humansareok1

11 ms·

91.
by humansareok1 2y ago
> Still can't write a short paragraph without using the letter 'e' 'Crafting a paragraph without using a particular symbol (which is most common in all writing) is a daunting task; still, I will aim to comply. This ta
92.
by humansareok1 2y ago
I think Auctions exist explicitly to potentially buy or sell an Item with a delta on it's Market Value? I.e. Buyers want the chance to buy below and sellers to sell above. Neither really wants to engage in the transaction at all in the
93.
by humansareok1 2y ago
At the same time compute required is increasing like 10x each generation, its scaling faster than Moore's law.
94.
by humansareok1 2y ago
Bro given that the cost of frontier models is already at or past $100M I think that boat has already sailed. Unless you have a completely cracked team that can raise like $1B upfront you have no chance at competing.
95.
by humansareok1 2y ago
Auctions by default almost always undercut the actual market value so no not really?
96.
by humansareok1 2y ago
What's this thing actually worth? Current bid is ~3k.
97.
by humansareok1 3y ago
If it costs 200$ to add the device and modify the saw to accept it and the original saw cost 300$ you've got a pretty massive increase. Also from some deep dive I saw apparently SawStop has basically cornered the entire premium market
98.
by humansareok1 3y ago
Do you risk losing fingers by repairing your Iphone?
99.
by humansareok1 3y ago
You hit a knot in your 2x4 while on a table saw and you might be surprised when you see your fingers laying on the ground and you thought you were being safe too.
100.
by humansareok1 3y ago
It's not expensive to replace, the brake is like 100$ and it sure beats a 10,000$ hospital bill and a couple digit amputations.
101.
by humansareok1 3y ago
Cynical take is that the SawStop feature adds enough cost to budget table saws that they will no longer be economically viable and you can only purchase mid-high end tables saws going forward.
102.
by humansareok1 3y ago
Great response! I wonder if part of my disagreement is in thinking of GPT4 as a System rather than just as an LLM. If the system has the ability to solve problems in code it feels like intentionally handicapping it by forcing it to solve pr
103.
by humansareok1 3y ago
Does this not fall under the US CFAA Laws? I.e. intentionally adding malicious code?
104.
by humansareok1 3y ago
Why would they, imo signs point to a nation state or at minimum a very sophisticated APT so it seems incredibly unlikely they're going to make a mistake like that.
105.
by humansareok1 3y ago
>> To study how the parallel structure of transformers might limit their capabilities, the pair considered the case where transformers didn’t feed their output back into their input — instead, their first output would have to be the f
106.
by humansareok1 3y ago
>LLMs have no concept of truth or correctness really, they are just probabilistic appending to output to match learned ideal patterns. I think they do have those concepts, like they have many others but they are not bound to them in any
107.
by humansareok1 3y ago
Theres a net gain between credit card rewards and higher prices for reward card users?
108.
by humansareok1 3y ago
>All evidence suggests that current LLMs don't scale (EDIT: they do scale, but at a certain point somewhere around GPT4 the scaling slows down very quickly) There seems to actually be no evidence that LLMs dont scale. All data from
109.
by humansareok1 3y ago
But this is just current state. Token costs continue to go down and contexts will continue to get larger.
110.
by humansareok1 3y ago
If you're running a 1M context LLM I assume you are well past memory constraints since the weights alone are 10s of GBs.
111.
by humansareok1 3y ago
Well I guess the assumption is that you can just fit everything into the context and there'd be nothing to retrieve anymore.
112.
by humansareok1 3y ago
But it wasn't caught because it was open source? It was caught basically by chance by an engineer using the distro who admits they only caught it because of a series of coincidences.
113.
by humansareok1 3y ago
Is there any real point to further RAG work given extremely large contexts are clearly on the way with 1M token contexts already proven?
114.
by humansareok1 3y ago
This represents a massive failure of the Open Source model and it's worth thinking carefully about as more and more people advocate for fully open source AI models. People could spend a lot of time on token safety features only to have
115.
by humansareok1 3y ago
>whose main theorem is given at the most general possible level with no applications at all. Outside of Applied Math why would this be an expectation at all?
116.
by humansareok1 3y ago
So you think whatever process produces a GPT4 update is completely equivalent to pretraining and RLHF'ing a brand new model with new architecture, more data, etc??
117.
by humansareok1 3y ago
You're maybe proposing another line where no spying is legal at all and we should just submit ourselves to the whims of terrorists or other lunatics? Surely there is actually a line where there are tradeoffs between security and privac
118.
by humansareok1 3y ago
As another poster mentioned this is literally why we have courts. There is a clear line for obtaining a search warrant for example. Precedents exist.
119.
by humansareok1 3y ago
Go ahead, no one is saying to stay with GPT4. But its disingenuous to compare a gpt-4-march-update to a completely new pretrained model like Claude 3 Opus.
120.
by humansareok1 3y ago
I don't support ubiquitous spying at all but are you hanging out with known ISIS members or members of White Nationalist Militias regularly? Because I'm pretty sure that's where the line begins.

More ›