adroniser

7 ms·

31.
by adroniser 2y ago
That doesn't imply that it's GPT-5. A GPT-4 training run probably doesn't take them that long now they've acquired so many GPUs for training GPT-5.
32.
by adroniser 2y ago
Agree on hoping for an intelligence update, but I think it was clear from teasers that this was not gonna be GPT-5. I'm not sure how fair it is to classify the new multimodal capabilities as just a gimmick though. I personally haven&#x
33.
by adroniser 3y ago
Lol there's a massive difference between a framework that generates javascript, a language which has existed for 30 years at this point, and a magic LLM that no one on earth understands the internals of.
34.
by adroniser 3y ago
Absolutely. As far as I know both Fujitsu and the Post Office were aware of the bugs in the system, and deliberately covered them up.
35.
by adroniser 3y ago
Given a fixed level of precision for input and output, and a function on this discrete space, we can construct a continuous extension of this function to the reals. Now using the paper we know that there is a neural network with continuous
36.
by adroniser 3y ago
But continuous depends on the topology of the space you are working over. If the topology is discrete any function is continuous. I've found some papers that argue that simply from an error corrections standpoint the signals in the bra
37.
by adroniser 3y ago
It's worth saying we still haven't seen a single model trained post-hype cycle from openAI. GPT-4 was summer 2022 stuff.
38.
by adroniser 3y ago
You have to be a lot clearer about what you mean by continuous here. An LLM technically does not produce a continuous function on the reals because its inputs are floating point numbers with finite precision. Any function on discrete inputs
39.
by adroniser 3y ago
I guess the disagreement you have with doomers is the probablility you assign to us being able to create human level AI. For the record most "doomers" don't pull a pascal's wager here. Standard EA doctrine is that if we
40.
by adroniser 3y ago
At the very least, when compute passes a certain threshold we are just gonna be able to simulate human brains.
41.
by adroniser 3y ago
Suppose you wanted to train an LLM to do addition. An LLM has limited parameters. If an LLM had infinite parameters it could just memorize the results of every single addition question in existence and could not claim to have understood any
42.
by adroniser 3y ago
I'm saying that it's always going to be in the back of your head that it's an AI regardless of how it acts.
43.
by adroniser 3y ago
Do you think people would listen to Kanye if it wasn't Kanye?
44.
by adroniser 3y ago
I am definitely generalizing. It does seem to be a common sentiment expressed when this gets brought up though. Whether it's just an instance of me clinging on to anything to distinguish from AI or if it's a legitimate problem rem
45.
by adroniser 3y ago
Don't you think simply knowing it was an AI would mean you can't take it seriously? Or do you think this is just a cultural thing that will fade away in coming years.
46.
by adroniser 3y ago
You don't have much control over what is being generated. And if you did want to have fine control then what would be the difference compared to just dicking around on a synthesizer? A lot of the enjoyment of music comes from connectin
47.
by adroniser 3y ago
yes???
48.
by adroniser 3y ago
What do you mean be replaced wholly?
49.
by adroniser 3y ago
It seems like you agree with me then?
50.
by adroniser 3y ago
I don't like where we are headed at all. I acknowledge we face two dystopian options which is either contribute power in the hands of a few corporations who hopefully you can regulate, or have open source models which ends up deliverin
51.
by adroniser 3y ago
And many countries ban the possession or distribution of those blueprints and the united states had a ban on re-publication of those 3d designs from 2018 until trump reversed it, and even now it requires a license to post blueprints online.
52.
by adroniser 3y ago
Yes but if we had the ability to download a gun from the internet anonymously with no way to feasibly get the identity of the person downloading the gun I think we would be right to be concerned. Especially if you could then shoot that gun
53.
by adroniser 3y ago
How can you effectively enforce anything if the models are open source? How do you draw the line if a deepfake is not defamatory (making someone say something they didn't say) but in fact just makes someone look silly https:/&#x2
54.
by adroniser 3y ago
You should worry about both problems. You're telling me that AI isn't going to improve it's video capabilities in the next 4 years enough to make convincing deepfakes?
55.
by adroniser 3y ago
I would personally feel pretty guilty about making use of the tool knowing i'm using someone else's time. Also privacy reasons I guess, it feels like there's a difference between a person seeing your photo and an ai seeing it
56.
by adroniser 3y ago
I'm not claiming anything about what causes consciousness to arise. I'm not claiming it doesn't or that it does. I'm saying it's irrelevant. That is all. You can come up with all sorts of theories about what causes
57.
by adroniser 3y ago
Again when I say consciousness I mean a subjective experience. If you define consciousness to literally just mean models that plan then of course tautologically if you can't reach consciousness you can't get to a certain level of
58.
by adroniser 3y ago
I think we have different definitions of consciousness and this is what's causing the confusion. For me consciousness is simply having any subjective experience at all. You could be completely numbed out of your mind just staring at a
59.
by adroniser 3y ago
Because consciousness is metaphysical? You can't test scientifically if one person's red is the same as another's.
60.
by adroniser 3y ago
I think you're right it could have been created without AI. I'm trying to think of the right way to say it. Maybe it wouldn't have been created without AI? Or AI has made it so simple to express this idea that the idea has be

More ›