Working with arm64, cuda drivers, and shared ram design can be a bit of a learning experience but overall was smoother than I expected.

The Spark is a remarkably capable machine for the price. Not just for LLMs (like gpt-oss-120b), but cuda-accel polars, & even as a shared jupyterhub

Like so many of us, I've tried just saying this in previous semesters to justify our not using it, with very limited success.

And the students can use it well for simple, well scoped tasks. but outsourcing our thinking on open-ended assignments goes badly.

this lesson we must learn by experience

I don't know what clicked. I'm sure we will hit the walls again. With a large class for only 3 of us there's always a lot of variation. But it's the first time in some time I've witnessed the students know the tools are there when they are stuck, but they also know when they're better

a week ago we'd thought we'd lost them completely. The assignment was too ambitious, too open-ended; students kept trying to guess for one 'right' answer. Among our teaching team, we debated if our expectations were just too high.

we all walked out of the last session thoroughly impressed.

But as I wonder the buzzing room, or peer over shoulders, they are talking only to each other.

At my prompting, one team puts a complex query to the agent. But while it spins away, they start tapping out a solution by hand, pure ibis code, tight and elegant - it's done before GPT can reply.

What is happening?

They are re-evaluating global fisheries declines result of Worm et al's classic paper using the latest RAM legacy stock data.

I've encouraged them to use the LLMs instead of memorizing syntax.

this matters because 'AI', like any technology, is designed, owned, operated by companies that make choices.

In other tech - iPhones, TikTok - we quickly attribute design choices to specific companies. But we still discuss "AI" as if it was some disembodied discovery, more uranium than software.

sure, but sycophantic affirmation is hardly a pattern it got from memorizing the internet.

Alignment is trained. RL is the smiley face in front of the monster.

(from www.nytimes.com/2025/09/26/w...)

these tools can be useful, even transformational or foundational, but I think more in the 'duct tape + pvc piping' is foundational than in the 'one ring to rule them all' view. If even the companies are proceeding with tool use, this is something we too can build for ourselves.

As an ecologist, just putting here for the record that we are never going to have an AI model 'predict the future of biodiversity'.