meroes

8 ms·

91.
by meroes 1y ago
Math RLHF already has verifiable ground truth/right vs wrong, so I don't what this distinction really shows. And AI changes so quickly that there is a breakthrough every week. Call my cynical, but I think this is an RLHF/RLVR
92.
by meroes 1y ago
https://work.mercor.com/jobs/list_AAABljpKHPMmFMXrg2VM0qz4 https://benture.io/job/international-math-olympiad-participa... https://job-boards.greenhouse.io/xai/jobs/45387
93.
by meroes 1y ago
Latent?? If you looked at RLHF hiring over the last year, there was a huge hiring of IMO competitors to RLHF. This was a new, highly targeted, highly funded RLHF’ing.
94.
by meroes 1y ago
They didn’t actually compete.
95.
by meroes 1y ago
In the RLHF sphere you could tell some AI company/companies were targeting this because of how many IMO RLHF’ers they were hiring specifically. I don’t think it’s really easy to say how much “progress” this is given that.
96.
by meroes 1y ago
DOA doesn’t mean forever and always. But certain claims like the advent of living beyond 200, humans on Mars, etc can just be immediately dismissed outright for several decades. What you’re talking about is unsupervised LLM therapy. Even wh
97.
by meroes 1y ago
My bank’s fraud department uses text shorthand like “Stop2end” and “call ph#” and their dates lack spaces “24Jun” in their texts to me. Is this some kind of meta-level play to sound less fake?
98.
by meroes 1y ago
There's no chance LLMs have a sufficient training set of effective therapists-patient interactions, because those are private. Ergo, there is no need to wait, it's DOA. Anything else is feeding into LLM hype. It's that simple
99.
by meroes 1y ago
Driving through the Central Valley I’ve seen signs for 10 for $1. No idea if they are extra small, but typically it’s $1-$2.50 per
100.
by meroes 1y ago
LLMs are “taught” two kinds of “truth”. One is 100% adherence to a reference text. If the text says the Coliseum is in Antarctica or 1+1=716, model must too. The other is adherence to reputable outside sources. Not sure if it’s embarrassing
101.
by meroes 1y ago
> arithmetic (why would use LLM for that ?) Because people ask LLMs all of these things, including arithmetic. People were saying the same about the number of r's in strawberry. Why ask and LLM that!?!? But the big AI companies want
102.
by meroes 1y ago
> there exists a string of words (not necessary ones that make sense) that will get the LLM to a better position to answer exactly. The opposite is also true. You might supply more clarifying information to the LLM, which would help any
103.
by meroes 1y ago
The problem is these meetings are so low information density even an AI summary is not worth my time. And it’s not some elitist mindset. It’s like the entire reason there are these regular meetings is to make some mid level person feel bett
104.
by meroes 1y ago
I don't buy the token explanation because RLHF work is/was filled with so many "count the number of ___" prompts. There's just no way AI companies pay so much $$$ for RLHF of these prompts when the error is purely i
105.
by meroes 1y ago
And the way to claw back rank on the world scene surely takes investing, and investing in science is among the best options for long term returns.
106.
by meroes 1y ago
The most disagreeable thing is they are psychologically insidious.
107.
by meroes 1y ago
You can cut out auto foam cups without much change compared to redoing the entire country’s transportation infrastructure that was built with unprecedented federal funds.
108.
by meroes 1y ago
The argument is more like our most successful sciences like physics use our most successful math, which is ZFC. And thus incompatible reals are part of this package deal. Maybe it’s possible to do physics without any packaging of lots of th
109.
by meroes 1y ago
> absolutely nothing requires an incomputable continuum of reals. In a sense this is wrong. Some say the indispensability of modern math, which includes the incomputable reals, shows that such abstract objects are required for science.
110.
by meroes 1y ago
Doesn't scale contract out/aggregate many smaller companies, and that's where the bespoke solutions are?
111.
by meroes 1y ago
What? AI proofs are still atrocious.
112.
by meroes 1y ago
Solid article
113.
by meroes 1y ago
There are middle-men companies like Scale that recruit thousands of remote contractors, probably through other companies they hire. There are of course other less known such companies that also sit between the model companies and the contr
114.
by meroes 1y ago
Ya I don’t think I’ve seen any article going in depth into just how many low level humans like data labelers and RLHF’ers there are behind the scenes of these big models. It has to be millions of people worldwide.
115.
by meroes 1y ago
Hmm how can I use being forced to use Chrome for work, for me tax wise… If I’m a contractor forced to use Chrome and mobile devices, can I deduct a separate work phone? I really hate having it my iPhone, at least maybe I can claw something
116.
by meroes 1y ago
Yep. This is exactly the conclusion I reached as an RLHF'er. Reasoning/LRM/SxS/CoT is "just" more context. There never was reasoning. But of course, more context can be good.
117.
by meroes 1y ago
Ya for Alzheimer's, the two people in my family who've had it had bad diets (think dining out lavishly every night or sugar-addiction) and poor hygiene (like having every single tooth pulled). I also suspect chronic lack of sleep
118.
by meroes 1y ago
Sounds like all ways to skirt around the actual learning bit. Consuming the classics while gaming? Really?
119.
by meroes 1y ago
> LLMs can write a large fraction of all the tedious code you’ll ever need to write. But, you still have to read it: > Reading other people’s code is part of the job...I have to read the code line-by-line anyways. So instead of writin
120.
by meroes 1y ago
I don't think the big bang or the universe ever had a center.

More ›