I've tried all of the models available right now, and Claude Opus is by far the most capable.
I had an assertion failure triggered in a fairly complex open-source C library I was using, and Claude Opus not only found the cause, but wrote a self-contained reproduction code I could add to a GitHub issue. And it also added tests for that issue, and fixed the underlying issue.
I am sincerely impressed by the capabilities of Claude Opus. Too bad its usage is so expensive.
I've been on ChatGPT Pro plan since introduced, and also used codex-rs since it was made public, never hit a limit. Came close last week, not sure if the limits were recently introduced or there always was but they got lowered, but I think that's as close to "unlimited" as you can get without running your own inference.
I've tried Anthropic's Max plan before, but hit limits after just a couple of hours, same with Google's stuff, but wasn't doing anything radically different when I tried those, compared with Codex, so seems other's limits are way lower.
I finally bit the bullet and got a $200 Claude subscription last month. It's been a busy month and I've used it a lot. More than is healthy, more than I sustainably could for more than a few weeks. I've managed to hit a 5 hour limit exactly once (20 minutes before it refreshed) and I've never hit more than 80% of a weekly limit.
But if I did - and I could imagine having some specific highly parallelizable work like writing a bazillion unit tests where I send out 40 subagents at a time - then the solution would be to buy two subscriptions. Not switch to API billing.
While that sounds impressive, a $200 subscription is still not pocket change. Do you have any approximation of the amount of tokens you use on average, and how much would it cost on a per-million-of-token billing?
Good question. I bought the subscription on 16th of January. I used a tool called ccusage.com. Assuming it's accurate, since then I would have racked up $1976.64 in API charges. There's been one single day that would have cost $324.82.
This is actually a great way to foster the learning spirit in the age of AI. Even if the student uses AI to arrive at an answer, they will still need to, at the very least, ask the AI to give it an explanation that will teach them how it arrived to the solution.
No this is not the way we want learning to be - just like how students are banned from using calculators until they have mastered the foundational thinking.
That's a fair point, but AI can do much more than just provide you with an answer like a calculator.
AI can explain the underlying process of manual computation and help you learn it. You can ask it questions when you're confused, and it will keep explaining no matter how off the topic you go.
We don't consider tutoring bad for learning - quite the contrary, we tutor slower students to help them catch up, and advanced students to help them fulfill their potential.
If we use AI as if it was an automated, tireless tutor, it may change learning for the better. Not like it was anywhere near great as it was.
There is research that shows that banning calculators impedes the learning of maths. It is certainly not obvious to me that calculators will have a negative effect - I certainly always allowed my kids to use them.
LLMs are trickier and use needs to be restricted to stop cheating, just as my kids had restrictions on what calculators they could use in some exams. That does not mean they are all bad or even net bad if used correctly.
> There is research that shows that banning calculators impedes the learning of maths.
I've seen oodles of research concluding the opposite at the primary level (grades 1- 5, say). If your mentioned research exists, it must be very well hidden :-/
> Do calculators threaten basic skills? The answer consistently seemed to be no, provided those basic skills have first been developed with paper and pencil.
So, yeah, there are no studies I have found that support any assertion along the lines of:
>>> There is research that shows that banning calculators impedes the learning of maths.
If you actually find any, we still have to consider that things like this meta-study you posted is already 74-studies ahead in confirming that you are wrong.
Best would be for you to find 75 studies that confirm your hypothesis. Unfortunately, even though I read studies all the time, and even at one point had full access via institutional license to full-text of studies, and spent almost all of my after-hours time between 2009 and 2011 actually reading papers on primary/foundational education, I have not seen even one that supports your assertion.
I have read well over a hundred papers on the subject, and did not find one. I am skeptical that you will find any.
> There is research that shows that banning calculators impedes the learning of maths.
Please share what you know. My search found a heap of opinions and just one study where use of calculators made children less able to calculate by themselves, not the ability to learn and understand math in general.
Yeah; it gets steps 1-3 right, 4-6 obviously wrong, and then 7-9 subtly wrong such that a student, who needs it step by step while learning, can't tell.
An understanding of READ_ONCE() and WRITE_ONCE() is important for kernel developers who will be dealing with any sort of concurrent access to data. So, naturally, they are almost entirely absent from the kernel's documentation.
/*
* Yes, this permits 64-bit accesses on 32-bit architectures. These will
* actually be atomic in some cases (namely Armv7 + LPAE), but for others we
* rely on the access being split into 2x32-bit accesses for a 32-bit quantity
* (e.g. a virtual address) and a strong prevailing wind.
*/
It's funny how easy Plan 9 would make all this. Just mount the work dir as readonly in Cowork's filesystem namespace and mount a write-only dir for output.
We can still do this via containers, though. But it does have some friction.
How realistic are they?
reply