"The generated code might not always be correct. In that case, run it again lmao" is the best documentation I've read all week.
"Monkey around with the prompt and pray" has replaced unit testing.
than this allegedly miraclous students are wasting there talents studying in IITs!
IIT students are amazing, i've worked with them. Far more creative and competent programmers than the average Ivy League grad in my cohort.
Would citizens of the EU and US not want to go to one of the best schools regardless of locale?
Not that it’s the same, but have you looked at either of these countries’ pasts? Even relatively recently— the Marshall Islands, or still ongoing, in Mauritius/Diego Garcia.
I say this, having studied at a European university as well as visited IIT and a top US university. These are different traditions with different strength and weaknesses but my point was not to say that IIT is better, it was to put the OP comment into a context.
IITs are way more teaching focused than research focused, and are have a very light international student presence - two factors that enormously influence rankings.
So just because someone ostensibly "passed" an exam doesn't mean they actually did.
Of course, India's education system is far from the only one where cheating is rampant, which is just one reason to look on credentials and accolades everywhere with a skeptical eye.
[1] - https://www.economist.com/asia/2022/05/26/indias-exams-are-p...
Your rich dad endows a building in an Ivy and you are in - and apparently it's all good!
I just finished writing up a long post [1] that describes how this can work on local models. It's a bit tricky to do via API efficiently, but hopefully OpenAI will give us the primitives one day to appropriately steer these models to do the right things (tm).
Perhaps this is infeasible with the current cost per generation but a multiple choice display could be handy (perhaps split the screen into quadrants and pick the one that most fits what you prefer).
There’s two approaches I use, I’m sure there are more.
* Do multiple completions and, filter the ones that successfully run, and take the most common result.
* Do a completion, if it fails, ask the Llm to find and correct the bug.
It could be inaccurate with high precision - like a bow that consistently undershoots, and you would have a harder time trying to correct it.