ogogmad

23 ms·

421.
by ogogmad 3y ago
You're partially right. It's obvious that the solution is to combine traditional programming with AI, using traditional programming wherever possible because it's greener. Assuming you want things to turn out well in every po
422.
by ogogmad 3y ago
Might be a silly question, but if you want determinism in this, why don't you get the LLM to write the deterministic code, and use that instead? Interesting experiment, though! In fact, what about a hybrid of what you're doing now
423.
by ogogmad 3y ago
Can be used to buy things illegally. Think ADHD meds or sanctions breaking. In an ideal world, the ADHD meds where I live should be made available by GP prescription to reduce the backlog. Right now, it takes years to get a diagnosis free o
424.
by ogogmad 3y ago
It is. That said, I know a bunch of people who use it to buy ADHD medication, who can't otherwise afford to get a prescription in a timely manner. And I don't even know that many people! I suppose there's many downsides to en
425.
by ogogmad 3y ago
It does make sense. An explanation of NNs must help people choose hyperparameter values ***based on*** the data.
426.
by ogogmad 3y ago
This is a misunderstanding. You're right that you can always find some weights that approximate any reasonable (that is, continuous) mathematical function. The problem is, you have to find those weights in a small amount of computation
427.
by ogogmad 3y ago
One model? So there may be models where it doesn't happen. Or perhaps that very same model makes other, verifiably wrong predictions.
428.
by ogogmad 3y ago
> You simply can't get past what Gödel and Rice proved with current technology. Not this again. Those theorems tell you nothing about your concerns. The worst case of a problem is not equal to its usual case.
429.
by ogogmad 3y ago
Might be relevant: https://www.nature.com/articles/s41586-023-06924-6 Mathematical discoveries from program search with large language models
430.
by ogogmad 3y ago
SGD keeps the matrices small, I think.
431.
by ogogmad 3y ago
You can do a mid-point construction using just a straight edge and some means of drawing parallels. Good example of projective geometry. ;)
432.
by ogogmad 3y ago
How flexible and performant is that in different situations?
433.
by ogogmad 3y ago
If you read what I wrote, you'd see I never claimed originality. Anyway, that's cool. I'll need to check how much abstract algebra stuff they use. Using semirings (uh, rigs) alone isn't impressive. Do they consider semir
434.
by ogogmad 3y ago
I suspect a lot of graph theory can be reduced to abstract algebra. You consider matrices over lots of different scalar types. The scalar types should be: - Rigs (rings without negation), - idempotent (that is, where x + x = x for all x), -
435.
by ogogmad 3y ago
Minor nitpick: The title is confusing. The first word should be substituted with "Machine-Learning" or "Statistical-Learning". Ideally, the author of the piece should do that at some point.
436.
by ogogmad 3y ago
This might be insultingly simplistic, but I always thought the phrase "conservation of information" just meant that the time-evolution operator in quantum mechanics was unitary. Unitary mappings are always bijective functions - so
437.
by ogogmad 3y ago
You might want the same piece of code to work correctly under different numerical types. Answer: Op overloading + generics. E.g. Dual numbers for autodiff; 2x2 matrices of quaternions for texture mapping; matrices of differing floating poin
438.
by ogogmad 3y ago
Operator overloading is about syntax and type safety. Brainfuck is Turing complete.
439.
by ogogmad 3y ago
Autodiff requires interpreting the same piece of code using "dual numbers" that would normally expect "real numbers". This needs operator overloading combined with generics.
440.
by ogogmad 3y ago
RETVRN TO GEOMETRY https://en.wikipedia.org/wiki/Planar_ternary_ring If you look at the section to do with geometry, you'll see that Fused Multiply-Add T(a,m,c) is the value of y given x where y=mx+c, where the la
441.
by ogogmad 3y ago
The world seems like it's moving to ML. That needs matrices, complex numbers, autodiff, different fixed/floating point types. All four of these need operator overloading.
442.
by ogogmad 3y ago
Lack of support for operator overloading is a bit sad. As a result, Go can't get used in numerical applications, unless you like writing `matrix_plus(A, B)` instead of `A + B`. Is there a programming language that has a "simple&qu
443.
by ogogmad 3y ago
OK, I've noticed that the pseudo-code above is vectorised, and so there's no recursion. The SSM function is actually described at the start of the paper, and an efficient hardware-aware implementation is suggested in section 3: h
444.
by ogogmad 3y ago
That might explain the motivation for why the Δ variable is used and varied; but not the "Selectivity", which the article says is expressed by how the matrices B and C vary while consuming input. Something I've noticed is tha
445.
by ogogmad 3y ago
To be fair, they said at the end that their path-finder is not yet competitive with the SOTA. The paper tests how well a transformer does at predicting an execution trace (like in a JIT compiler) and whether this can help improve heuristics
446.
by ogogmad 3y ago
Floating cities?
447.
by ogogmad 3y ago
> They'll write: "look, people used to think we'd have blockchain-based economies and that AI was about to take over the world! How silly they were." I get the wider point of your comment -- which is that most people
448.
by ogogmad 3y ago
> facilitates money laundering, the sale of illegal goods Some illegal goods should perhaps be legal. The cost in time and money of certain diagnoses and drug prescriptions is too high at the moment, and some of us can't put our liv
449.
by ogogmad 3y ago
It seems like a very long input to an LLM can act as additional training data. A possible future is that LLMs will be used to bootstrap AIs that will be "trained" by feeding them a giant prompt at the start. Might end the era of &
450.
by ogogmad 3y ago
One possible alternative is to combine PROGN with LET, which you might then call BLOCK, with the macro definition of BLOCK looking for `VAR`s. The VAR syntax wouldn't make sense anywhere else. More exactly: (block (var first-

More ›