BobbyJo

17 ms·

331.
by BobbyJo 1y ago
What risk/return are they expecting employees to accept? What risk/return are they willing to accept? Any advantage they give themselves over the employee, to me, means they don't care.
332.
by BobbyJo 1y ago
The core issue there isn't that the LLM isn't building internal models to represent its world, it's that its world is limited to tokens. Anything not represented in tokens, or token relationships, can't be modeled by the
333.
by BobbyJo 1y ago
Which is an even better argument when you look at how yields have been behaving. AI is sucking the air out of the room.
334.
by BobbyJo 1y ago
A times table test like we all took in grade school. A test applying Newton's method. Any test where the calculator can do the work instead of you, and a TI-89 can do quite a bit of work.
335.
by BobbyJo 1y ago
I feel like you're glossing over the contexts of "acceptable" and "questionable". The tool itself is not questionable or acceptable, it becomes questionable or acceptable depending on the usage. A pencil and paper c
336.
by BobbyJo 1y ago
This ignores that the internet exists now, and ideas spread millions of times faster. No other tech has needed that much time to mature and find use since.
337.
by BobbyJo 1y ago
Bitcoin was created 17 years ago. The "the use cases are coming" argument doesn't work anymore.
338.
by BobbyJo 1y ago
I don't think its any less democratic than any other government organization run by appointees. Manufacturing's decimation had far more to do with globalization than banking regulation IMO.
339.
by BobbyJo 1y ago
The Federal Reserve is an example of a well run and focused organization, because they have a very short, very simple mandate: maximize employment and ensure price stability. Everything they do is in service to those two goals and every emp
340.
by BobbyJo 1y ago
Putting the first people on the moon, the first people on any foreign celestial body, as a show of power during the cold war isn't what I'd consider "token". Putting people on the moon today is a lot less substantive, an
341.
by BobbyJo 1y ago
Not anymore, but people were just doing whatever like 100 years ago.
342.
by BobbyJo 1y ago
> Tariffs are not now, have never, and will never lead to an increase in domestic production. I disagree. Its a pretty straightforward market force. > The US has done this multiple times already and the result is always the same. Pers
343.
by BobbyJo 1y ago
> Tariffs are a demand-side change, there's also subsidies for supply-side. Yes, but subsidies are typically more complex to implement than tariffs. Imports are already controlled via customs, so you're piggybacking off existin
344.
by BobbyJo 1y ago
> I'd call the first part utterly counter-productive and stupid. I don't think there is a better method of onshoring than tariffs, personally. You change the cost of offshoring and see what the market does. IMO its the least co
345.
by BobbyJo 1y ago
I mean, i wouldn't call 300% increase in collected tariffs and a tax break for onshore capital investment lip service...
346.
by BobbyJo 1y ago
> I read his point as, "if we have an obligation to fix other people's problems, we shouldn't be subject to complaints when our foreign adventures go wrong". That's a bad faith reading imo. They said "action
347.
by BobbyJo 1y ago
> But is it really so strange to you to think that the richest and most powerful country might have some role to play in international problems that arise from comparative poverty? I think this is a mischaracterization of parent's p
348.
by BobbyJo 1y ago
It's his hackernews username.
349.
by BobbyJo 1y ago
I think you missed the point of parent's comment. > All life on Earth uses water to transport chemicals and ions across its cells. Period. It's one immutable requirement of life on Earth. They were pondering this. With billions
350.
by BobbyJo 1y ago
> I cant see any reasonable reason, in a broader context, this would have a meaningful impact. A significant amount of software dev employment is in startups. Companies that are spending on development, but aren't making much money
351.
by BobbyJo 1y ago
100%. Most of the planet is cheaper than the US, and has been for decades. That being the case, how are there so many knowledge workers here still?
352.
by BobbyJo 1y ago
This ignores the other financial and non-financial costs of offshoring: legal, cultural, temporal... a lot of the time, those close the gap. On paper, offshoring has made sense the entire time, and yet here we are in 2025 and companies stil
353.
by BobbyJo 1y ago
What do you base that belief on?
354.
by BobbyJo 1y ago
Original passing didn't cause much trouble because the provision didn't take effect til 4 years later.
355.
by BobbyJo 1y ago
That's less a versioning of software and more a versioning of AI's role in software. None -> Partial -> Total. Its a valid scale with regard to AI's role specifically, but I think Karpathy was intending to make a point
356.
by BobbyJo 1y ago
The versioning makes sense to me. Software has a cycle where a new tool is created to solve a problem, and the problem winds up being meaty enough, and the tool effective enough, that the exploration of the problem space the tool unlocks is
357.
by BobbyJo 1y ago
> The problem isn't that this is necessarily incorrect, it is how do you find this mythical figurehead? I think, historically, the problem has also been: what happens when they die? Then you need lightning to strike twice.
358.
by BobbyJo 1y ago
That was my point. The fact that we call people wearing CGI suits hopping around green room, and people on stage playing a character for a crowd, both acting doesn't account for the fact that doing one doesn't mean you can do the
359.
by BobbyJo 1y ago
Sitting in a moving car and sitting on a moving horse are both called "riding", but I think we can all appreciate how useless it is to equate the two.
360.
by BobbyJo 1y ago
I worked on CUDA and OpenCL in the 2010-2014 timeframe, well before buttcoin and LLMs were profit centers, and Nvidia was already well ahead in the "GPUs as general compute" area. Literally everyone doing highly parallel HPC wante

More ›