omolobo
omolobo on Hacker News
So what? Why should the application talk over the Internet to begin with? And why isn't that functionality off by default under a settings option that clearly warns the user of the consequences? I think you're missing the forest for the trees here.
And the claims that this is good privacy/security are not at all obvious either. And who are those third-parties anyway? Did you verify each one of them?
How is this comment down-voted? Makes absolutely no sense. These are simple facts stated here.
> "I think he was too naive and too innocent to understand this dirty corporate world."
Clearly, if you work at OpenAI, you must be either naive or psychopathic.
I wonder what led him to the extremes of suicide, though. Clearly he must have been influenced and possibly coerced.
So the first email was from a Protonmail account, and the second one was spoofed and obviously had incorrect headers. Are you saying Youtube doesn't check for these when processing take-down requests?
Yes. These limitations are not theoretical at all. The author touches on compositionality --- how a problem/program be decomposed into smaller, orthogonal problems/programs, reasoned about and tested separately, and then abstracted away in an interface that hides the implementation details. This is the essence of programming and software engineering at large, whether you're programming in assembly, Java, or Haskell. To divide and conquer so that we can fit an isolated aspect of the program in brain cache so that we can reason about it. This is a fundamental limitation and will not change until the year 40,000 when we have Space Marines.
A neural network, conversely, is a big ball of mud. Impossible to reason about and to test except for whole-system, end-to-end testing, which is impossible to do exhaustively because of the size of the state space. It is, by design, unexplainable and untestable, and therefore unreliable. It's why you use globals in C only judiciously. (I am just rephrasing the article here, not saying anything new.)
And the evidence that it causes practical problems to usefulness is already out there; "hallucinations" are simply errors, just that corporate PR likes to pretend that it's a "feature" and not a bug. This is delusional. A society seeking digitalization should run away from this level of stupidity.
You are clearly delusional. You have no idea how software engineering works. Sorry, but have you read the article? The author has decades of research in formal methods and establishes very good points about the fundamental limitations of NNs. He also lays out where formal methods fall short and how "informal" ones have succeeded in pushing the field of software engineering forward. Please stop parroting corporate PR, especially when you haven't read the article.
You are putting it too lightly. From the author:
> When all its techniques are based on testing, AI safety is an intellectually dishonest enterprise.
Or, like the old Internet adage says, "never go full retard."
So what have you gained in the process, other than wasting significantly higher amounts of energy in the form of heat and other emissions? It is nothing like software engineering; clearly you speak out of ignorance.
And then you say you attach 0 credence whatever, but you give no reasons for why others should buy your points. You don't really seem to have much of a point, anyway.
> I would call ‘LLM-functionalism’: the idea that a natural language description of the required functionality fed to an LLM, possibly with some prompt engineering, establishes a meaningful implementation of the functionality.
My boy. More people need common sense like this talked into them.
It should also be noted that those elements are dynamic and update in pseudo-real-time. In case my boy is still trying to make sense of them in the context of the article, lol.
Meanwhile, every YC startup these days is some unimaginative variant of "let's automate this with LLMs".
Humanity racing towards maximum idiocy.
It probably works better when the reflected ray is almost tangent to the surface. But that should be an epsilon case.
It's a mega-kernel, so you'll get poor occupancy past the first bounce. A better strategy is to shoot, sort, and repeat, which then also allows you to squeeze in an adaptive sampler in the middle.
> // No idea where negative values come from :(
I don't know, but:
> newRay.origin += sign(dot(newRay.direction, geometryNormal)) * geometryNormal * 1e-4;
The new origin should be along the reflected ray, not along the direction of the normal. This line basically adds the normal (with a sign) to the origin (intersection point), which seems odd.
Poor's man way to find where the negatives come from is to max(0,...) stuff until you find it.
As far as I know, it is popular in bioinformatics because of
https://bioperl.org/ Universities use that often in their curriculum.
It somewhat resembles the encircled lambda in LOL.