frud

no profile record

コメント

2 年前議論

I think that, more generally, intelligent people don't get arrested for crimes for several reasons. First, because they are smarter, they just don't get themselves into jams where murdering someone seems like the best way to get out of the jam. Second, because they are more successful they have more to lose in terms of wealth, happiness, good living situation, so they risk more when choosing crime, so they're less likely to choose it. Only thirdly is actual proficiency in the planning and execution of the crime.

2 年前議論

The raw words written to the drive are actually re-encoded into slightly larger codewords with nice properties like not having too many zero or one bits in a row, and error detection/correction.

Plus I think that the 0/1 bits are not encoded as "no magnetism"/"some magnetism", but instead as "north magnetism"/"south magnetism" since magnetic fields have a direction.

And I don't think the magnetic fields on the platters have any appreciable effect on the head besides the electromagnetic effects at the sensor.

2 年前議論

The game goes through several phase changes as you progress.

2 年前議論

Any form that doesn't have loose fibrils, microparticles, or dust.

I'd compare it to wood from walnut trees, which is perfectly safe to handle and use in dining tables, etc., but dust from its' woodworking is toxic.

2 年前議論

I used to work for a company that sold factory automation technology, and had hundreds of manuals for all the products they sold. In the front matter of every manual was a disclaimer that nothing in the manual was warranted to be true. This was automation equipment running things like steel mills, factories, petroleum plants, where failures could result in many deaths and grave financial losses.

2 年前議論

The real story here is that Air Canada's lawyers argued, among other things, that the chatbot was a separate and independent legal entity from Air Canada and therefore Air Canada was not obligated to honor the made up policy.

In other words, this was possibly the first historical argument made in a court that AI's are sentient and not automated chattel.

2 年前議論

For basically forever compilers and languages have been designed with a total ordering on operator precedences. The Yacc compiler generator tool maps operator precedence to integers, as do most handwritten compiler parsers.

A total ordering has a definite answer to the question "is x greater than, equal to, or less than y?" for all x and y. A partial ordering will answer "I don't know" for some x and y.

It's quite possible to implement operator precedence using a partial ordering. When the parser has to resolve precedence between two operators and their relationship is not defined, throw an "ambiguous precedence" error.

You can implement the partial ordering by putting all the operators into a DAG of sets of operators. If two operators are in the same set, they have equal precedence. If there is a path through the DAG from one operator to the other, that defines the precedence relation. Else there is no relation.

Say "*" is defined to have a higher precedence than "+", and they both have an undefined precedence relation with "&". Then "1 + 2 * 3" should compile into "1 + (2*3)", but "1 + 2 & 3" should throw an "ambiguous precedence" syntax error.

2 年前議論

Even if it's stable in a game theory sense, it depends on the political class being partitioned into exactly two parties. Otherwise the two biggest parties would have an incentive to smother all smaller competitors.

I think the answer has to lie in an algorithmic solution dependent on a high-resolution population density map.

2 年前議論

Fish + fire = sushi

2 年前議論

Wow, cool. A paper that costs $20 to read. I'll be sure to hop on that.

2 年前議論

That's quite an injection if it can cure six children at once.

3 年前議論

I have an idea. Why don't we give new experimental AI systems their own off-grid nuclear power plants so they can't be switched off. There's no way that could go wrong.

3 年前議論

Some are made from dissimilar metals. https://www.coin-database.com/series/japan-47-prefectures-co...

3 年前議論

After 30 seconds of furious googling: https://developer.nvidia.com/cuda-fortran

3 年前議論

One big thing I remember is that it's illegal to have multiple array arguments reference overlapping storage, so storage regions can't alias. This means functions can be optimized more aggressively.

My fingers can't type Fortran anymore, so I'm going to use C as an example.

Imagine you have this function:

    /* Add entries in arg1 to arg2, put result in result */
    void add_vec(int arg1[], int arg2[], int result[], int length) {
        for(int i = 0; i < length; i++) { 
            result[i] = arg1[i] + arg2[i];
        }
    }

In C, it's perfectly legal to call this like so:

    int array[101];
    // ...pretend array is initialized...
    // for the first 100 elements, set a[i] = a[i] + a[i+1]
    add_vec(array, array+1, array);

The C compiler has no choice but to iterate through the array one element at a time, doing things in the exact order that the source code spells out. No loop unrrolling or vectorization can occur.

In Fortran, the compiler knows that the arrays are not aliased, so it can do much more reordering, unrolling, and vectorization to speed this code up.

3 年前議論

From the link: "The same exact GPU model and program versions must be used for compression and decompression". So it seems to me that you need a standardized and deterministic LLM execution model before this becomes useful.