What if 'good code' is just 'code optimized for humans who can't hold much in working memory'? The model doesn't need breadcrumbs if it can see everything at once. If context windows 100x, think some of this may be less relevant. Big IF, have no idea tbh, hard to predict.
Europeans spend all their time complaining rather than building the next Google. That's why they have no Googles and we aren't going to let them in anymore. Regulation is also bad. /s
Always took it to be synonymous with "enlightenment values", created in Europe and by Christians. (Who I believe were at least somewhat secular). I am unsure if we are, at present, a bastion of said values.
Good article though rather than philosphizing about lost souls, we probably should go back to the future.
Pen-and-paper exams only. No take-home essays or assignments. Assessments done in person under supervision. No devices in class. Heavily reduced remote learning or online coursework. Coursework redesigned so that any out-of-class work is explicitly AI-collaborative. Frequent low-stakes in-class writing to verify student voice and baseline ability. And when resources permit have oral exams and presentations as a means of assessment.
We did this for decades when tuition was a fraction of today's cost. Any argument that we can't return to basics is bollocks.
If you're trying to hawk education for $$$$$$, probably need to offer some actual human instruction, not Zoom and Discord sessions that anyone could run from their bedroom.
If they can't, then the rot and capture really is as bad as this makes out, and to update Will Hunting: the kids might as well save $150k and get their learning for $20/month on ChatGPT.
This is the correct answer. I work at a CSU (non-faculty) and the issue here is many of the faculty like using online and automated systems to dish out the work and the grading. Going back to doing it the old fashioned way will provoke a pushback from the faculty who will complain about workloads etc...
My parents had oral exams in university. I feel like that actually is a better format that does not rewarding cramming, but is interactive and over quicker. It means that there is a dynamic that actually allows for grading problem solving over regurgitation.
I agree, though I shudder to imagine how cringey the switchover would be. A significant number of students already had poor diction and linguistic skills when I was in college, and recent evidence shows this situation has likely become worse.
It's ironic. In Italy we always had constant oral exams (and still do!) from elementary school all the way to Uni. At least 2 per week in high school.
In an effort to standardize European systems many courses are trying to get rid of them because foreign students are particularly weak in an oral defense.
was the system I was educated under, when I took orals it was a result of being a scruffy kid that wore no shoes but passed general high school and math talent exams better than all but three others my age in the state.
( For interest, the three that ramked higher than myself that year in Tertiary admissions exams were all educated in expensive private schools in the capital city- I got by on School of the Air, a bunch of books and a few years at a smallish remote high school in far north W.Australia
it's almost as though what we need is to entirely ditch the university scam and create something new that more closely resembles what its real purpose was always supposed to be.
I don't think this is an LLM bug. I think this is an implementation bug. The model thinks it is 2024 due to training cutoff date, though it is "connected to Search", so conflicting information is getting in the prompt.
Even since the very early versions of Sydney and ChatGPT their system prompt has started with the current date/time (and username, user geolocation etc), it seems odd they'd leave it off here.
A better tell IMO is an unnatural huge amount of editorialized h2s / h3s. Often they are overly lofty.
reply