I think that the first revolution to come from LLMs will be in education. Tutoring is a HUGE industry that creates massive achievement gaps between haves and have-nots. China went as far as banning tutoring because desperate parents would bankrupt themselves trying to give their kids a leg up. To be sure: GPT-3 makes mistakes. It struggles with leading questions, perfect/correct code, sparsely trained subject areas. But when it comes to trying to help give intuitions to a high school chemistry student, the stakes are low and the training data is good enough.
Imagine a world where every child alive had access to unlimited 24/7 pseudo tutoring offered by robots who were almost as smart/correct as actual tutors? How many millions students just "give up" every day because while they can find thousands of pages of facts/guides/lessons about a subject, there are zero results that are custom tailored to their exact gap in understanding? As a student,
-->Google can give you endless facts about a subject that you can ram your head against until a subject makes sense.
-->Chat-GPT can give you intuitions and explain the exact part that confused you with as much detail/complexity as you want!
You wouldn't even need to call it AI or a tutor (so students don't wrongly give 100% trust). It could be a peer. You could call it "My Study Group's Best Guess". How much further could students get in high school with a 24/7 tutor beside them?
~~"Hi study group, I don't understand how my teacher jumps between steps 3 and 4 for this proof in Geometry class. Can you explain it in simple terms?"
~~"Hi study group, why did my flask turn blue when I did xyz in chemistry class? Everyone else had theirs turn pink."
~~"Hi study group, I wrote "cosin - sin = co" on my math test but my teacher put a big x on my problem without telling me why."
~~"Hi study group, why is the x "number of cars" on this homework problem XYZ"
Millions of students give up trying to understand challenging subjects every day. LLMs can dramatically lower the difficulty of absorbing new ideas. In 10 years, LLMs can likely raise SAT/ACT scores 20-30% equivalent and help more students finish the key math/science prerequisites for STEM careers in high school instead of college, lowering the barrier of entry for everybody!
This. This is what I use it for most often at the moment (helping out my rusty physics). Its super helpful. If I was 20 again, teaching myself maths on Khan Academy, with this grade of chatbot in the corner ready to help out? Heaven.
This comment sounds like when electricity was just discovered and people would discuss how it could improve the speed of windmills.
We have to think broader and get out of the confines of how things work now and just think that technology only makes what we have now better. Some technology just completely changes the landscape.
LLMs will make education as we know it obsolete. There won't be need for tutoring if LLM is good enough to do anything. Anyone can ask anything on the fly. There won't be need to study anything except for fun.
One of the plot-lines is getting the Primer mass produced for the seven float orphanages. The Primer is designed to react to its owner's environment and teach them what they need to know to survive and develop. It can do so by using human actors (in the theatrical sense), but it is not the case that a human is the tutor using the Primer like an encyclopedia, as your post seems to suggest.
That would be fantastic, but I think it’s very far away. Right now LLMs are far too dumb and unreliable to be useful as tutors. Ironically, they’re only useful in domains you already know very well, because then you can detect BS.
Ya people on this forum are a small subset of the population that know how to interpret ChatGPT. The general public, let alone kids, will have a hard time distinguishing fact from linguistic hallucination.
young students still have the tendency to remember things their tutor said as hard facts. couple this with confidently wrong chatgbot ... this issue needs to be fixed first
Most likely an autodidactic exercise. I'd love to see the source, even if, or especially if, it's half-finished. Seeing someone work their way through the snarls in such a project would be useful.
FSD stack is disabled on highways. You are using the years old code. Beta v11 when it comes out will enable OP's FSD referenced improvements for highways.
I used to mock my relatives for not wanting to put their kids in a Tesla. But yeah, I'm not subjecting my family to beta software for a critical safety feature, nor should anyone else.
They are standing firm on the vision-only approach because it is the correct approach. FSD cannot be perfected unless the Tesla team puts 100% exclusive focus on perfecting vision based models that don't have lidar as a fallback. Tesla can only use lidar again for redundancy only after vision is fully solved problem.
