Henry Merrilees
@henrym.bsky.social
Software at Diderot — LA+Chicago
My much cooler twin: @kristinmerrilees.bsky.social at phonetime.news
My much cooler twin: @kristinmerrilees.bsky.social at phonetime.news
It’s a difficult also when AI is often used to mean “ML that the speaker doesn’t like”
November 9, 2025 at 9:14 PM
It’s a difficult also when AI is often used to mean “ML that the speaker doesn’t like”
Fascinating shift of language from when many ML practitioners were skeptical of being associated with symbolic AI to now having users want to distinguish the part of ML itself (LLMs) as AI from the rest of ML
November 9, 2025 at 9:09 PM
Fascinating shift of language from when many ML practitioners were skeptical of being associated with symbolic AI to now having users want to distinguish the part of ML itself (LLMs) as AI from the rest of ML
Has always been the case! And historically even broader beyond ML to include predominantly symbolic methods!
November 9, 2025 at 9:03 PM
Has always been the case! And historically even broader beyond ML to include predominantly symbolic methods!
Your stuff is so awesome! Inspired me to start using my plotter again!
August 17, 2025 at 6:35 PM
Your stuff is so awesome! Inspired me to start using my plotter again!
They have Kahnmigo, which AFAICT does not claim any of these capabilities today. In the context of the interview (he’s on the podcast/interview circuit to sell his new book), i don’t want to speculate on what he thinks Kahn Academy will do in the future without reading it, which I probably won’t :/
June 27, 2025 at 7:11 PM
They have Kahnmigo, which AFAICT does not claim any of these capabilities today. In the context of the interview (he’s on the podcast/interview circuit to sell his new book), i don’t want to speculate on what he thinks Kahn Academy will do in the future without reading it, which I probably won’t :/
And to address more directly, I agree completely that not all problems are software engineering problems. Especially here. But with regard to the initial claim about technical capability, that is a software engineering problem, though necessarily also invoking many other disciplines.
June 27, 2025 at 7:01 PM
And to address more directly, I agree completely that not all problems are software engineering problems. Especially here. But with regard to the initial claim about technical capability, that is a software engineering problem, though necessarily also invoking many other disciplines.
But “Kahn Academy is not well suited to realize these speculations, descriptively or ethically” is very different from “nobody will ever,” which was the initial claim.
June 27, 2025 at 6:57 PM
But “Kahn Academy is not well suited to realize these speculations, descriptively or ethically” is very different from “nobody will ever,” which was the initial claim.
I might be lost here but I thought this is just Sal Kahn speculating as to what could be done as opposed to any company making claims as to what they were going to do. I agree wholeheartedly if it were the latter, and for the former, I think his speculations are premature.
June 27, 2025 at 6:54 PM
I might be lost here but I thought this is just Sal Kahn speculating as to what could be done as opposed to any company making claims as to what they were going to do. I agree wholeheartedly if it were the latter, and for the former, I think his speculations are premature.
(End excerpt)
Obviously this is extraordinarily difficult to get right and potentially misses a lot of what teachers do in a traditional classroom (including the bad stuff!) but my point is simply that evidence is needed to get from there to “no pathway.”
Obviously this is extraordinarily difficult to get right and potentially misses a lot of what teachers do in a traditional classroom (including the bad stuff!) but my point is simply that evidence is needed to get from there to “no pathway.”
June 27, 2025 at 6:48 PM
(End excerpt)
Obviously this is extraordinarily difficult to get right and potentially misses a lot of what teachers do in a traditional classroom (including the bad stuff!) but my point is simply that evidence is needed to get from there to “no pathway.”
Obviously this is extraordinarily difficult to get right and potentially misses a lot of what teachers do in a traditional classroom (including the bad stuff!) but my point is simply that evidence is needed to get from there to “no pathway.”
And then, they're able to distill all of that and communicate to the parents. It's not once a term. It's almost real time.
