@shortstein Flip k times, let S be the number of heads. If S < k/2, return coin has heads with prob 0.4; otherwise, return coin has heads with prob 0.6. The error prob is max{ P(Xโฅk/2), P(Yโคk/2) }, with X ~ Bin(k, 0.4), Y ~ Bin(k, 0.6). This error prob is at most 5% when k โฅ 67.
@KWargamer @Hasen_Judi ahh that last point especially helps make sense of this. any idea of how prevalent this is at a place like google/meta? they are big enough companies where I could imagine some teams operating like this and having people like that.
@KWargamer @Hasen_Judi Iโve heard of people like this but I always wonder how they get there, whether they are actually a prodigy developer or something else is going on.
@KWargamer @Hasen_Judi I see your thought process and I think thereโs some truth to it for many situations. There do exist problems that require a bit of thought to architect solutions for tho, not because of the software side but due to the constraints of the system and ensuring correctness.
@KWargamer @Hasen_Judi I see, makes sense. Feels weird tho, never seen such an environment in the wild. I can appreciate someone who has good ideas, but I cannot imagine these engineers have many if their lack of skills means they do not have a precise sense of pros and cons of ideas they have.
@Hasen_Judi @KWargamer Iโve seen computer science students on Reddit crying about the fact they relied way too much on LLMs, to the point they are leaving uni with barely any actual programming skills to help them got a job. It even hurts them in learning broader computer science concepts.
@KWargamer @Hasen_Judi Someone with the inability to do such a basic problem just makes me wonder if an LLM would be more useful than them. I have to imagine for such a high fail rate, most of these people must have never actually learned to program and are just hoping to fly under the radar.
@adad8m I made slides [1] a bit ago along this topic and a key result implied for any fixed โ > 0 and for any random sample of n = O(exp(d โ^2)) points from the d-ball, with good prob they are all almost unit vectors and their inner products are โค โ.
[1]: https://t.co/GThB9YzA6v
@stevenstrogatz The equation is equivalent to 0 = mn-n-2m with positive integers m,n. Substitute m=x+1 and n=y+2 and get new equation xy=2 with positive integers x, y. The only two solns for (x,y) then are (1,2) and (2,1) which implies (m,n) has only solns (2,4) and (3,3).
@alz_zyd_ can find plenty of math at the grad level for CS folks who do theoretical CS or machine learning theory, to name a couple, so itโs still not like ME is inherently doing tougher math stuff. Most SWEs may not do much math on the job but most ME engineers donโt either. 6/x
@alz_zyd_ Now if you start to talk about grad school, things change some. Some ME folks get into scientific computing or control theory and have to dive into more mathematical areas like functional analysis or dynamic systems, so some MEs get into more math. But you 5/x