Is this the fourth time I've rewritten this just this week? Yep. Everything is moving fast and breaking my newsletter But I'm hoping that this will be an evergreen analogy.

Recently, lots of people have been talking about how micro LLMs are the answer to anyone's concerns about LLM/genAI resource overconsumption. Because they're still cool toys like the big models, but they use less of everything. So isn't that a good thing?

Maybe?

I'm going to use excerpts from my side of a recent conversation on this topic.

I know that people who work in AI tell me that there are lower impacts from tools that use limited data set. But then I also see other info from outsiders — largely coming at it from the back door using inferred information based on resource consumption of known hardware — and they say things like "yeah, it's less, but it's still a staggering amount." So I'm operating under the assumption that it's still significant.

We're at a global point where we should be doing what we can do mitigate emissions across the board, and yet people are adopting AI tools broadly and using them wildly. And then the "leaner" LLMs are being hyped as being the answer, but this feels to me like eating one sleeve of Thin Mints instead of the whole box and acting like that is the portrait of restraint.*

It's like a retailer who marks up the price of an item and then offers a discount to bring it back in line with normal prices, but making you think you got a special deal.**

Micro-LLMs aren't about finding the most efficient way to get the job done. They're about making people feel better about AI adoption.

Meanwhile, someone who I know uses AI tools extensively across their work and personal lives (company with AI integration, writing and brainstorming using AI, and even having an AI "friend" that they ask for advice) gave me shit for making the annual lasagna for my son's birthday that used both beef and cheese, because I was destroying the planet with non-vegan food.

It just makes my brain ache. Just because an AI tool is better doesn't mean it's good for the planet. Or efficient. Or even accurate.

We're rushing headlong into something that we know isn't accurate. We're taking professional (reputational), personal (privacy), and planetary (environmental) risks that simply aren't necessary. And I worry that in less than three years since ChatGPT hit the mainstream, we've already done more harm than good.
​

* For those who live outside the 50 states, Thin Mints are Girl Scout Cookies, available only for a limited time each year, and somewhat legendary for being consumed well beyond the listed portion sizes.

** It reminds me of the street bazaar scene in Casablanca where the vendor is trying to sell linens to Ilsa for 700 Francs. Rick tells her she's being cheated. Vendor: "Ah, the lady is a friend of Rick's? For friends of Rick's we have a small discount. 700 Francs did I say? You can have it for 200."

​Rice is becoming more resilient, hopefully securing the food supply against a warming climate.

NGFS — a group of the world's central banks — released a report about near-term financial risks associated with climate change. "delaying transition efforts increase the economic costs of transitioning and could cause additional financial stress."

Some good news: China's emissions have begun to fall. The country has brought a tremendous amount of renewable energy online to power its growth.

​MIT Technology Review talks about what it's going to take to power AI. AI adoption is growing exponentially, but the power supply isn't. So how are we going to keep up with that demand? By falling back on fossil fuels.

🗒️

Consulting

​Reduce your Scope 3 emissions​

Alisa Bonsignore

Founder, Strategist, and Author