meroes

23 ms·

181.
by meroes 2y ago
Seems like every installation had a backdoor password hardcoded too from that article.
182.
by meroes 2y ago
If everyone 30+ bmi can get to 30 for “free” (not sure where the subsidizing stops, for me it’s free if I’m over 30 bmi), that’s just too tantalizing to pass up, even if the moral applies. At least it takes a load off one problem (obesity r
183.
by meroes 2y ago
Or, keep the info but change our perceptions.
184.
by meroes 2y ago
Well said. As someone with only a math undergrad and as a math RLHF’er, this speaks to my experience the most. That craving for an understanding an elegant proof is nowhere to be found with verifying an LLM’s proof. Like sure, you could put
185.
by meroes 2y ago
Glad it’s not just me. People saying the FF market is small, well why does YouTube work flawlessly on other small browsers like Brave or Opera?
186.
by meroes 2y ago
Saltwater crocs bridge the gap! (idk only eaten alligator but it was surprisingly “terrestrial”)
187.
by meroes 2y ago
A baby learns to walk and talk in 1 year. Compared to the number of PHDs and compute training these models, the baby is so far ahead in efficiency I marvel way more at their pace.
188.
by meroes 2y ago
Aren’t prompts seeking to offload reasoning though? Is that really a fair data point for this?
189.
by meroes 2y ago
Scale AI is a big one who owns companies who do this as well, such as Outlierai. There are many other AI trainer job companies though. A lot of it is gig work but the pay is more than the vast majority of gig jobs.
190.
by meroes 2y ago
At a certain level they are identical problems. My strongest piece of evidence is that I get paid as an RLHF'er to find ANY case of error, including "tokenization". You know how many errors an LLM gets in the simplest grid pu
191.
by meroes 2y ago
Well you got it wrong since there are no "other defibrillation devices" since these are computer models.
192.
by meroes 2y ago
ChatGPT and others can’t reliably sum several integers together still, when is an LLM going to replace anything meaningful in higher math? I spent more time deciphering ChatGPT’s mistakes than programming something in python to sum some int
193.
by meroes 2y ago
Many of these devices aren’t actually fungible commodities though.I tried to buy a phone in the $200 range and Android phones were so much worse than used iPhones. I they would drop calls and freeze on the dial pad during long calls where I
194.
by meroes 2y ago
It made it into their data set via RLHF almost assuredly. Wild these papers are getting published when RLHF'ers and up see this stuff in the wild daily and ahead of the papers. Timeline is roughly: Model developer notices a sometimes h
195.
by meroes 2y ago
Irrelevant info is taught in grade skill and is a skill for the SAT for example. Basically any kind of model (not just LLMs/ML) has to distill out irrelevant info. The point is having an answer that you can defend logically and most pe
196.
by meroes 2y ago
I don’t think it’s about junk vs not junk here. It’s whether this is novel enough data compared to the billions of images and hours they already possess. Since others have already said these are fundamentally different than pointing and sho
197.
by meroes 2y ago
Game changing for billions? Please give an example. I’m fine to be proven wrong but I don’t know what you could mean.
198.
by meroes 2y ago
And some Mongols, Romans, etc probably said continually expanding their boarders would give them comparable contemporary heights.
199.
by meroes 2y ago
Wow that quote is absolutely crazy. “Discovery of all of physics”. I think that’s the worst puffery/lie I’ve ever heard by a CEO. Science requires experiments so the next LLM will have to be able to design CERN+++ level experiments dow
200.
by meroes 2y ago
This was probably the influential topic I had in English class in HS (ads work on everyone, perhaps even especially on those who think they have no effect). I took it to heart and consume extremely little media with ads, because they are so
201.
by meroes 2y ago
Yes and it was never a point, it was just much much denser but still infinite. And so uniform there was no net direction for collapse before expansion took over.
202.
by meroes 2y ago
I’m a layman but I think I’ve understood this enough to repeat. The early universe is hypothesized to have been so uniform (gravity pulling in every direction) there was no net direction for anything to collapse to. Because the expansion wa
203.
by meroes 2y ago
Well ya once each project, e.g. “grad level math”, “k—12 math”, “undergrad math”, “k-12 chemistry”, etc is sufficient they are all fed into a larger more powerful model. Maybe my RLHF work does make it harder for me to dream, but I teach mo
204.
by meroes 2y ago
I mean paying several hundred to thousands of grad students to RLHF for several years and you get a corpus of grad-student text. I'm not surprised at all. AI companies hire grad students to RLHF in every subject matter (chemistry, phys
205.
by meroes 2y ago
Before LLM's and the contemporary advances, my good friend argued "AI" could never be funny, because humor requires subversion. Now I think AI humor is both here and art. I'm not an AI evangelist by any means, but the &q
206.
by meroes 2y ago
MWI is not local according to all the big names I read; Lev Vaidman, Tim Maudlin, and Im pretty sure David Wallace too.
207.
by meroes 2y ago
So wanting is a key driver then, since it makes the chemical changes more sticky?
208.
by meroes 2y ago
If we are that self-deceptive as you are claiming, don’t we need to be extremely careful we aren’t deceiving ourselves into less autonomy? How can we?
209.
by meroes 2y ago
How do you explain some of the most drastic and lasting changes in many peoples lives coincide with wanting them very deeply? We just happened to want most what we achieve?
210.
by meroes 2y ago
It's either true as an axiom or true as derived from other axioms and/or theorems. In neither case does latching onto Euclid's other common notions/postulates/theorems as the selection it must be proved true from

More ›