Post

An early argument for support was solving big problems for the good, e.g. cancer cure. Bespoke models are more efficacious for many of these and current capabilities seem adequate to speed less daunting problems. That moves the constraint to the physical testing etc. Are these assertions true?

could we advance LLM technology at 1/100th the scale, committing a fraction of the externalities if the firms didn't have to race to have the best model and win a monopoly share on the market?

btw - If we had programmers that refused to use (or code/test) error handling functions, they wouldn't have a job because we wouldn't have customers/employers. #AI is code!

Like all high capex businesses in history early builds always lead to a collapse & consolidation. We can be absolutely certain that much of we build will become stranded assets. China is ensuring that their clean energy & grid assets will maintain their value. We could do the same but aren’t. Why?

How can we break the oiligopoly?

Q2 - the downtown Vancouver data centre proposal would decarbonize the [nat gas based] steam system. A rare symbiosis. Are any US district energy systems (NYC?) even big enough to absorb all of a hyperscale [GW] data centre's waste heat at midwinter peak? resourceworks.com/data-centres...

Has everyone forgotten why we ditched mainframes for personal computers in the 1980s? I think of data centers as the new mainframes. So the question is: do you really need giant data centers for this? Or is just a way for the VC world to suck more money out of everybody?

Is there significant evidence that data center investing and associated energy investments are crowding out investments in renewable energy?

Is there a single built one that hasn't been a net negative? (Also a question I'd submit for your guests with more-optimistic takes on what data centers can do for the grid).

@ketanjoshi.co do you know of any politicians anywhere in the world who have made an effort to increase their tech literacy to really understand their decisions? Do their staff know and actually advise them? Or do they drink any Kool-Aid with the words ‘increased productivity’ on the label?

my question is maybe dumb but why aren’t more data centers run on clean energy. seems they would be less friction for them is they used more solar wind and batteries

Indeed :) Needs strong govt regulation I’d suggest. (Currently none)

If all data centers were required to build on-the-grid additional clean energy and storage (or VPP) to match their 24/7/365 hourly load and pay for grid upgrades before interconnection permit (and behind the meter BYOP was illegal), would the data center boom be a good thing for climate progress?

Ask him if he understands the paradoxical moral vs material survival drives of the human brain. When people understand how cognition works enough to see the global-emotional system, the profiteer simulator brains (AI) trigger moral disgust because of their material harms. That's moral clarity

How much of any benefit from AI is flowing to women? How much of the downside?

From someone in WI, can you cover water impacts? Ex: water consumption, water pollution, impact on Great Lakes Even if Data Center energy is fully clean, the water stuff worries me

There is some research ongoing about emissions abatement, using reclaimed/oilfield produced water for cooling, etc. How much of a difference can we realistically make on emissions and water use reduction vs nibbling around the margins?

Is it only ai datacenters we are concerned about? How about crypto datacenters? Or cloud hosting datacenters? Or networking datacenters?

Noah Smith: "...some sort of restriction on data centers. In other words, the economic danger of AI isn’t really that it’ll take all our jobs; the danger is that it’ll gobble up all the land and energy, leaving too little for human use." www.noahpinion.blog/p/plentiful-...

Do we have the technology available to have data centers that don’t emit GHGs, don’t pollute air & water, don’t add heat & noise to neighboring communities. i.e. can build them with no adverse environmental health impacts

What's their plan after they've covered the world in data centers? Are you guys going to get into network states and freedom cities at all? Are you familiar with Ann Pettifor author of The Global Casino? Are you going to talk about credit/debt at all?

Is there any reason not to laugh anyone who spouts eventual geo engineering as an eventual pie in the sky climate solution out of the room, or at least avoid giving them any credibility with a platform for their dangerous, denialist, delays to excuse burning fossil fuels for AI?

Would he reconsider his stance on AI if it was reprogrammed to create armies of self-replicating robots that systematically stole everyone’s ICE vehicle tire rims and melted them down to make new pv module frames and server racks?

Tangential, but... talking point was used a lot that AI would solve climate change. Is there anything there? Technology or process that AI could conceivably help develop that would have an impact? Like drug development where there is at least real potential. Or is that all just full on BS?

How will companies or governments handle the increase of e-waste generated by AI data centers? I’d love some discussion of environmental aspects that goes beyond energy/water and into the materials used/discarded due to the build out.

