3 ms·

Could you explain in detail why the power consumption calculations were mistaken? It sounds like you're saying that the calculations weren't mistaken, but that

Could you explain in detail why the power consumption calculations were mistaken? It sounds like you're saying that the calculations weren't mistaken, but that she didn't mention the carbon neutral status of Google as a whole.

random314 5y ago
Yes, I never said that the power consumption calculations were wrong. The paper wasn't about power consumption, but CO2 emissions leading to global warming. She compared the power consumption of model training to jet fuel exhaust and used jet fuel exhaust to tie the power consumption to CO2 emissions. Of course, she didn't mention that Google is carbon neutral and its data centers are primarily powered by green energy. I have not seen a more dishonest paper in CS. Electricity from the grid can come from green sources, and in the particular case if Google, it actually does. Comparing electricity consumption for a one time training with jet fuel usage is absurd. Moreover, the paper doesn't mention any possible benefit of building such nlp models. Increased accessibility of web content across languages and for disabled people. Google, simply through its search, maps and navigation saves a lot of wasted carbon. Complaining about the power consumption of ML models is absolutely ridiculous. Her paper reeked of desperation to find something, anything within the AI space that could be posted as an ethical dilemma. Reading that paper convinced me that Timnit could bring absolutely no value to AI as a field, in terms of ethics or otherwise.
random314 5y ago
Consider this paragraph "We first consider environmental risks. Echoing a line of recent work outlining the environmental and financial costs of deep learn- ing systems [129], we encourage the research community to priori- tize these impacts. One way this can be done is by reporting costs and evaluating works based on the amount of resources they con- sume [57]. As we outline in §3, increasing the environmental and financial costs of these models doubly punishes marginalized com- munities that are least likely to benefit from the progress achieved by large LMs and most likely to be harmed by negative environ- mental consequences of its resource consumption. At the scale we are discussing (outlined in §2), the first consideration should be the environmental cost." This whole paragraph is such a reach, and so absurd that it seems like a parody. Are we supposed to believe that marginalized communities are being harmed by Google training NLP models using green energy? Or will they benefit from better translation from English language content into marginalized languages? Timnit has zero interest in what actually benefits marginalized communities. She uses them as a prop to build a career in high paying AI field where, apparently, the sub field of ai fairness has a low bar for publication.

random314 5y ago

Yes, I never said that the power consumption calculations were wrong. The paper wasn't about power consumption, but CO2 emissions leading to global warming. She compared the power consumption of model training to jet fuel exhaust and used jet fuel exhaust to tie the power consumption to CO2 emissions. Of course, she didn't mention that Google is carbon neutral and its data centers are primarily powered by green energy. I have not seen a more dishonest paper in CS. Electricity from the grid can come from green sources, and in the particular case if Google, it actually does. Comparing electricity consumption for a one time training with jet fuel usage is absurd. Moreover, the paper doesn't mention any possible benefit of building such nlp models. Increased accessibility of web content across languages and for disabled people. Google, simply through its search, maps and navigation saves a lot of wasted carbon. Complaining about the power consumption of ML models is absolutely ridiculous. Her paper reeked of desperation to find something, anything within the AI space that could be posted as an ethical dilemma. Reading that paper convinced me that Timnit could bring absolutely no value to AI as a field, in terms of ethics or otherwise.

random314 5y ago

Consider this paragraph "We first consider environmental risks. Echoing a line of recent work outlining the environmental and financial costs of deep learn- ing systems [129], we encourage the research community to priori- tize these impacts. One way this can be done is by reporting costs and evaluating works based on the amount of resources they con- sume [57]. As we outline in §3, increasing the environmental and financial costs of these models doubly punishes marginalized com- munities that are least likely to benefit from the progress achieved by large LMs and most likely to be harmed by negative environ- mental consequences of its resource consumption. At the scale we are discussing (outlined in §2), the first consideration should be the environmental cost." This whole paragraph is such a reach, and so absurd that it seems like a parody. Are we supposed to believe that marginalized communities are being harmed by Google training NLP models using green energy? Or will they benefit from better translation from English language content into marginalized languages? Timnit has zero interest in what actually benefits marginalized communities. She uses them as a prop to build a career in high paying AI field where, apparently, the sub field of ai fairness has a low bar for publication.