lossolo

33 ms·

541.
by lossolo 2y ago
I wouldn’t consider it a major problem, especially with the coming robotic revolution. Even if the population declines by half, that would still leave 700 million people so twice the population of the U.S. According to predictions, the firs
542.
by lossolo 2y ago
> Low growth For 2024 prediction is 2.6% US and 4.8% China. I don't see how it's low compared to US. > high unemployment 5.1% China vs 4.1% USA > huge debt from infrastructure spending What do you mean by "huge"
543.
by lossolo 2y ago
In what world is the cost of 2.5 GB of RAM per 1 million connections an issue? Are you telling me there's a service in this world handling, for example, 100 million active connections, and they can't afford 250 GB of RAM? It'
544.
by lossolo 2y ago
If you are running $2 trillion deficits, then of course you will have trade deficits. You are an importer of goods and an exporter of USD. The problem will arise when your debt becomes unsustainable and alternatives to the USD emerge for se
545.
by lossolo 2y ago
CMP + JMP have a branch, which can be mispredicted and then you may pay 5-10x more than for cmp + cmov.
546.
by lossolo 2y ago
Well, it's your site, so you can do what you want with it, but I don't believe what you just wrote is logical at all. Sometimes, you just want to see, in general, how one language compares to another when one uses intrinsics and t
547.
by lossolo 2y ago
Cool! I didn’t see it before, maybe because I mostly use the language vs. language feature. There is no such section there, but it would be very helpful IMO, instead of clicking every solution to check which one is intrinsics free.
548.
by lossolo 2y ago
You should really create a way to filter solutions with SIMD intrinsics and without them.
549.
by lossolo 2y ago
I don't know why, but your message moved me. Thanks for sharing! I'm happy for your daughter and wish her many friends, as many as she wants.
550.
by lossolo 2y ago
> I search around, and inevitably end up with "by combining clickhouse and Kafka" Those are probably some old sources of knowledge. You need to use Kafka if you want it to handle batching for you. But Clickhouse can handle batc
551.
by lossolo 2y ago
Livebench updated https://livebench.ai Model | Global | Reasoning | Coding | Math | Data | Language | IF ------------------------------|---------|-----------|---------|---------|---------|-
552.
by lossolo 2y ago
What if they just sit and talk? What is the description of this? What if only part of the communication is relevant? What if it's not relevant at all because they reacted to atmospheric changes? Or electromagnetic signals, that can&#x2
553.
by lossolo 2y ago
Their architecture is not fault-tolerant. If one server goes down and the whole system goes down, then it was not designed to be fault-tolerant. I have been running fault-tolerant systems spread across multiple dedicated servers (inside sys
554.
by lossolo 2y ago
We could train a transformer that could predict the next token, whether it's the next sound from one animal or a sound from another animal replying to it. However, we wouldn't understand the majority of what it means, except for t
555.
by lossolo 2y ago
> good strategy for them. For everyone, besides OpenAI etc.
556.
by lossolo 2y ago
And Meta is using a lot of GPUs for offline ML and online ML features on Instagram, FB etc. So nothing is "wasted".
557.
by lossolo 2y ago
Pruning takes forever in Restic, which is why we migrated away from it.
558.
by lossolo 2y ago
Knowledge gained from experience that is not included in documentation is also significant part of SO. For example "This library will not work with service Y because of X, they do not support feature Y, as I discovered when I tried to
559.
by lossolo 2y ago
> I'd argue Rust mpsc channels are actually more powerful than Golang's Do they support Go select functionality, buffered channels etc without using 3rd party libraries?
560.
by lossolo 2y ago
You must have done something really wrong if they don’t even want to resolve this with you. Normally, you would receive an abuse notice from them — we did many times with one of the services I worked on. I guess your service is somehow con
561.
by lossolo 2y ago
That's a very naive view of the world. There are predators waiting for deregulation to flood the market with snake oil and profit from everyone. By the time the FDA can launch any campaign, people will already have been scammed or even
562.
by lossolo 2y ago
The FDA exists to prevent us from repeating history[1] and to minimize the risk of people being killed, developing tumors, or buying placebos instead of real cures. 1. https://en.wikipedia.org/wiki/Snake_oil
563.
by lossolo 2y ago
Bob McGrew, head of research just quit too. "I just shared this with OpenAI" https://x.com/bobmcgrewai/status/1839099787423134051 Barret Zoph, VP Research (Post-Training) "I posted this note to Open
564.
by lossolo 2y ago
> For comparison, GPT-4o is currently $5/million input and $15/million output and Claude 3.5 Sonnet is $3/million input and $15/million output. Google is the only one of the three that has its own data centers and cus
565.
by lossolo 2y ago
Nuclear plants often operate at 90%+ capacity factors, while solar can range from 10-25% depending on location. So you might need 4-9 times more solar capacity to match total energy output. So depending on the region, you might need 4-9 gig
566.
by lossolo 2y ago
I think you are overly optimistic, this is an older process, which means that next iphones will need to source their CPUs from TSMC fabs in Asia, not from US. There also will be no knowledge transfer, that was not part of the deal. It'
567.
by lossolo 2y ago
These workers didn't create the fab equipment and don't know how to design or create the machinery used there. They also don't have access to the software source code. Most US workers hold maintenance or managerial positions,
568.
China Is Rapidly Becoming a Leading Innovator in Advanced Industries(itif.org)
63 points by lossolo 2y ago|161 comments
569.
by lossolo 2y ago
> Visiting the same number of nodes, but because the nodes are referenced from an array, when you load `a1` you're [probably] also going to load `a2` into the cache. Yep, the problem with a linked list is that you don't know th
570.
by lossolo 2y ago
It seems like o1 is a lot worse than Claude on coding tasks https://livebench.ai

More ›