June 27, 2025 at 6:44 PM
And then, they're able to distill all of that and communicate to the parents. It's not once a term. It's almost real time.
And then, they'll report back to the teacher and say, 'Hey, I noticed Katty is not as engaged as she was yesterday' or 'Sal's really engaged today. Did you know that he's really into baseball? Let's make the next example about that just for Sal!'…
June 27, 2025 at 6:44 PM
And then, they'll report back to the teacher and say, 'Hey, I noticed Katty is not as engaged as she was yesterday' or 'Sal's really engaged today. Did you know that he's really into baseball? Let's make the next example about that just for Sal!'…
Relevant section:
When class starts, those grad students, along with the teacher, are going to be able to walk around and help your children when they need it. They don't have to wait for that help….
When class starts, those grad students, along with the teacher, are going to be able to walk around and help your children when they need it. They don't have to wait for that help….
June 27, 2025 at 6:44 PM
Relevant section:
When class starts, those grad students, along with the teacher, are going to be able to walk around and help your children when they need it. They don't have to wait for that help….
When class starts, those grad students, along with the teacher, are going to be able to walk around and help your children when they need it. They don't have to wait for that help….
Sal Kahn interview in which he is optimistic about AI in education and speculates about potential uses.
June 27, 2025 at 6:43 PM
Sal Kahn interview in which he is optimistic about AI in education and speculates about potential uses.
Incredibly low! Which can be readily admitted without at all conceding either “no pathway” (initial claim) or “no ethical or responsible way it can be done” (normative claim, distinct in my view but certainly worth addressing).
June 27, 2025 at 6:39 PM
Incredibly low! Which can be readily admitted without at all conceding either “no pathway” (initial claim) or “no ethical or responsible way it can be done” (normative claim, distinct in my view but certainly worth addressing).
Any engagement metric could indeed be used to marginalize neurodivergence. Should be expected to, like you said. But these are conditions of our vocationally-oriented education system as it exists, not an ethical impossibility inherent to the technology that would justify “no pathway” as claimed.
June 27, 2025 at 3:22 AM
Any engagement metric could indeed be used to marginalize neurodivergence. Should be expected to, like you said. But these are conditions of our vocationally-oriented education system as it exists, not an ethical impossibility inherent to the technology that would justify “no pathway” as claimed.
I wasn’t clear enough, but my response was intended to hinge on that current tech does not have the capabilities you mentioned.
June 27, 2025 at 3:07 AM
I wasn’t clear enough, but my response was intended to hinge on that current tech does not have the capabilities you mentioned.
Feels like the goalposts are shifting quite a bit here. I can’t find reference to correction anywhere in the original interview.
June 27, 2025 at 3:03 AM
Feels like the goalposts are shifting quite a bit here. I can’t find reference to correction anywhere in the original interview.
Maybe this is a software-person take, but I wouldn’t read that as supposing anything beyond current capabilities with conventional tech, aside from natural language formatting.
June 25, 2025 at 4:51 PM
Maybe this is a software-person take, but I wouldn’t read that as supposing anything beyond current capabilities with conventional tech, aside from natural language formatting.
“All those with the funding to pull this off have horrible incentives and can be assured to do it poorly and roll it out prematurely” is much much more credible than that it simply cannot be done. It’s not magic, it’s a software engineering problem.
June 25, 2025 at 4:47 PM
“All those with the funding to pull this off have horrible incentives and can be assured to do it poorly and roll it out prematurely” is much much more credible than that it simply cannot be done. It’s not magic, it’s a software engineering problem.
The biggest gap in the task described is software engineering and cost. I have already seen LLM-based feedback working to great effect in an upper-div OS course! Not without significant work by the professor (surprise, she’s good at software engineering).
June 25, 2025 at 4:43 PM
The biggest gap in the task described is software engineering and cost. I have already seen LLM-based feedback working to great effect in an upper-div OS course! Not without significant work by the professor (surprise, she’s good at software engineering).