> "Let’s break down what it takes to create just one course – say, an introductory course on core pre-algebra concepts:
- 50+ core concepts to teach
- 20+ problems per concept
- → ~1,000+ individual problems per course
That last number kept us up at night. You can’t just explain a concept once and move on. You need enough variations to let learners truly master each idea, enough edge cases to build real understanding, and enough of a ramp in difficulty to create that perfect learning curve.
Designing the right game and sequence of concepts, so that learning feels like flow, is the fun part. But then you need to make a thousand carefully calibrated problems. And that part is a lot less fun – and it takes a long time."
This seems like a practical and appropriate use of LLMs. It reminds me of a similar application I heard about recently from a friend who teaches language classes using ChatGPT or equivalent to generate dozens of example sentences to teach specific grammar rules.
I find it refreshingly limited in scope compared to many projects that aim to outsource the creative process altogether rather than focusing on automating the legitimately menial and repetitive tasks.
> It reminds me of a similar application I heard about recently from a friend who teaches language classes using ChatGPT or equivalent to generate dozens of example sentences to teach specific grammar rules.
I'm not surprised that ChatGPT is being used this way. I used to teach ESL via the JET Programme and a lot of my job was about creating games and exercises that allowed students to get varied practice. I created or extended at least a hundred games in my 5 years there. These games weren't all successful, but the ones that were hit a cross section of:
1. repeatability (enabling pattern matching)
2. variation (just different enough to make it not mindless pattern matching)
3. difficulty curve (range of problems)
4. and fun.
As I learned the hobbies and personalities of all my students, I was able to get a higher success rate on 4. Being able to reuse games across multiple years helped me to fine tune 3. But nailing down 1 and 2 were always a challenge because of scale. A lot of times that was solved through instructors jumping in and adding new wrinkles on the spot. That worked well in the small schools I taught at (sometimes 3 of us for 20 students) but I don't think it would have worked as well in larger classrooms. I get the sense that using LLMs in creative ways has the potential to help solve the scaling problem.
What are "bullet points" as named in this article? It's difficult to find anything on Google not referring to the typographical symbol of the same name.
Context: "Archaeologists also found a hammer, tweezers, pliers, keys, knives, bullet points, and a completely preserved 14th century gauntlet, in addition to fragments of its counterpart worn on the other hand."
It could very well be referring to bullets in the "pew! pew! pew!" sense.
I'd recommend The Military Revolution by Geoffrey Parker[1] if you're curious to know more about the transition from knights to guns and all the strategic ramifications.
Ah, I'd been thinking it must be a device somehow related to manufacturing bullets (but how? was the question) -- it hadn't occurred to me that it could mean the pointy tips of bullets themselves. And thank you for the book recommendation.
Something's off here - bullet shaped bullets didn't appear until 1820 or so, musket balls preceeded .. but this is an excavation of a 14th century workspace with pre ballistic armor.
I'm tempted to think "bullet points" is either something lost in translation OR (it happens) a transcribed into text NOTE that indicated the list should have been made as bullet points.
Interesting that you raise the connection to schizotypal thinking. I recently learned there is a quite explicit connection between excessively instrumental, abstraction-obsessed thinking and schizophrenia. I recommend looking into the work of Iain McGilchrist if you want to learn more. A brief sample: https://www.youtube.com/watch?v=jkfMnaLpU7s
I second the recommendation for Nongbri's "Before Religion". It definitely improved my understanding of what it is we even mean / have meant by "religion".
For a more evolutionary understanding of religion, I'd recommend Joseph Henrich's "The WEIRDest People in the World" and "The Secret of our Success".
Finally though certainly not least importantly, I'd recommend Iain McGilchrist's "The Master and His Emissary" and "The Matter with Things". Both touch on how it is we see and understand the world and how the current default perspective differs from those of other times and places.
Isn't this true of all new strains of all contagious pathogens? They are constantly undergoing change. We estimate low probability that mutated viruses or bacteria descended from strains we are familiar with will have dramatically more harmful long-term effects, but we don't actually know for certain that any given year's new strains won't have very different risk profiles until much later. We can make probabilistic models based on historical data, but they are unavoidably vulnerable to underestimating the risk of black swan events.
