Mark Betnel

What work is this from?

Even if you include *all* energy usage (per capita, even though that includes industrial uses), the human still uses 1/30th as much energy. He's right that the energy for a single query is probably less than the energy for a human to give the same response - but it's still the wrong comparison.

Even accepting his premise (which I don't), the scale is still completely off -- estimates of gpt4 training cost were 50GW-hrs, while a human operates on 100 Watts, or about 20 MW-hrs up to age 20 (when we apparently "get smart"). That's about 2500 times more energy to train the LLM than a human.

Another Killing on ICE The right wing propaganda machine can’t spin the killing of Alex Pretti

Within 4 hours of ICE (actually CBP) killing Alex Pretti, the data suggested that the right wing spin machine was having trouble gaining traction, even on their home turf over on X.

Here's a Substack I wrote analyzing X data from that time. katestarbird.substack.com/p/another-mu...

Just a random thought after some conversation about AI, "workplace readiness", and education.

Learning isn't a transfer, it's an effort. It does not work without effort.

It's wrong to quantify that effort as a measure of the learning that has happened, but it's also just misguided to imagine learning happening without effort or eliminating the effort.

Speaking for myself, as a teacher, the kind of AI I want is the kind that would make the copier and printers work every time I want to use them. THAT would free up time to focus on what matters.

Why Didn't It Stick? Students forget things. What can we do to help more students remember?

fivetwelvethirteen.substack.com/p/why-didnt-... An awesome collection of wisdom about teaching for memory and transfer from @dylanpkane.bsky.social

Yeah -- I think "useful for what?" is an important question, and the answer will change w/ development. I focus on the conceptual a lot because I think it will be a better guide in the future when the teacher isn't there and the problems are real and unstructured, about what calculations to do.

I like that example for highlighting a distinction between conceptual and "mechanical"(?) understanding. There's a student who would plug those coordinates into the distance formula, say that the diagonal is bigger, and then get mad if I tried to say there was a deeper idea to get.

Screenshot of a popup ad offering free access to the Washington Post if one uses the Comet browser from Perplexity.

This seems problematic? ... To offer "free" access to the news if one uses a particular AI agent web browser?

I appreciate some of the rest of this thread, but strongly disagree with this. Making students feel stupid doesn't make them rise to the occasion. It's possible to challenge them to be confident _and_ competent.

I use them. I insist on putting arrowheads on both ends of every line to combat that, and find that they help in understanding when to make a 2nd (or more) fbd to really understand a system and bookkeeping in accounting for every interaction. Also helpful in talking about potential energy

#teach180 day 16. Just wrapped Back to School Night, felt good talking to parents about the physics class and the AI & Ethics class. Lots of nodding heads about the AI class. And I succeeded in exactly filling the 10 minutes.

#teach180 day15 Wrapped up the medical diagnosis discussion last class (itmeson.github.io/ai-explainer...) and introduced false positive/false negative distinction. Today started the brief programming unit with python turtle graphics (used pythonsandbox.com/turtle to simplify install) Always so fun!

#teach180 day11 Only had the AI class today. Played with the TicTacToe demo (itmeson.github.io/tictac-demo/) and discussed how a simple computer program can use shifting probabilities to "learn" and why it doesn't necessarily learn the correct lessons.

As a teacher, how can I create the perfect conditions? Remove the barriers/antagonists? Provide friendly pace setters? Convince them they can do more than they think they can? Show them that it's necessary to do more?

It's a lot to think about.

Bannister and Landy both worked alone at first, but found mentors and role models who showed them that it was both possible to do more than they thought they could, and that it was *necessary* to do more. Bannister had friendly pacesetters, Landy had an invitation to Finland and perfect conditions.

Santee was ready first in early 1954, and stayed at that level longer, into 1955. But he never had the right race in the right conditions with the right competition. He had rainouts, no friendly pacesetters, and active antagonists who cut his career short and banned him from competition.

There's all kinds of lessons about extreme performance and running, but since I don't think my role is primarily about pushing students to the edge, I gravitate to the story of their relationships with mentors, about needing models of behavior, and the role of luck / removing barriers.

All three went to the 1952 Helsinki Olympics, all three had disappointing performances, and all three went home determined to take their running to the next level. In 1954 Bannister made it first (3:59.4), Landy got there weeks later (3:58.0), and Santee never made it (4:00.5).

Just finished _The Perfect Mile_ (Neal Bascomb, 2004). It's about the parallel efforts to break the 4 minute mile by Roger Bannister, John Landy, and Wes Santee, between 1952 - 1955. As a former runner and as a teacher and coach, it was a fascinating dive into life at the edge of human performance

Voices Carry I'm truly at a loss of what to say today. (Last week. It lingers, doesn't it.) So (I guess) let me just repeat a point I've made repeatedly: one of the saddest rationales for using "AI" is for "brain...

Today's bird is the screaming piha. Just a wee little thing, but one of the loudest birds in the world. 2ndbreakfast.audreywatters.com/voices-carry...

But as a counterpoint on that tension: idlewords.com/2010/03/scot... an amazing deep dive into the discovery of Vitamin C and how wrong metaphors for the correct treatment for scurvy in the 18th century lead to *losing* the cure in the 19th. Sometimes the details *do* matter.

Do students need to know how LLMs work, or to predict how they'll act? A little venting about what educational explanations are for...

mikecaulfield.substack.com/p/do-student... This is excellent. I think it's critical to think through the implications of the explanations we help students build -- the important question is whether they are productive for the student, not whether they are precisely "correct" in their ontology

a cartoon of mickey mouse in a wizard outfit Alt: a cartoon of mickey mouse in a wizard outfit leading a broom that has been magically animated to carry buckets of water

www.oneusefulthing.org/p/on-working...

Pretty interesting piece from Ethan Mollick. I've got some more complex thoughts brewing on it (it has several useful nuanced points), but I can't help thinking immediately of this:

#teach180 day 5 (yesterday) More graphing stories and first problem solving practice in physics. Finished AlphaGo doc in AI. Did learning walk through four world language classes -- amazing work from my colleagues in creating a fearless class climate in the first few days.

Yep. Just checked with my students and officemates, none of them have seen this show up yet. Gemini is now available in Google Workspace for Ed accounts, but it's turned off by default unless your workspace admin turns it on, but I don't think this "Homework Help" is connected to that.

Screenshot of a Chrome browser tab showing part of a school LMS page and a popup offering to search anything on the page using Google lens.

#teach180 day 4. Just noticed this chrome tool that shows up only when I'm on a page in our school LMS. What do you wonder about that? #aiethics #aieducation

Day 3. Broke out the vernier sensor carts and had physics SS replicate position - time graphs. Pushed them to be specific about the scales. In AI we introduced train/test/deploy vocab and started watching AlphaGo documentary.