Edit history

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

If you want a better comparison, multiply the numbers I gave by the numbers of people in world that have a smart phone. Then compare costs and harms.

One person uses one full phone (assuming they have 1 device; some people have more). The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU cluster, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

That’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

If you want a better comparison, multiply the number I gave by the numbers of people in world that have a smart phone. Then compare costs and harms.

One person uses one full phone (assuming they have 1 device; some people have more). The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU cluster, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

That’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

If you want a better comparison, multiply the number I gave by the numbers of people in world that have a smart phone. Then compare costs and harms.

One person uses one full phone (assuming they have 1 device; some people have more). The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

That’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

If you want a better comparison, multiply the number I gave by the numbers of people in world that have a smart phone. Then compare costs and harms.

One person uses one full phone (assuming they have 1 device; some people have more). The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

If you want a better comparison, multiply the number I gave by the numbers of people in world that have a smart phone. Then compare costs and harms.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison. You’ve cited whole industry numbers against my per capita numbers.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device.

But one person cannot be said to bear the full cost of a GPU, a data centre, or a training run.

Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before associated social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation, proliferation of gig economy…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone (at all) without incurring its full lifecycle cost, unless you’re assuming 1 phone shared amongst n individuals.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on (eg: adolescent mental health, attention economy design, surveillance capitalism, algorithmic radicalisation…)

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

And that’s before social harms enter the ledger, which the phone industry has a considerable head start on.

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation, social harms and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generation, and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.

That’s not a 1:1 comparison.

One person uses one phone. The embodied cost - energy, materials, waste - maps directly to that one device. You can’t use a phone without incurring its full lifecycle cost.

But one person doesn’t bear the full cost of a GPU, a data centre, or a training run. Those costs are distributed across every user of that system. If we’re arguing per-capita impact, the LLM side of the ledger needs to be divided by its actual user base, not compared whole against a single handset.

You’d need to compare either one phone to one person’s proportional share of LLM infrastructure, or the entire phone industry against the entire LLM industry.

And given rare earth extraction, cobalt supply chains, labour abuse, e-waste generated, and global market scale, that industry-level comparison probably doesn’t favour phones as cleanly as you’d expect.