dilap

17 ms·

121.
by dilap 2y ago
llama.cpp does support something like this -- you can give it a grammar which restricts the set of available next tokens that are sampled over so in theory you could notice "```python" or whatever and then start restricting to val
122.
by dilap 2y ago
I believe just executable files, right? (Still terrible of course.)
123.
by dilap 2y ago
I was curious too: > An "airplane" in conkers is when a player swings their conker in a wide, sweeping, horizontal motion, typically at about shoulder height.
124.
by dilap 2y ago
Biggest issue w/ emacs is once you start adding packages everything is half-broken. Ideal emacs replacement would preserve the ease of extension and flexibility but have more of a static time verification systems (types and ...?) to pr
125.
by dilap 2y ago
I guess if you're doing something like the original post's "type Iterator[V] struct" you could somehow extend it capture the extra information to make an efficient Reverse work. (FWIW, in C#, the IEnumerable extension me
126.
by dilap 2y ago
Thanks, good points.
127.
by dilap 2y ago
Yeah, that all makes sense!, but I think it's not relevent to the code in question, which is this: func (i *Iterator[V]) Reverse() *Iterator[V] { collect := i.Collect() counter := len(collect) - 1 for e := range i.
128.
by dilap 2y ago
The Reverse implementation seems off to me -- it runs through the iterator twice, once collecting into a slice, and then a second time filling the same slice in reverse. (So basically the first Collect call is only being used to find the le
129.
by dilap 2y ago
it can be fixed in theory if the model knows-what-it-knows, to avoid saying things its uncertain about (this is what (some) humans do to reduce the frequency w which they say untrue things). theres some promising research using this idea, t
130.
by dilap 2y ago
A related mental trick I use to move on from some mistake I'm stuck on is to think the following: In the future, I will make another mistake, and I will be regretting that instead of this. So I should focus on the present and the futur
131.
by dilap 2y ago
There's an interesting whack-a-mole dynamic at play here, since, from what I've read, the proliferation of huge trucks is partly driven by regulations in the first place -- fuel efficiency standards are less stringent for large tr
132.
by dilap 2y ago
Scrolling (much nicer/smoother feeling 120hz).
133.
by dilap 2y ago
They suspended Tim Peters? That is insane .
134.
by dilap 2y ago
I think it's completely reasonable for FB to take the position of "it is not our job to block links to newspapers". Similarly I don't blame FB for failing to block links to the Steele dossier, even though I think it was
135.
by dilap 2y ago
Spam is a real problem, but when your platform is doing things like disallowing linking to a NY Post article on the Hunter Biden laptop or mentioning the possibility that COVID originated from a lab-leak, then I think pretty clearly the ter
136.
by dilap 2y ago
It's not even true in a facile way for non-base-models, since the systems are further trained with RLHF -- i.e., the models are trained not just to produce the most likely token, but also to produce "good" responses, as deter
137.
by dilap 2y ago
ha totally forgot about that! what a crazy feature. i wonder if anyone uses it in the wild?
138.
by dilap 2y ago
I think the closest equivalent in Java or C# would be to use reflection to pull out the Swim and Quack methods.
139.
by dilap 2y ago
Ah, ok, thanks, now I see the point the article was trying to make. Yes -- supporting structural typing + also runtime casts means you support duck typing.
140.
by dilap 2y ago
Confusing post; as I see it, it doesn't really have much to do with duck-typing. It's more an illustration that with Go you can, if you choose, take "any" as your argument type and then later cast to the actual type you
141.
by dilap 2y ago
Looks like generic-brand macOS. All that work but that can't find anyone to give it a distinctive look?
142.
by dilap 2y ago
Isn't your example showing an issue w/ the opposite approach, where someone is getting bad output w/ an earlier openAI json mode that worked via training rather than mechanical output restriction to conform to a schema? FWIW
143.
by dilap 2y ago
You just sample from a grammar and you automatically get 100%; who knows but it seems the most likely thing they are doing. llama.cpp has supported this for a while ( using a BNF-style grammar -- https://github.com/ggerganov
144.
by dilap 2y ago
Then there's the approach taken by my university's physics department, where they made it a point of pride to always have you using math before you'd learned it from the official math classes...
145.
by dilap 2y ago
On mac/iOS, you get this using the AVAudioEngine API if you set voiceProcessingEnabled to true on the input node. It corrects for audio being played from all applications on the device.
146.
by dilap 2y ago
I hate those -- they instantly make me nauseous.
147.
by dilap 2y ago
I found my pistol squats got better after a few months of just training boxing -- so I think even just hopping around and doing jump-rope is probably pretty decent exercise for your legs.
148.
by dilap 2y ago
groq is super fast and cheap, but it seems to have (for normal plans) a very-low 30 requests/minute rate limit; to get a higher-limit, it's "contact us" enterprise pricing (which I haven't done, so I'm not sure
149.
by dilap 2y ago
There's a trade-off between doing things you enjoy (or don't enjoy) and the risk of death, and it's a personal trade-off, not something that can be assessed rationally.
150.
by dilap 2y ago
Thank you for providing a nice side-by-side. This makes it clear to me the voices are not very similar at all. If Johansson had agreed, I have to imagine they would've been able to make a much closer (and less annoying!) voice.

More ›