DougBTX

12 ms·

121.
by DougBTX 3y ago
A big area of research in NLP is translation between natural languages, I’d go so far as to say that “translation tool” is a better mental model than “token predictor”. For example, think of summarisation as translation between verbose lang
122.
by DougBTX 3y ago
It will lead to “AIO”, AI Optimisation, the set of techniques to smuggle content marketing through AI summaries
123.
by DougBTX 3y ago
> There's a difference between being in public and being available for public use. Books in a book shop are "in public", but only those out of copyright are available for public use Copying books can be fair use if it is s
124.
History of Philosophy without any gaps(historyofphilosophy.net)
72 points by DougBTX 3y ago|49 comments
125.
by DougBTX 3y ago
You can pass in any command line arguments you like, so it must be open source
126.
by DougBTX 3y ago
It probably changes from person to person, but while I find wrapping like that harder to read than seeing it all normally, for code it is still much better than scrolling back and forth for every line.
127.
by DougBTX 3y ago
> a separate charger After the last round of EU charger regs, it was just a cable!
128.
by DougBTX 3y ago
It is frameworks vs libraries. An agent decides when to call your code, or possibly someone else’s code.
129.
by DougBTX 3y ago
This is an interesting design decision between protobuf and cap’n proto, one has “repeated field of type X” while the other has “field of type list of X”. So one allows adding a repeated/optional field to a schema without re-encoding a
130.
by DougBTX 3y ago
You should try the intro to Physics lab work where students are asked to measure the length of a piece of string, and get a different result each time!
131.
by DougBTX 3y ago
If schema migrations take non-zero time and you want zero-downtime deployments, then the both service versions need to be compatible with either the new schema or the old schema, regardless of service size.
132.
by DougBTX 3y ago
> You need to have the discipline of a good CI/CD Good CD is even more important if there are more services to deploy
133.
by DougBTX 3y ago
This is mixing two meanings of confidence which could lead to confusion. The OP is using confidence to describe how high the per-token probability scores are, while you are talking about the confidence expressed in the tone of voice of the
134.
by DougBTX 3y ago
Even if the manufacturing process is the same, pre-80s ones may be better on average due to survivorship bias: the weaker ones broke already.
135.
by DougBTX 3y ago
FWIW, it is possible to pipe data faster than memcpy: https://codegolf.stackexchange.com/a/236630
136.
by DougBTX 3y ago
This article isn’t making a distinction between the interface provided by an abstraction and the implementation details of that abstraction, which I think causes it to come to the wrong conclusions. A bad abstraction is an interface which c
137.
by DougBTX 3y ago
Honestly I couldn’t tell you whether copying code out of Stack Overflow or out of ChatGPT is more legally suspect. For SO, you don’t know where the user got the code from either (wrote it themselves? Copied it out of their work repo? Copied
138.
by DougBTX 3y ago
Which book would be the equivalent to an Effective Modern Python?
139.
by DougBTX 3y ago
Previous discussions from last time they posted this: https://news.ycombinator.com/item?id=36856815 If they’re going to rank themselves so much higher than their competition, they might as well call that out up front and ex
140.
by DougBTX 3y ago
Love it: > The following built-in types inherit from Cool: Array Bool Complex Cool Duration Map FatRat...
141.
by DougBTX 3y ago
Rumour is that the first paper was from a disgruntled ex employee.
142.
by DougBTX 3y ago
The numbers measure different things, so hard to compare. One of the biggest sources of variation listed is the weather, while the first paint was done in a workshop (presumably where there is no influence from the weather). Also, a paintin
143.
by DougBTX 3y ago
It looks like those were chosen to get a opposite numbers for the comparison, +5% and -5%: > the expected value ... grows by 5% in every round ... but, crucially, a head and a tail experienced sequentially is different from two different
144.
by DougBTX 3y ago
The major difference with rust is that the core language has strong backwards compatibility guarantees, and the package manager supports installing multiple versions of transitive dependencies so packages can be updated incrementally.
145.
by DougBTX 3y ago
In theory, yes, however the vast majority of tables will have some form of unique ID in each record... so in practice, there’s usually no difference. But if it must work for all tables...
146.
by DougBTX 3y ago
This is C++ rather than C, but a substantial portion of PyTorch is written in C++, and they provide a C++ interface: https://pytorch.org/tutorials/advanced/cpp_frontend.html In other words, you can absolutely use
147.
by DougBTX 3y ago
Here's another way to look at it. The test set is an approximation for how the model will perform against production data, but the actual performance of the model is how it performs for actual end-users. So real _actual_ results are al
148.
by DougBTX 3y ago
There's info about that here: https://replicate.com/a16z-infra/llama13b-v2-chat > Run time and cost > Predictions run on Nvidia A100 (40GB) GPU hardware. Predictions typically complete within 9 seconds.
149.
by DougBTX 3y ago
The dimensionality is a challenge here - these are 2D islands in a, what, 2000D space? Imagine a cross section of a cylinder, that would look like an island if it went through the cylinder, but not along it. I wonder what it would look like
150.
by DougBTX 3y ago
This isn’t just a problem with custom types, it affects generic iteration types too. With a next() -> Option<T> method, an Iterator can iterate over any type of element.

More ›