So the beatings will continue until morale improves, eh?
This is like saying we need to get to the moon using steam power and ONLY THEN we can improve by using rockets. What if it's completely infeasible to solely use vision for FSD?
Even humans use audio processing to augment the vision.
Combing multiple sensory inputs is a hard problem in itself. Say Tesla uses LIDAR with vision...they can have a great vision experience, a great LIDAR experience, but fuses those two experiences together can create lots of problems that wind up being worse than either by itself.
Is it really equality to perfectly equalize decision-making while completely slanting the resulting the burden-taking?
I realize how dangerous/potentially harmful a "taxpayer contribution === influence" is. But I think it doesn't have to be a binary. Perhaps a "three house" style system would be an improvement for representative democracy here:
One house for representatives of each state
One house for representatives of vote by population
One house for representatives of taxpayers by net tax burden carried
Given that a large number of those "states" are net tax beneficiaries, that first house needs to be reformed.
As for having reps of taxpayers by net tax burden, what the hell does "net tax burden" mean? Tesla is useless without roads. Boeing is useless without airports. How do we apportion the tax paid against benefits received?
It's nonsense that "taxpayers" are worthy of further representation than any other citizen. The fact that corporations pay tax should not give them rights to representation in the legislature. Unless you think of corporations as one way to "represent taxpayers by net tax burden" because they are a collection of taxpayers.
I recommend every single person I know to screen themselves for Sleep Apnea -- My doctor estimates 30% of the population has it and most people go their whole lives without ever knowing.
Personally, my CPAP has boosted by daily energy/wakefulness tremendously, I feel at least 40% more energy each morning when I get a perfect CPAP adherence overnight, and I only had a minor/moderate case. If you snore, you probably have it without knowing.
Please get checked and make your life much better!
30% of the population do not have Apnea. Initial estimates in research were 3%, then a few years ago we started seeing numbers in the 8% to 12%. I have never seen a research paper that suggests anywhere near 30%, but that number gets batted around a bunch.
I'm not going to say Apnea isn't a problem. It is. I'm happy it made your life better.
I tried my best to fill in the blanks for this story, and my working understanding is this:
1) Client requests service.
2) Service is graphQL, which makes the "thousands of calls" Musk is referring to towards other backend APIs.
3) There was an incorrect international cache/CDN configuration which hurt response time of these APIs for some countries.
4) Musk claims its the API calls causing slow performance, engineers point out that its technically the cache.
Debate is about if using graphQL is even worth it if it needs such complex caching to work well
Imagine a world where every child alive had access to unlimited 24/7 pseudo tutoring offered by robots who were almost as smart/correct as actual tutors? How many millions students just "give up" every day because while they can find thousands of pages of facts/guides/lessons about a subject, there are zero results that are custom tailored to their exact gap in understanding? As a student,
-->Google can give you endless facts about a subject that you can ram your head against until a subject makes sense.
-->Chat-GPT can give you intuitions and explain the exact part that confused you with as much detail/complexity as you want!
You wouldn't even need to call it AI or a tutor (so students don't wrongly give 100% trust). It could be a peer. You could call it "My Study Group's Best Guess". How much further could students get in high school with a 24/7 tutor beside them?
~~"Hi study group, I don't understand how my teacher jumps between steps 3 and 4 for this proof in Geometry class. Can you explain it in simple terms?"
~~"Hi study group, why did my flask turn blue when I did xyz in chemistry class? Everyone else had theirs turn pink."
~~"Hi study group, I wrote "cosin - sin = co" on my math test but my teacher put a big x on my problem without telling me why."
~~"Hi study group, why is the x "number of cars" on this homework problem XYZ"
Millions of students give up trying to understand challenging subjects every day. LLMs can dramatically lower the difficulty of absorbing new ideas. In 10 years, LLMs can likely raise SAT/ACT scores 20-30% equivalent and help more students finish the key math/science prerequisites for STEM careers in high school instead of college, lowering the barrier of entry for everybody!