Links

Three findings from this week's UN report on AI vs climate crisis:

🌿 Majority of AI energy use (80-90%) is not in training models, but in day-to-day use.

🌿 Generating an image is 1450⨉ more energy intensive than text, generating video much worse.

🌿 We need to monitor water and land footprint, as well as carbon.

(From: Rising Emissions, Depleting Water and Vanishing Land—UN Scientists: AI Is Threatening Natural Resources for Billions | United Nations University)

The WebKit team responded to WebMCP in standards-positions.

What stood out to me:

When a site's actions are hard for an agent to use, that is a gap in the page's own semantics, and the fix, in our opinion, is to close it in the platform's shared layers (HTML and ARIA), where the user, assistive technology, and agents all benefit. Describing those actions a second time as JavaScript tools would not deliver the reliability it promises: the agent still selects and interprets a tool from its natural-language name and description, which the specification itself concedes are ambiguous and unverifiable, with "no guarantee that a WebMCP tool's declared intent matches its actual behavior."

(From: WebMCP · Issue #670 · WebKit/standards-positions)

Another reason to disallow LLM contributions in open source projects would be that it gets in the way of the project nurturing “new, confident, trustworthy contributors“, says (even) Simon Willison:

It doesn't matter if the LLM helps you submit a perfect PR to Zig - the time the Zig team spends reviewing your work does nothing to help them add new, confident, trustworthy contributors to their overall project.

(From: The Zig project's rationale for their firm anti-AI contribution policy)

(via Cassandra)

Of course kids are bypassing age verification:

Kids used to draw on their faces for fun. Now, they're doing it to play Roblox.

(From: Mustache Mischief: Kids Bypass UK's Digital Age Barriers - CNET)

In another great piece about what LLM usage actually means in the workplace beyond promise of ‘efficiency’ and ‘profit’, Dave Rupert notes:

I’d rather have a human-to-human conversation with you, not a chat with Claude by proxy. What Claude said is an okay chunk of “anecdata”, but it’s not a substitute for our working relationship.

(From: I don't want a screenshot of your Claude conversation - daverupert.com)

As part of my work for the W3C's Advisory Board (AB), I co-wrote a short overview of reasons why or why not to use LLMs in the standards process.

My own personal opinions are a bit more extreme than what is captured here, and that's fine, the goal of this document is to capture what the AB mostly agrees on when it comes to usage of these tools. We wanted to be balanced, too, which is why a sceptic (me) and an enthusiast (Elena) decided to collaborate on it.

As we wrote:

we want to highlight considerations around different ways in which LLMs can be useful or problematic when it comes to leveraging them in standards work at W3C.

(From: Use of Large Language Models in Standards Work)

My opinion on "Artificial Intelligence" | ./axel.leroy.sh) gives an overview of developer's Axel Leroy's thought process on the question whether to use LLMs at work.

Dave wrote a great piece about friction, and Matthias pointed out something rather important that followed from this:

The frictionless version of creative work isn’t faster creative work. It’s no creative work at all.

(From: The Shape of Friction · Matthias Ott)

I could not agree more. Bringing in experience and judgment (and, imo, intentions) is what makes works meaningfully creative, removing that leaves us with very little left.

Simon P. Couch wrote about energy consumption of LLM usage and shares some of his napkin calculations (in lieu of data being made available my large AI vendors).

He concludes that it isn't too bad:

Personally, I don’t know that this scale of energy (and, ostensibly, water) use is significant enough to make me decrease my use of coding agents

(From: Electricity use of AI coding agents | Simon P. Couch – Simon P. Couch)

When you're not into ITC Garamond:

The most distinctive element of [ITC Garamond] is its enormous lower-case x-height. In theory this improves its legibilty, but only in the same way that dog poop’s creamy consistency in theory should make it more edible.

(From: I Hate ITC Garamond - DesignObserver)

(via Eric Bailey)