jhrmnn

13 ms·

61.
by jhrmnn 2y ago
Even in cognitive tasks expressed via language, something like a memory feels necessary. At which point it’s not a LLM as in a generic language model. It would become a language model conditioned on the memory state.
62.
by jhrmnn 2y ago
On the other hand, can I pay them already now a few thousand dollars a year to remove my share? That would be the first actual carbon offsetting.
63.
by jhrmnn 2y ago
Interesting. To me music is a form of communication. So then obviously I care about who’s on the other side. I guess how people perceive this is quite subjective.
64.
by jhrmnn 2y ago
The article makes the point that the random variable map and the underlying sample distributions can change independently. That’s just not the case in quantum mechanics.
65.
by jhrmnn 2y ago
In quantum mechanics, the measurement and observation are two sides of the same coin, and the sample space is _defined_ by the random variable (observable) of interest, so it makes a little less sense to separate the two. (There is no hidde
66.
by jhrmnn 2y ago
In general, how does CPU utilization correlate with CPU power draw?
67.
by jhrmnn 2y ago
Indeed, it's along these lines: https://www.microsoft.com/en-us/research/blog/ai4science-to-...
68.
by jhrmnn 2y ago
Microsoft | Senior data engineer, AI for Science | On-site: Berlin, Germany Interested in driving the engineering side of a large-scale data operation in computational chemistry that unlocks AI for science? https://jobs.careers.m
69.
by jhrmnn 2y ago
It‘s just https://en.wikipedia.org/wiki/Baumol_effect , right? With AI, it now seems plausible at least on paper to use technology where it wasn‘t possible before in a hope that it will increase productivity, escaping t
70.
by jhrmnn 3y ago
Maybe it’s the difference between tracking a bug (abnormal operation) vs understanding behavior of a complex system (normal operation)?
71.
by jhrmnn 3y ago
Spotify should then be run as ad business—charge artists to let them advertise on Spotify.
72.
by jhrmnn 3y ago
Note that this is also how our own world works. Typical time scales of processes get faster as the length scales shorten. Somehow it feels inevitable for a dynamic multiscale structure.
73.
by jhrmnn 3y ago
The problem with this is that it could be a universal argument against any skepticism. So unless skeptics are always wrong, it doesn‘t really work
74.
by jhrmnn 3y ago
Interesting, exactly the same experience here. I’m now contemplating getting a desktop computer again once my laptop dies.
75.
by jhrmnn 3y ago
I’ve never remembered anything by memorization. Only at school and university for exams, and only to promptly forget it one week later. The stuff that I have remembered for years I learned organically by using it, as part of some mental pro
76.
by jhrmnn 3y ago
I think the intuition here is that an “intelligent” system would lean much more heavily on the heuristics than on the search, like humans. It’s hard to quantify in this case, but certainly in the case of chess, when engines were about as go
77.
by jhrmnn 3y ago
They also say “I left one training for a few more days and it reached 1500 ELO.” I find it quite likely the observed performance is largely limited by the spent compute.
78.
by jhrmnn 3y ago
What does „intuit“ mean to you then?
79.
by jhrmnn 3y ago
The problem is that that failure of the hypothetical scaling up is not priced in (an externality). In other words, individual choices are subsidized by the debt of the whole society. If we could agree on a tax that takes care of that, the
80.
by jhrmnn 3y ago
This just made me ponder again—where does the assumption that the Internet should allow unconstrained anonymity come from, other than that’s how it used to be for some time? The real world doesn’t allow that. It’s hard to remain anonymous i
81.
by jhrmnn 3y ago
In math specifically, one could easily imagine a reward signal from some automated theorem proving engine
82.
by jhrmnn 3y ago
Is it then similar to how utilizing a GPU fully is much harder than utilizing a CPU fully?
83.
by jhrmnn 3y ago
> I am seriously concerned that we'll end up killing ourselves before our scientific, social and cultural knowledge starts to compensate for what can only be described as recklessness with just about everything we touch. Beautifully
84.
by jhrmnn 3y ago
Condensed matter physics and materials chemistry certainly become inseparable in certain areas
85.
by jhrmnn 3y ago
I’d think that any harmful interaction with the physical world always affects also persons. In this case it at the least gravely endangers people
86.
by jhrmnn 3y ago
True. But then why do we almost never hold persons criminally accountable when they do harm through a company they control. Shouldn’t this be reciprocal?
87.
by jhrmnn 3y ago
This made me think—what if we postulated that companies cannot be victims of a crime. Only physical persons can. So if all you do is harm to a company, it cannot be a crime by definition, the act can only be treated as a company doing harm
88.
by jhrmnn 3y ago
That’s basically the claim of the “Blurry JPEG of the Web” article. I agree that to a large degree that’s probably what’s going on here. But it’s not all there is to it. Large language models apparently can build some world models. See Ot
89.
by jhrmnn 4y ago
Yeah, that was supposed to be just my personal feeling for the thing. And it’s not even a description of consciousness but rather of something that I imagine might have a capacity for consciousness. I don’t think we’ll really know until w
90.
by jhrmnn 4y ago
I’d think it’d need to be something like a diffusion model that’s continuously fed noise which it turns into outputs (thoughts). It’d also need to be getting continuous feedback and to have the ability to incorporate that feedback over all

More ›