espe

8 ms·

1.
by espe 3y ago
and the other way round - have it built in for self-fuzzing and healing the infra.
2.
by espe 3y ago
yes that sounds rational. central bankers are humans with a target on their back though.
3.
by espe 3y ago
the "perfectly rational markets" expect the central banks to lower interest rates. which actually assumes that something economically worse happens in the meantime because as it is now, the fed is _not_ lowering.
4.
by espe 3y ago
it's a play on n-wave feminism and how white lady feminism was/is different than that of the "subalterns".
5.
by espe 3y ago
yes, it's the nvms one.
6.
by espe 3y ago
they will sell. the big guys are gobbling up _anything_ they can get their hands on.
7.
by espe 3y ago
> My fear is that eventually you'll start seeing government services where identity and auth are handed off to private companies like Google and Apple. Imagine having your real identity tied to an attestation by one of those compani
8.
by espe 3y ago
yes i'd second deepseek-coder and check out wingman. it's a relatively new extension, snappy and with good ux.
9.
by espe 3y ago
yes. from the perspective of anthropology, humans using tools are cyborgs. and language and texts are ai. or "alien intelligence" as burroughs (and a few others) would have put it.
10.
by espe 3y ago
you both describe valid approaches, similar to a breadth- or depth-first search. one will get into deep waters with "some great big system", regardless of the specifics.
11.
by espe 3y ago
this looks really cool. might be perfect for an old odroid. hope that doesn't stray too far from the happy paths.
12.
by espe 3y ago
true. i call them "prompt stimuli". prompts can give structure and context and whatnot but a dash of original domain gives the edge.
13.
by espe 3y ago
source: a law prof specialized in the matter. but yeah, ymmv.
14.
by espe 3y ago
just remember that feedback loop implicates us, our language, psyche, culture. i guess it will be a challenge _not_ to unwittingly converge with LLMs.
15.
by espe 3y ago
i think it's safe to assume most "copyrights" on model weights will be annulled, on either side of the pond.
16.
by espe 3y ago
thanks for the clarification! i'd still be in for a fossil t-shirt :)
17.
by espe 3y ago
> containing the downsides of Silicon Valley’s latest export through which political lense do i have to look in order to put it like this? there is decades of european (continuing the bloc thinking) research and braindrain (and wasted op
18.
by espe 3y ago
yes the baseband blobs are still underappreciated.
19.
by espe 3y ago
interesting question. anyone know of a paper that looks at power law distributions in llm vs. human text?
20.
by espe 3y ago
maybe check out https://flet.dev
21.
by espe 3y ago
well, in a way language itself is a semantic timebomb.
22.
by espe 3y ago
python3 -m llama_cpp.server --model /path/to/model.gguf
23.
by espe 3y ago
nice. 1995 ibm aptiva keyboard here. it’s still the same as on day 1.
24.
by espe 3y ago
not sure it helps but - soulseek is pretty alive
25.
by espe 3y ago
that should also apply to vscode extensions and most other plugin systems. and this seems to get worse? one area where apple's strategy of copying the functionality of smaller extensions/apps makes sense security-wise..
26.
by espe 3y ago
yes. at first glance it looks like a windows app but it's actually very portable. it has some parameters for gpu offloading and extended context size that just work. it exposes an api endpoint. i use it on a workstation to serve larger
27.
by espe 3y ago
this. also, they have a payment feature that opens them up further on grounds of regulatory scrutiny.
28.
by espe 3y ago
those paratextual phenomena probably are important for the model's representations.. not to get rid of and not easily compressable either. have a look at predicitive features for authorship attribution in stylometry for example. whites
29.
by espe 3y ago
"retina" describes the orientation of the vector space? first time i hear it in this context.
30.
by espe 3y ago
i think they don't. given enough data those related forms will have similar colocation profiles (aka fingerprints). not a model of semantics.. at scale it might become a proxy though.

More ›