Post by @urlyman@mastodon.social

One of the discussion points in the conversation with Zeke was AI:

- how bad is its contribution to global heating?
- to what extent and in what domains can it help?

Zeke describes his habitual use of AI coding tools to be more “productive”. Whenever I see that word these days, I find myself reflecting on the hollowing out of its meaning.

What does it even mean to be ‘productive’ at modes of existence which are ratcheting down the probability of human thriving?

https://mastodon.social/@urlyman/117382285770993865

Jonathan Schofield

@urlyman@mastodon.social 7 hr. ago

Some #climateDiary estimates absorbed while listening to @hausfath

Part 1 of 2:

The 8 biggest recent historic El Niños have tended to have a cyclical global warming effect of ~0.2°C.

This year’s El Niño, which will probably peak late Nov to early Dec, looks to have a global warming effect of ~0.3°C.

The year following each of the 8 has always been substantially hotter than the El Niño year. So there is high confidence that next year will be the hottest year on record

https://overcast.fm/+BTumXqskJc

0 Boosts

0 Likes

Tim Ward ⭐🇪🇺🔶 #FBPE

@TimWardCam@c.im 5 hr. ago

@urlyman I'm sure "productive" is measured in "tickets per week".[#]

So if you commit some AI crap which results in three new bugs, hence three new tickets, and you throw some AI slop at them as well to "fix" them, then you've got 400% "productivity".

[#] OK, so I lied, I'm not sure. There supposedly were, and for all I know still are, places that measured how much work a developer was doing by "tokens burned".

Jonathan Schofield

@urlyman@mastodon.social 6 hr. ago

…In the coming years, I hope we will see more clearly the fundamental impasse between what we are doing and what we *say* that produces, and what it actually produces

https://mastodon.social/@urlyman/113980517150495691

Jonathan Schofield

@urlyman@mastodon.social 5 hr. ago

…Re AI and climate science, Zeke was asked:

“Has anyone used one of the most powerful models, and fed in all of the climate data and had some new inferences? Can AI help us make sense of IPCC-type work?”

Zeke answered with:

“So historically, using AI for physics-based processes has been a little challenging, or even stepping aside from AI for a little bit, using machine learning models for physics-based processes has been challenging because physics likes linear or nonlinear relationships…

Jonathan Schofield

@urlyman@mastodon.social 5 hr. ago

…“It likes relationships that are characterized by a smooth function. Machine learning often forms inferences in ways that are not necessarily linear. And so trying to create physics-based simulations from pure machine learning approaches have historically not always worked out that well.

Now, there are some exceptions to that…

Climate Jenny 2.1

@ClimateJenny@biodiversity.social 5 hr. ago

@urlyman The number of people wasting electricity trying to identify plants is probably too small to measure, but there’s a huge controversy in the iNaturalist world about the organization taking Google money. It’s probably just greenwashing, but it’s alarming.

Climate Jenny 2.1

@ClimateJenny@biodiversity.social 5 hr. ago

@urlyman Sigh. There ARE some places where machine learning produces useful results. For example, the plant identification apps can pick up key details from photos *really well* and use that to distinguish species. Even that has limits, though. Many of the more abundant species can’t be told apart by just a photo. False positives mislead beginners all the time.

This technology is now available at the grocery store self-checkout to identify your bananas without you needing to remember the code.

1+ more replies

Jonathan Schofield

@urlyman@mastodon.social 5 hr. ago

…“AI-based weather models are actually quite phenomenal, but that’s largely a sort of pattern and state inference based on a large training data of previous weather models.

We don’t have a huge set of experiments of the planet as a whole … that we can feed into AI models to learn from in the same way…

1+ more replies

Ken Milmore

@kbm0@mastodon.social 7 hr. ago

@urlyman If we roll back from the "AI" branding and think about machine learning as a possible form of mathematical modelling, no doubt there will be models that can apply to climate, water management, logistics and economics and might be used beneficially. Some of these models you might run on a PC, some might require a university-scale computing resource. Probably none of them require global scale data centre rollouts. They won't need LLMs or mass copyright infringement at all.

Jonathan Schofield

@urlyman@mastodon.social 6 hr. ago

@kbm0 agreed. These things have utility when they can be called upon robustly. Their utility changes as that robustness fractures and dissipates, and I see that applying to compute generally

Ken Milmore

@kbm0@mastodon.social 6 hr. ago

@urlyman Trouble is, they (the AI bros) are deliberately keeping it vague. And it frustrates me that hardly any of the journalists or politicians are asking the right questions. The different models, modes and scales of application need to be teased apart from each other, from the crypto grifter hype and from the hard realities of financialisation and its risks to the economy. Not to mention the fact that leaning *into* the climate crisis is most definitely optional.

Jonathan Schofield

@urlyman@mastodon.social 6 hr. ago

@kbm0 I share your frustration. I anticipate it getting worse

𝓼𝓮𝓻𝓪𝓹𝓪𝓽𝓱【ツ】☮(📍🇨🇦)

@serapath@mastodon.gamedev.place 7 hr. ago

@urlyman

i see it more as obviously money printing is justified by AI, so using it to "lower costs" and optimize and get results fast creates a "competitive advantage" and not leveraging it will be punished by customers everywhere.

so either we turn of the money printer and see what sticks ...or we all need to leverage AI eventually and should work towards "democratization" or decentralization of it

Jonathan Schofield

@urlyman@mastodon.social 7 hr. ago

@serapath I don’t see the point in leveraging a technology that cannot stand up in the environment it is helping to create.

I’m already doing that for things I’m habituated to. And those things won’t stand up either. I don’t have any desire to add a ✨new✨ self-terminating “competitive advantage”. But then I’m weary and getting old https://mastodon.social/@urlyman/111867603435972978

Mastodon

Jonathan Schofield (@urlyman@mastodon.social)

Attached: 1 image Helluva book cover 2026 update (thanks to Cameron in the replies): Christopher did have a website for his book at readblip\.com but that’s now been co-opted by some spamming endeavour. For an overview, see more lower down the 2024 thread https://mas.to/@jackofalltrades/111868294339918049

𝓼𝓮𝓻𝓪𝓹𝓪𝓽𝓱【ツ】☮(📍🇨🇦)

@serapath@mastodon.gamedev.place 6 hr. ago

@urlyman

so whats exactly the issue apart from threatening democracy whoch we already dont really have apart from layers and layers of intransparent complicated indirect delegation on top of delegation with always the same establishment coming up with more and more ridiculous laws ... but yeah, having a few companies controlling AI isnt great.

but for the majority of 8 billion ppl it provides them access to knowledge (even if not always perfect) which they had no access to before. empowering

Jonathan Schofield

@urlyman@mastodon.social 6 hr. ago

@serapath I had access to more information than I could ever integrate *before* 2023. And for all that corpus of data’s shortcomings I was generally getting it from a human source.

If you’re convinced that AI is a net positive, I’m not going to even attempt to persuade you otherwise

1+ more replies