angusturner

9 ms·

61.
by angusturner 4y ago
As someone working on singing synthesis, I know how hard it is to get that last 10% quality that makes a human listener instantly recognise if the voice is real or generated. These are really impressive results! For anyone interested, my te
62.
by angusturner 4y ago
I agree that adoption tends to be slow and there’s a lot of legacy systems that will need humans to look after them for a good while. But I think you are selling GPT a bit short. Partly because transformers are incredibly general architectu
63.
by angusturner 4y ago
If you haven’t seen it, I’d highly suggest looking at some of the recent research on in-context learning and reasoning in LLMs. It is becoming increasingly clear that there is a phase-change in behaviour when these models get large enough,
64.
by angusturner 4y ago
At this point I'm convinced we could have sentient humanoid robots living among us, and there would still be a contingent of internet users assuring us that "It's not really intelligent..." (presumably bc something abo
65.
by angusturner 4y ago
I mean, in a sense yes the training set is gospel. But these systems are also (generally) tested against held out data. When you have to model 100TB of images with 4gb of weights there is no way this is possible without learning some kind o
66.
by angusturner 4y ago
I feel like a broken record on this topic lately, but I strongly believe that training ML models on copyrighted works should be legal. It is clear to anyone that understands this tech that it is not simply "memorizing" or "co
67.
by angusturner 4y ago
the octopus would like a word with you
68.
by angusturner 4y ago
I think any argument against generative AI should not hedge on there being a fundamental difference in how humans and generative models work. I mean sure, maybe humans aren't "just" doing sophisticated pattern matching, but t
69.
by angusturner 4y ago
I get the impression that many peoples' grievance with generative AI (text, code, images etc.) isn't _really_ about the data provenance. Or at least, it feels secondary, compared to the general disruptive nature of the tech. If to
70.
by angusturner 4y ago
Only tangentially related, but there is some cool work on using reinforcement learning to learn an optimal tax rate. See: “AI Economist”.
71.
by angusturner 4y ago
That just feels like such a pessimistic forecast to me. Of course, the current trajectory of improvements in model efficiency and better commercial GPUs / ML-accelerators may hit a wall. But I would not be surprised if this was trainab
72.
by angusturner 4y ago
A few months ago this task was virtually impossible. Then it was possible, but extremely expensive and pay-walled behind the "Open"AI' website. As of about a week ago this tech runs on consumer GPUs. The weights have been dow
73.
by angusturner 4y ago
It seems pretty well-agreed that the HN moderation works because of dedicated human moderators and community guidelines etc. I think spaces that effectively moderate AI art content will be successful (or not) based on these same factors. It
74.
by angusturner 4y ago
Developing models that can predict if stuff is harmful ironically makes it easier for people to optimize for harm. e.g. the one line of code in Stable Diffusion that predicts if stuff is NSFW, can be inverted to generate only NSFW stuff.
75.
by angusturner 4y ago
Relevant earlier discussion about this issue: https://news.ycombinator.com/item?id=32436203
76.
by angusturner 4y ago
My bad, I misunderstood you! I agree with you.
77.
by angusturner 4y ago
Where I think this analogy breaks down is that these models are not simply copying / memorizing the training data. They are using the data to fit a model, highly compressing the specifics while holding on to general patterns and abstra
78.
by angusturner 4y ago
These are fair points. I suppose I am more optimistic about all the potential ways this tech could augment or enhance human creativity. Like, isn’t it kind of wild to think that the average person could have access to models that have learn
79.
by angusturner 4y ago
I don’t think it makes sense to talk about these systems as though they have any agency (and I think I was careful not to). Of course there is nothing that necessitates letting these systems do what humans do. What I am saying is that I am
80.
by angusturner 4y ago
Feel like people would have said the same with images 6 months ago tbh. (I mean sure, you could get a fake photograph of a person, but the text control that clip has unlocked is just wild…)
81.
by angusturner 4y ago
The question that fascinates me about this whole debate is on what basis (if any) one draws the line between a machine learning from copyrighted data vs a human learning from copyrighted data. And I do think “learning” is the correct metaph
82.
by angusturner 4y ago
Good on them. As a researcher it is so disheartening when GPUs get hoovered up by people participating in proof of waste pyramid schemes. Likewise, if google decides it wants to preserve its service for students and researchers, then I hope
83.
by angusturner 4y ago
Isn't this just proof of work?
84.
by angusturner 5y ago
This is awesome. I use the "self control" app for Mac OS X, which is pretty effective, but obviously doesn't stop me using my phone. Would be interested to learn more about this approach.
85.
by angusturner 5y ago
A lot of crypto trade (especially on shadier exchanges without KYC or the option to use fiat) is denominated in tether. Tether keeps printing billions of dollars with no oversight or regulation. They lied about it being backed by cash reser
86.
by angusturner 5y ago
This is actually quite good for Bitcoin.
87.
by angusturner 5y ago
This is such a good point. This is my first thought whenever I see a new "blockchain for x", where x involves some real-world product or activity. Suddenly the promise of "trust-less" makes no sense, and we are just left
88.
by angusturner 5y ago
As someone familiar with the old react API, I agree 100% about useEffect being super arcane. There's just too much magic going on, which is the same thing that makes me reluctant to write Vue or Svelte (after reading this)
89.
by angusturner 5y ago
Wow, a lot of people are upset about this. Curious what HN thinks of the argument that Layer 2 scaling + POS will solve / mitigate the environmental problems with ETH at the moment.
90.
Diffusion Models as a Kind of Hierarchical VAE(angusturner.github.io)
5 points by angusturner 5y ago|2 comments

More ›