hexaga

13 ms·

91.
by hexaga 3y ago
Ah, I see. You are unfamiliar with how SOTA language models are actually trained in current day, what RLHF is in detail, and how it differs from other training phases. I'd encourage you to read through the GPT3 [0] and llama2 [1] paper
92.
by hexaga 3y ago
Are you trying to argue that RLHF is the same as the algorithm used to pretrain and finetune models, because they are both searches in model space that reward good answers and penalize bad answers? If you only look at the very high level su
93.
by hexaga 3y ago
> [...] modulated the logic behind the Weighting methods. I fail to see how this is not literally the case. The 'logic behind the weighting methods' refers to something like cross-entropy loss minimization via AdamW (or insert
94.
by hexaga 3y ago
Speaking frankly, this seems like a misread to me. Ascribing an author's intent to the output of a program they wrote is not the same as believing they manually performed the actions of the program.
95.
by hexaga 3y ago
It's fairly clear that they understand that much. The objection is that the word 'learn' is generally used in reference to how humans learn. There is a lot implied in this usage, and a long history of use by the general pop
96.
by hexaga 3y ago
> don’t your unspoken thoughts take form and exist as language in your mind? Not really. More often than not my thoughts take form as sense impressions that aren't readily translatable into language. A momentary discomfort making me
97.
by hexaga 3y ago
Lead added to fuel is in the form of an organic compound, so I'd expect the opposite effect. Obviously there's more distribution post combustion, but from a bio-availability standpoint spraying tetraethyl lead around in equivalent
98.
by hexaga 3y ago
> Prove that the dynamic instability that causes a speed wobble is the same thing that causes a high side. This is a broader claim than the one I made. > A high side happens when the rear tire breaks traction at lean, and is no longer
99.
by hexaga 3y ago
Prove what, specifically? I'm honestly unsure which part is a controversial enough claim as to require sources. It seems trivial to me that in both cases a perpendicular force applied at the road acts to fling you, and that said force
100.
by hexaga 3y ago
Well, yes. How 'high' the side is depends on many factors, but it's a gradient. A death wobble when going in a straight line happens to be relatively weak with regard to how much force the rider is ejected with should it full
101.
by hexaga 3y ago
I don't disagree, but I also didn't mention low sides at all so I'm unsure of what you're trying to express here. My point is that a high side is what happens if you lose and regain rear traction, landing you in an unsta
102.
by hexaga 3y ago
They're effectively the same process occurring to different degree. A high side is what you get when you arrive at a point of unrecoverable inflection in such a wobble.
103.
by hexaga 6y ago
I don't know much js, but it seems fairly straightforward. Here's some comments: // create an element with 'tag' const elem = document.createElement(tag) // copy properties into the new
104.
by hexaga 6y ago
It is difficult to overstate how strongly I want to download and poke through that file, even knowing it's a bad idea.
105.
by hexaga 6y ago
Imo the problem is that 'high' is a relative term. It's calibrated according to whatever the commonly used abstraction level is for the term-speaker. The other problem is that level doesn't really mean anything, or it me
106.
by hexaga 6y ago
I don't think that's necessarily true, because I don't think emotions are 'what consciousness is made of'. For example, I'm conscious of color. Imo consciousness is an ability to witness subjectively, in a very
107.
by hexaga 6y ago
True enough.
108.
by hexaga 6y ago
Infiniband and friends aren't cpu interconnects, they're alternatives to ethernet. Network latency is important and it's not that hard to write code whose performance is bounded by io latency/throughput.
109.
by hexaga 6y ago
From my perspective, thoughts are the sensory experience of 'my mind moving'. Similar to arm movements - when you move your arm, where does the motion come from? I'm reminded of a time in my childhood when I realized that I c
110.
by hexaga 7y ago
That is, only when said memory location is accessed mutably. Immutable references are perfectly ok to follow concurrently.
111.
by hexaga 7y ago
It works the same way with actual (non-contrived) problems. You'd identify a property you want to test, and assert that it holds for randomized inputs. Say you have a CRUD api (TheTrackerOfWidgets), and you want to test a create Widget
112.
by hexaga 7y ago
IME, random data is extremely useful. Consider fuzzing, property testing ala quickcheck, automated fault injections, etc. The key thing to make it work well is to keep track of your prng seeds to make things deterministic. If you find that
113.
by hexaga 7y ago
The terminology and practice comes from the time when drones were primarily built by hobbyists. Rotors are dangerous for fleshy things, and it's easy to leave controls in an unsafe state (throttles especially, they usually don't h
114.
by hexaga 7y ago
> can we agree that the copy would believe that it's you? Absolutely. I think this is a presupposition of the entire scenario. Practically speaking, the only thing that is in question is whether the original's subjective experi
115.
by hexaga 7y ago
[citation needed] Somewhat tongue in cheek, but the point remains. What leads you to believe this? As another reply touches on, there's a straightforward way to verify whether that is the case (for oneself, and not an external observer