Does anyone know who all the new data center capacity is supposed to serve? Commercial / research / military ?

Picture the morning the API key dies. The team that still has eval fixtures, prompt files, and a local runner in git can keep working that afternoon. The team that only had the SaaS tab is stuck. That folder is usually what is worth open-sourcing.

/is it even possible?

If these companies are hemorrhaging money investing in the infrastructure, how can/why are they investing in so much political spending (which I believe @kevinjkircher.com posted)?

might actually be helping establish the playbook for opposition to the many other components required to "electrify everything" that will also require a lot of power, land, grid connection, heat dissipation etc. [and, separately, I wrote a separate multi-post suggestion for the interview, which...

... maybe Anthony is not being sufficiently hateful enough for her taste, and Anthony going on to clarify as below👇. And that Anthony is demanding that of others doesn't mean that's necessarily an obligation of Ketan to do this, but I would just point you to the podcast... bsky.app/profile/anth...

I assume you'll cover what we're talking about when we say "data center" & why they're being built & why these locations. Are they really needed at all to do what they're supposed to do, is there a way to do that w minimal impact? Comparable to ordinary light industry. Or is heavy industry closer?

Maybe some perspective on the case for datacenters in Europe, such as the Google-case in Finland. Why do tech companies even bother trying to build there, with all their governance? And also; is there truly a way in which (ai) datacenters can be net positive for net stability when using smr's?

As someone who worked in government to resolve 2 banking crises, I’d like to know if this bubble is repeating the mistakes of the past - that being financial institutions loaning money for things they don’t understand? Or is this mostly a kind of check kitting scheme - loaning money to each other?

Q: Some AI (eg for research) can make sense. So some data centres capable of AI processing can make sense. But how to *practically* prioritise those over AI-slop data centres? Through public ownership of the former?

@volts.wtf @ketanjoshi.co is: why don't we tax the shit out of them? an acceptable question?

BAAJA lame duck?

One scenario of how the data center gold rush ends (or winds down) is with one or more major companies going under. That way lies risks of recession, bailouts, etc. Another scenario involves democratic decision-making that manages the scale/pace/locations of buildout. What might that look like?

No steel in the ground yet, but the power purchase agreement with Google's 1 GW data center in Michigan has them paying for at least enough renewable and batteries to serve the facility. Likely they'll be subsiding some renewable grid power. (Depends on their actual usage.)

For comparison, how many permanent jobs would $30 to 60 billion worth of factories making batteries, solar panels, or EVs bring to a community.

50-100

One of the issues with AI data centers is apparently very rapidly fluctuating load. This seems like a readily solvable issue that is likely only a problem because the techbros can’t be arsed to do anything that slows things down one iota, but I dunno. I’d be interested to hear his perspective.

Related: Some people say the current surge in electricity demand growth comes only partly from data centers, and also comes from electric vehicles, heat pumps, induction stoves, and so on. How much of the current surge in electricity demand growth comes from those other things vs data centers?

What fraction of growth in pollutiom from US natural gas power plants is attributable to AI data centers?

Good point.

Many energy people who used to describe themselves as climate activists or renewable champions have chased the AI gold rush and are now getting rich off of the data center boom, pollution be damned. Should anyone take these people seriously ever again when they talk about clean energy or climate?

No.

(Note all the off-ramps here for a switch from on-site gas turbines to grid power not actually being helpful. Not switching to grid power because gas is faster or manlier or whatever; just buying vanilla grid power, however coaly or gassy it might be; non-additional PPAs; etc.)

Follow-up: What could or should policymakers be doing now to prepare for a possible collapse of a massive AI company? In particular, what alternatives are there to using taxpayer money to bail out "too big to fail" AI companies, chip makers, or other firms entangled in the web of AI finance?

FYI, nice discussion here on progress (and barriers) toward building truly open LLMs. www.argmin.net/p/the-world-...

Yes. The enabled emissions campaign is probably best known for this work of late - they had a paper come out recently https://www.enabledemissions.com/

but if it works it will all be worth it - an authoritarian surveillance state powered by good ol 'muricun oil & slave labor.

software under the same umbrella "AI" term as LLMs affects how people think and feel about AI, especially its potential to do good stuff for climate?