One could argue that certain features of covid make it riskier with regard to long-term effects, but that is not a proposition that is well developed in the public conversation, especially by proponents of the zoonosis hypothesis. The lab origin hypothesis with its accompanying assumptions of serial passage and direct gene modification would in my eyes strengthen the case that covid's long-term effects were less likely to conform to historical data on other viral infections, though interestingly the intersection of those who find the lab origin more convincing with those who have serious concerns about long-term harms is a pretty small set.
> Isn't this true of all new strains of all contagious pathogens?
Yes. The distinguishing factor there is that most new strains do not kill millions of people within the first year or two of discovery. Compare, for example, the H1N1 variant that caused the 2009 flu pandemic, which killed "only" around 300,000 people (based on best excess death estimates).
> One could argue that certain features of covid make it riskier with regard to long-term effects, but that is not a proposition that is well developed in the public conversation, especially by proponents of the zoonosis hypothesis.
This has long been an established part of the messaging: we're more or less confident that short term effects to young, otherwise healthy individuals are minor. The guidance has still been to avoid infection, because we're not confident that mild short term guarantee or protect against serious long term effects. Chickenpox (and subsequently shingles) exemplify this.
I understand the intuition that a non-zoonotic origin would lend credence to the possibility of long term risks, but I don't think the epidemiology actually supports the intuition: my understanding is that viruses that jump the species gap tend to have higher variability in terms of their harm to the new species.
I've noticed the same strange misaligned covid explanation/behavior. Lab leak would make me more fearful of the virus itself, pure zoonotic origin and I say Jesus take the wheel and by that I mean countless generations of evolution tuning my immune system against similar virus for familial survival take the wheel.
This mis-states the relationship between your immune system and novel viruses, especially ones that cross species boundaries. Viruses adapt to avoid the adjustments the immune system makes, and zoonotic transmission means that your immune system is "seeing" all kinds of novel adaptations for the first time.
This all depends on how you're approaching the question. Why do you want to read the news?
What do you hope to get out of it? You're absolutely right that the NYT has changed in recent years, but that change has been part of systemic shifts that have affected all of journalism as it existed before the internet.
For most outlets, "the news" is now substantially more infotainment and in-group sermonizing than it used to be. Buzzfeed-y clickbait and Facebook-y rageporn juice engagement numbers like little else. Competition for your attention online is fierce and, with access to alternatives at the click of a button, audiences have very little appetite to continue reading an outlet that publishes things they don't agree with.
Are you looking to stay abreast of conversational topics in certain social circles? Are you looking for high-quality information that will help you form more accurate predictions about the world? If so, what types of predictions are most important to you? Economic? Social? Political?
Do you like your facts presented in an editorialized fashion or do you want events to be reported without being nudged to feel one way or another about them? Do you just want a news source that provides fun stuff to read? Do you read the news because you're not very into sports and books are too long? (I've been there in life.) How important are polish and sleek UIs to you? Do you want to read content that goes down easy or do you want to survey a broad set of views on subjects to know what other people are thinking even if some of that makes you angry or confused?
This is maybe not quite what you're looking for, but I'd use this decision as an opportunity to step back and reflect on the bigger question of why you want a subscription to a news publication in the first place.
Once you have the answer to some of those questions, that should help narrow down your search.
I don't want to make myself too identifiable, but it's a fairly big place and I lucked into it more than anything. Can't give any specific advice how to find a similar job besides looking for remote listings at non tech places.
- 50+ core concepts to teach - 20+ problems per concept - → ~1,000+ individual problems per course
That last number kept us up at night. You can’t just explain a concept once and move on. You need enough variations to let learners truly master each idea, enough edge cases to build real understanding, and enough of a ramp in difficulty to create that perfect learning curve.
Designing the right game and sequence of concepts, so that learning feels like flow, is the fun part. But then you need to make a thousand carefully calibrated problems. And that part is a lot less fun – and it takes a long time."
This seems like a practical and appropriate use of LLMs. It reminds me of a similar application I heard about recently from a friend who teaches language classes using ChatGPT or equivalent to generate dozens of example sentences to teach specific grammar rules.
I find it refreshingly limited in scope compared to many projects that aim to outsource the creative process altogether rather than focusing on automating the legitimately menial and repetitive tasks.