the8472

30 ms·

481.
by the8472 3y ago
If anything medical devices, especially implants, should be modifiable. Otherwise you're stuck with an unmaintainable device (possibly in your body) once the manufacturer goes out of business or the support window ends or whatever. R
482.
by the8472 3y ago
That only prevents analog copy degradation. It doesn't give you reproducibility. Reproducibility means running the same process twice with the same inputs and getting the same outputs. E.g. to later prove that something came from an LL
483.
by the8472 3y ago
To my knowledge it's still the harnesses that perform those actions, not baked into the weights (other than perhaps special tokens denoting that an action should be taken)
484.
by the8472 3y ago
Model weights are data , they shouldn't contain general purpose code with access to the host environment. Sure, any library/tool that harnesses models might contain malicious code. The models themselves should not be able to. Thi
485.
by the8472 3y ago
Afaik there's no guarantee that the compiler won't perform such changes. Even if it happens to generate the expected instructions today it could change with the next compiler release. I think on some x86 cpu tuning levels this can
486.
by the8472 3y ago
The issue with writing crypto code in anything above assembly is that there's a risk that optimizing compilers could turn your finely crafted constant time, constant power code into something which is not that.
487.
by the8472 3y ago
If the game leads to negative-sum outcomes it is the game that needs to be fixed, not the players.
488.
by the8472 3y ago
Exact copies aren't just about precision but also about reproducibility.
489.
by the8472 3y ago
Bits are copyable without data loss. Analog properties of individual transistors are less so.
490.
by the8472 3y ago
What does it mean for future hardware if it's not using floating point matrix multiplication units?
491.
by the8472 3y ago
I think the argument is that if land value were ~zero you could put the new panels next to the old ones that are still producing power.
492.
by the8472 3y ago
At least some of the language features motivated by async are also useful in other places, e.g. RFCs 3425 and 2033 And the "bit more wasteful" part is a non-starter because people want to use async in embedded contexts.
493.
by the8472 3y ago
Don't worry, AlphaJudge will provide swift justice at scale.
494.
by the8472 3y ago
At least in the workstation segment cloud doesn't compete. We use Threadrippers + A6000 GPUs at work. Getting the equivalent datacenter-type GPUs and EPYC processors is more expensive, even after accounting for IT and utilization.
495.
by the8472 3y ago
If you can do syscalls on behalf of another process, sure. But that's more than "just" CreateProcess.
496.
by the8472 3y ago
The thing is not about whether it's theoretically possible to configure it but more about evolving OS APIs and all those APIs having to be mirrored in the process construction API. E.g. all the security/namespacing stuff.
497.
by the8472 3y ago
The setup part needs to be turing-complete. A simple struct won't do. What you're talking about sounds more like posix_spawn. io_uring_spawn on the other hand would allow that because the parent process could execute the logic bet
498.
by the8472 3y ago
fork+exec is great in so far as it lets you do arbitrarily complex process setup between those syscalls. APIs like posix_spawn are far more restrictive. The issue is the overhead and the restricted post-fork environment in a multi-threaded
499.
by the8472 3y ago
Individual DHT nodes should only see a trickle of packets. A few kilobytes per second. Even less per remote IP. So they can set fairly strict rate limits. And ISPs are in a much better position to solve this problem anyway. They should ask
500.
by the8472 3y ago
> Which of these methods is not obtaining the address of a node in a swarm, or hitting a tracker for a list of nodes in a swarm? The "find bootstrap nodes via a bunch of DNS records, then persist long-lived contacts locally" ap
501.
by the8472 3y ago
That doesn't make sense because those are concurrent processes, not serial ones.
502.
by the8472 3y ago
> You have to be participating in a torrent swarm in order to bootstrap yourself into DHT at all. This is not correct. > Bittorrent's DHT is not a network independent of torrent swarms. This is also wrong. People have used the DH
503.
by the8472 3y ago
We had a discussion here https://github.com/bitmagnet-io/bitmagnet/issues/11 With the related changes bitmagnet shouldn't have the blatant misbehavior of magnetico (anymore). Though I haven't looked
504.
by the8472 3y ago
That's not correct. Torrent swarms and the DHT are separate. Each torrent basically forms its own small network of TCP connections to exchange data specific to that one torrent. While the DHT is a network shared by all clients that spe
505.
by the8472 3y ago
This is hardly speculative. RL-trained agenty models just don't grab the headlines. They have been growing over years. Think AlphaGo, OpenAI Five, This Model Without A Catching Name[0], AlphaCode are all in that category. And they&#x27
506.
by the8472 3y ago
I think that's a too-narrow focus on LLMs. Of course they're not good at problem-solving because that's not what they're optimized for. That they can approximate any problem-solving at all is already impressive. Consider
507.
by the8472 3y ago
Not really. We have way more data points than one on AI development. It has been incremental progress for more than a decade.
508.
by the8472 3y ago
> I'm not going to worry too much about my career being snuffed out by AI until AI actually exists. Can you explain what underpins this position? Is it about timelines, likelihoods, not taking preparative actions in general or what?
509.
by the8472 3y ago
MRAM is available as a niche product, it just hasn't achieved the promise of being cheap and dense as dram, fast as sram, nonvolatile and higher write endurance than flash at the same time. It ticks maybe two or three of those boxes, w
510.
by the8472 3y ago
"by this technology" does a lot of heavy lifting. Look at the pace of AI development and extrapolate 10 years.

More ›