of about 0.34 watt-hours, about what an oven would use in a bit over one second, or a high-efficiency gentle bulb would use in a few minutes. It additionally makes use of about 0.000085 gallons of water; roughly one fifteenth of a teaspoon,” Altman stated in a put up.
That is in step with prior outdoors estimates. Epoch AI printed the same determine in February. The agency stated, “a GPT-4o question consumes round 0.3 watt-hours for a typical text-based query, although this will increase considerably to 2.5 to 40 watt-hours for queries with very lengthy inputs.”
However taking a look at AI vitality utilization on a mean question foundation grossly oversimplifies issues concerning the expertise’s environmental influence, given the huge variety of queries customers are getting into – over a billion a day as of final December, in keeping with the corporate.
When the MIT Know-how Evaluate explored AI vitality utilization just lately, the conclusion didn’t align with Altman’s declare that “Intelligence too low-cost to meter is nicely inside grasp.” Quite, the publication cited analysis from the Lawrence Berkeley Nationwide Laboratory estimating AI-specific functions in knowledge facilities will eat between 165 and 326 terawatt-hours of vitality in 2028 – sufficient to energy 22% of all US households.
OpenAI’s o3 mannequin is not too low-cost to meter, however attributable to an optimized inference stack, it is 80 p.c inexpensive than it was once: Enter: $2 per 1M tokens; Output: $8 per 1M tokens. However there are nonetheless many cheaper fashions.
General, Altman’s musings skew towards techno-optimism – shock! He posits a flood of wondrous discoveries a decade therefore arising from AI superintelligence, no matter that’s.
“Perhaps we’ll go from fixing high-energy physics one 12 months to starting house colonization the subsequent 12 months; or from a significant supplies science breakthrough one 12 months to true high-bandwidth brain-computer interfaces the subsequent 12 months,” he stated.
Perhaps. Or perhaps not. We word that fellow futurist Elon Musk, who as soon as cautioned about releasing the AI demon, predicted in 2016 that people would land on Mars by 2025. Tech leaders merely pay no price for misprediction.
However Altman has extra ideas to share.
Within the 2030s, intelligence and vitality – concepts, and the flexibility to make concepts occur – are going to grow to be wildly plentiful
“Within the 2030s, intelligence and vitality – concepts, and the flexibility to make concepts occur – are going to grow to be wildly plentiful,” he opined. “These two have been the basic limiters on human progress for a very long time; with plentiful intelligence and vitality (and good governance), we will theoretically have the rest.”
Altman’s put up additionally exemplifies the slap-and-kiss that has characterised current AI evangelism, citing dangers however insisting all will likely be nicely ultimately.
“There are critical challenges to confront together with the large upsides,” Altman wrote. “We do want to resolve the security points, technically and societally, however then it’s critically essential to broadly distribute entry to superintelligence given the financial implications.”
Gary Marcus, an AI knowledgeable, writer, and critic, took the put up as a chance to match Altman to discredited Theranos CEO Elizabeth Holmes, now serving time for fraud.
Had Altman consulted his personal firm’s ChatGPT on the basic limiters of human progress, he’d have been introduced with a much more intensive record that features: Cognitive and Psychological Constraints, Sociopolitical Methods, Financial and Useful resource Constraints, Technological and Scientific Limits, Ecological and Environmental Boundaries, Cultural and Moral Constraints, and Temporal and Bodily Limits. And every of those classes comes with a number of bullet factors.
But the most important information to emerge amid Altman’s pollyannaish prognostication could also be that OpenAI, nurtured by billions from Microsoft, reportedly plans to increase mannequin availability via a partnership with Google Cloud. And it was solely a 12 months in the past that Microsoft described longtime companion OpenAI as a competitor.
OpenAI didn’t instantly reply to a request for remark. ®
