upghost

23 ms·

241.
by upghost 2y ago
Wow, I'm sorry you had to take a downvote for supporting this. Wild times we live in. Would the down-voter mind explaining their reasoning?
242.
by upghost 2y ago
Thanks for that. I really don't like to rock the boat or make comments like this. But the thing is that saying LLMs are "capital I" Intelligent moves them out of the realm of something we can reason about into the realm of pr
243.
by upghost 2y ago
No, this is false. Reinforcement learning also uses a dataset, it's just not an enumerable dataset. The dataset it is trained on comes from the Bellman Backup equation (or variants, including DL variants) being run against a simulation
244.
by upghost 2y ago
No, stop. Full stop. Categorial error. Title SHOULD be DATASETS (whether enumerable or supplied by proxy via model such as PPO) are increasingly being created that make humans think "AIs are faking alignment". PSA: An ideal tran
245.
by upghost 2y ago
I'm not even sure that the goal of Artificial Life is actually "life", although that may be the AGI equivalent of ALife -- AGL or "Artificial General Life"?. In practice I think the discipline is much closer to the
246.
by upghost 2y ago
Wow, it is really interesting the difference in comments between ALife and AI stories on HN. For some of you out there, there's a great book that really hasn't gotten enough attention called "The Self-Assembling Brain"
247.
by upghost 2y ago
Sure, but given the resourcing at OpenAI, it would not be hard to clean[1] the inputs. I'm just trying to be realistic here, there are plenty of ways around contractual obligations and a significant incentive to do so. [1]: https:&#x2
248.
by upghost 2y ago
Oh wow. No, that's heart breaking. I'll have to read up on this. Reminds me of David explaining the interesting and somewhat surprisingly insensitive language the oncology literature uses towards folks going through this. Its ther
249.
by upghost 2y ago
It's just a fancy word for clamping the new reward value to within some delta of the original value. Otherwise the model ends up "exploiting" outliers that make sense to machines but not to humans. They do the same thing wi
250.
by upghost 2y ago
It's so much worse than you could possibly imagine. I worked for a healthcare startup working on patient enrollment for clinical oncology trials. The challenges are amazing. Quite frankly it wouldn't matter if the data were in pla
251.
by upghost 2y ago
Ok, where do you nerds hang out and why am I not there?? I'm loving this discussion, y'all seem to be a rather rare breed of dev though. Where is the community for whatever this sort of dev is called? i.e., Clojure has the Clojuri
252.
by upghost 2y ago
Not sure why your comment was downvoted, but it certainly shows the pressure going against people who point out fundamental flaws. This is pushing us towards "AVI" rather than AGI-- "Artificially Valued Intelligence". Th
253.
by upghost 2y ago
Insightful comment. The thing that's extremely frustrating is look at all the energy poured into this conversation around benchmarks. There is a fundamental assumption of honesty and integrity in the benchmarking process by at least so
254.
by upghost 2y ago
Nope, makes sense to me. Seems unreasonable to conclude the dataset is not compromised now.
255.
by upghost 2y ago
I'm sure it was independently evaluated, but I'm sure the folks running the test were not given an on-prem installation of ChatGPT to mess with. It was still done via API calls, presumably through the chat interface UI. That means
256.
by upghost 2y ago
I didn't see anyone else ask this but.. isn't the FrontierMath dataset compromised now? At the very least OpenAI now knows the questions if not the answers. I would expect that the next iteration will "magically" get ove
257.
by upghost 2y ago
This is going to be bananas for libpython-clj[1]. One of the biggest limiting factors right now is that you can't mix Java/Clojure concurrency with Python concurrency, you need to have a really clear separation of concurrency mod
258.
by upghost 2y ago
"All You Need" in the title is apparently All You Need.
259.
by upghost 2y ago
Damn, wow. Good stuff. First of all, I assumed looping within kernels was "always bad" from a performance standpoint (not that I'd done any benchmarks on looping vs repeated kernel invocation). I also assumed it was "alw
260.
by upghost 2y ago
Er, to be clear, this just seems to be an interpreter -- not a meta circular interpreter. I don't see any evidence that it is able to evaluate its own definitions or its own API. It doesn't have any tools to handle class definiti
261.
by upghost 2y ago
Wow, great answer. That deserves more upvotes than it's probably going to get. So what that means for compute shaders is I will need to partition the vectors being multiplied into power of 2 sized partitions and repeatedly call the k
262.
by upghost 2y ago
I tried this (and backprop) in GPGPU compute shaders once. One issue that really tripped me up is matrix multiplication along dimensions that are not powers of 2. It will take me awhile to learn to read CUDA code -- is this just an easier t
263.
by upghost 2y ago
I tried making Dyalog bindings but wasn't smart enough to figure it out at the time based on the info available and I doubt much has changed regarding either my intelligence or information availability there. But it would be awesome!
264.
by upghost 2y ago
Jeeze, the overlap of people who are interested in Clojure and array langs couldn't exceed the low double digits. Well, if you are among that sort of masochist and you like Clojure and APL, you could play with libapl-clj: https:/
265.
by upghost 2y ago
of all the comments on the page, THIS is the one that gets downvoted!?
266.
by upghost 2y ago
Great point about the meaningful datasets, this makes perfect sense. Esp. in regards to SFT and RLHF. Although I suppose it would be somewhat easier to do pretraining on really long context (books, I assume?)
267.
by upghost 2y ago
So why are the context windows so "small", then? It would seem that if the cost was not so great, then having a larger context window would give an advantage over the competition.
268.
by upghost 2y ago
Cheers, friend! Thankful for HN and all the great perspectives from around the world. Also thanks for the phrase "Make AI Great Again"! :D
269.
by upghost 2y ago
this is a so-often overlooked point.
270.
by upghost 2y ago
In all seriousness it would be fun to know a little more about the story. Was this a rebellious act? Is this practice more common? Did this company do more things like this? Was the restraint in the message a sign of professionalism? Was

More ›