twothreeone

12 ms·

181.
by twothreeone 2y ago
That's because there is no such thing as "truly secure", there can only be "secure under an assumed threat model, where the attacker has these specific capabilities: ...". I agree that software engineering is gettin
182.
by twothreeone 2y ago
It continues to astonish me how little people care (i.e., it triggers the $%&@ out of me). I really appreciate the professionalism and cool rationale when faced with absolute ignorance of how shaky a foundation our "modern" so
183.
by twothreeone 2y ago
Yes, people are mixing salaries and total comp in this discussion which is unfortunately both common and misleading. It also depends very heavily on the total economic outlook, if you happened to join Google in the fall 2021 cohort your equ
184.
by twothreeone 2y ago
And if you have a "free market" the big utility companies are going to eat each other over time until you end up with de-facto regional monopolies which are then systematically run into the ground, extracting all inherent value, u
185.
by twothreeone 2y ago
I would imagine (a) keep watch over the lunar atomics and (b) fend off PLA officers stationed there permanently after Chang'e-42.
186.
by twothreeone 2y ago
Notice, however that it's not about _what_ to do, but _how_ to do it. And when it comes to to the how of human space flight, NASA has the experts, not Congress. Politicians rely on Engineers and Researchers to make informed decisions,
187.
by twothreeone 2y ago
That's not on congress though. If a budget for the agency only comes with those kinds of political strings attached, the right thing for the agency would be to say "please keep your money, the US won't be going to Moon or Mar
188.
by twothreeone 2y ago
> artisanally hand-crafted by a workforce that likes to get home before traffic gets bad. Ouch, that's gotta hurt.. I'm not saying I disagree, but I do wonder if a project is going "right" only when it starts to hit e
189.
by twothreeone 2y ago
You say that, but a world with more crows than tax payers honestly sounds kind of serene.
190.
by twothreeone 3y ago
The RDNA docs are just the tip of the iceberg.. in order to get to the compute engines (what you actually care about) you have to go through 3 layers of user-space cruft, a kernel driver, and then a firmware layer running half a dozen separ
191.
by twothreeone 3y ago
Figure 3 shows incident rate changes in percent by country, it seems there are stark increases in South American and African countries and mainly declining incident rates in western countries. Intuitively, I would associate increases in tho
192.
by twothreeone 3y ago
And even with population control technological advancements will inevitably lead to growing energy demands (e.g., see https://www.astralcodexten.com/p/sam-altman-wants-7-trillion to spark your imagination). Medium term
193.
by twothreeone 3y ago
I've been using chat apps "send to self" for this exact same workflow.. at work I just use Slack as it supports threads and basic formatting (e.g. render code blocks separately and clickable links). So every day has a few thr
194.
by twothreeone 3y ago
It's all relative, but supposedly the top 1% had $11M in the US in 2022 (e.g., via https://www.forbes.com/advisor/investing/financial-advisor/a... ). That means, 5 averagely priced houses do not put you i
195.
by twothreeone 3y ago
Specifically true for "Closed Access" cities like Los Angeles, San Francisco, New York, Boston and so on (see https://www.mercatus.org/system/files/erdmann_hamilton_and_f... ).
196.
by twothreeone 3y ago
It looks like https://osmocom.org/projects/baseband is still somewhat active.. they mainly develop for the TI Calypso (Motorola C123 and some others). My understanding is that all "modern" (meaning 3G and up)
197.
by twothreeone 3y ago
It's actually strictly worse: even if they had an IOMMU and the drivers were implemented correctly, phone vendors provide remote access tools as part of their firmware stack (e.g., see https://www.fsf.org/blogs/com
198.
by twothreeone 3y ago
Less experience overall (5+5 years), but I strongly agree. I think the main reason for this are the incentives, which halvarflake mentioned briefly in his post.. if you can directly affect valuation of your company (which in turn directly d
199.
by twothreeone 3y ago
https://www.cerebras.net/product-cloud#row-unique-7
200.
by twothreeone 3y ago
It's a bit sad to see people jump on the witch-hunt bandwagon every time something like this is published. That kind of reaction is an absolute disservice to the public. Even if this guys turns out to be a totally 100% evil fraudster w
201.
by twothreeone 3y ago
If someone on the internet says they want money to hand it to individuals in need I believe them. Especially when it is a global, for-profit, money-printing monopoly power like Google, who have repeatedly demonstrated through their actions
202.
by twothreeone 3y ago
I think the article got the very first number wrong.. 100 million neurons only makes 1/1000th of a human brain.
203.
by twothreeone 4y ago
The problem with your argument isn't lack of rationality, it's that it signals a lack of faith (destructive narrative). As others have mentioned, you can't "kill Mars" because who is going to be the hero then? You h
204.
by twothreeone 4y ago
Thanks, that's certainly helpful anecdotal evidence.. yeah it seems like there should be an "AllNorm" implementation that covers all cases and is just fast. I was wondering because I'm currently looking at math_group_nor
205.
by twothreeone 4y ago
> Right now in PyTorch, GroupNorm is 2x slower than BatchNorm How did you benchmark this? I think there are like 3 or 4 different GN implementations in PyTorch..