boshalfoshal

9 ms·

61.
by boshalfoshal 2y ago
I think the super played out twitter adage has some merit to it: "it makes 10x devs 100x devs." Those who already have a high level idea of what to do and roughly how to execute it benefit the most from LLMs at the moment. This is
62.
by boshalfoshal 2y ago
I seriously don't know why this is making rounds on HN as if it were some revelation. It should be pretty obvious that messing with your brain circutry in such an overt manner leads to various unintended consequences, just like every o
63.
by boshalfoshal 2y ago
I mean this is clearly an incorrect take? Google search cannot reason about its contents autonomously, nor synthesize out of sample data from the things it indexes. Google search is a query over user generated content. LLMs generate content
64.
by boshalfoshal 2y ago
Well thats the argument most people here are making - that current LLMs are not good enough to be fully autonomous precisely because a human operator has to "put the right thing into them to get the right thing out." If I'm s
65.
by boshalfoshal 2y ago
Maybe you're retired or not a SWE or knowledge worker anymore, but I have a decent amount of concern about this future. As a society, we have not even begun to think about what happens when large swathes of the population become unempl
66.
by boshalfoshal 2y ago
No, I'm mostly referring to buying talent and GPUs. I can't point to a single Andrej Karpathy or Chris Lattner or Jim Keller working at Mercedes, BMW, GM, etc (not to mention the not so big named people who are still very very goo
67.
by boshalfoshal 2y ago
Yes there are actual "legal" notions of driver assistance "levels," but out of those 3 companies Tesla seems to be taking it the most seriously. Elon has a high conviction in the product and has continued to pour billion
68.
by boshalfoshal 2y ago
I think it will be Tesla. Regulatory pressures for them just lifted, and they move faster than all the other driverless companies, courtesy of Elon's fairly hardcore mode of operation. It may not be a perfect driving solution (probably
69.
by boshalfoshal 2y ago
Supercruise might be comparable to Tesla's "Autosteer" product but its incomparable to Tesla's paid "FSD Supervised" product. The latter is closer to a Waymo or Cruise than it is to a Mercedes or GM driver assi
70.
by boshalfoshal 2y ago
I think at an even smaller scope than robotaxis, having a really convincing level 4 driver assist (dare I say "full self driving" mode) for general consumer vehicles is a decent value proposition. Both for prospective new car buye
71.
by boshalfoshal 2y ago
For (5), at least in most modern software companies, this isn't actually a "requirement," but places end up having some distribution of people from top schools anyway. This is usually just due to societal bias (oh, the guy fr
72.
by boshalfoshal 2y ago
> how are we to be confident that its safe? I hope you realize that these companies dont just push updates to your car like vscode does. Every change has to be unit tested, integration tested, tested in simulation, driven on a multiple c
73.
by boshalfoshal 2y ago
You are conflating two seperate problems (security vs functionality). "Firmware" can be open source and secure, but how does this translate to driving performance at all? Why does it matter if the firmware is validated by security
74.
by boshalfoshal 2y ago
Theres a difference between knowledge and intelligence. You might know these concepts, but could you apply them in novel situations under pressure in an interview? Could you maybe link disparate ideas you know together to solve a completely
75.
by boshalfoshal 2y ago
As much as people dunk on Tesla, they are the only "OEM" car company that actually takes software and self driving seriously. The FSD program is more akin to Waymo (talen-twise, pay-wise, investment-wise, etc.), than it is to Ford
76.
by boshalfoshal 2y ago
I think cache coherency protocols are less intuitive and less talked about when people discuss about caching, so it would be nice to have some discussion on that too. But otherwise this is a good general overview of how caching is useful.
77.
by boshalfoshal 2y ago
I agree with you, but I don't really see how this invalidates the style of interviews where you're presented with some timeboxed coding problem (of reasonably scaled difficulty) and are asked to solve it. There will be bad actors
78.
by boshalfoshal 2y ago
The problem you are describing is interview variance and hiring bias, not leetcode. This happens irrespective of interview style. Many companies have question banks that are specially designed to be fair/have some contextual relevance
79.
by boshalfoshal 2y ago
I see these posts a lot, and I sort of disagree with where they are coming from so I'll play devils advocate. Leetcode is used to select for one of (or both of these) 2 things for the average IC hire: 1. Do you work hard (i.e a person
80.
by boshalfoshal 2y ago
Unrelated but I see this sentiment a lot. Algorithms (and related courses) are supposed to be largely mathematical/theory based so it sort of has to be hardware agnostic. Imo an algorithms course doesn't really need to teach you a
81.
by boshalfoshal 2y ago
The thing is, at that point going to a larger public company would probably net you similar returns and less of the headache (immediately liquid compensation that you can invest into other things, no tail risk of option value suddenly becom
82.
by boshalfoshal 2y ago
Its not uncommon for PhD hires from top schools to be paid upper 6figs (i.e 600k+) of real money, according to a few friends of mine doing PhDs at top schools. OpenAI is probably the only one thats truly "competitive." I also (per
83.
by boshalfoshal 2y ago
Tesla AI orgs can pay well above market for AI talent. Thats about the only reason anyone would join. If you are insensitive to work hours but want to get paid, its not the worst option.
84.
by boshalfoshal 2y ago
> The companies that use them as a strong signal are the ones that will be absolutely demolished in engineering because they are on a fast track for a staff full of rote memorization rather than strong creative problem solving. What metr
85.
by boshalfoshal 2y ago
Maybe in certain parts it is. Last time I spoke with them they were transferring some of their on-robot stacks from C++ to Rust.
86.
by boshalfoshal 3y ago
Meta outpays pretty much all of FAANG right now, and thats saying a lot when FAANG companies are already in the top [0-5]% of compensation in the US. The uncertainty of joining an unknown team is a little annoying, I'll give you that.
87.
by boshalfoshal 3y ago
Your first point about AI changing markets was interesting until you put it in the context of automated trading, thats too small scale and kind of missing a bigger and more existential point. The long tail of AGI is that markets themselves
88.
by boshalfoshal 3y ago
Well I would classify ML, DL, statistical learning, etc. as "AI," yes. I would consider Waymo for example to be an "AI" system under a lose definition. Do you mean AI as in (proto) AGI? I think we might have slightly dif
89.
by boshalfoshal 3y ago
If you have used any modern product from a big tech company you have likely interacted with AI (probably some form of DL or statistical learning) in some capacity. Do you use gmail, youtube, or google search? All 3 products use AI for sever
90.
by boshalfoshal 3y ago
There aren't a lot of market makers or automated trading firms throwing around several hundred billions on a single stock. These types of strategies usually don't scale well with size. Thats usually flows from hedge funds, family

More ›