HNHacker News
TopNewBestAskShowJobs

angusturner

434 karma · joined September 9, 2020

ML Engineer

Interested in generative models, programming, music and philosophy of mind.

https://github.com/angusturner/

submissionscomments
angusturner··on Grok3 Launch [video]
Credit to the engineers that built this, but it fills me with rage that Elon has this sort of unchecked power.

How long before this starts getting deployed in safety critical applications or government decision making processes?

With no oversight because Elon seems to have the power to dismiss the people responsible for investigating him.

Anyone not scared by this concentration of power needs to pick up a book.

angusturner··on String of recent killings linked to Bay Area 'Zizians'
I am so so sorry you have to deal with this. As an Australian I have been watching on in horror this week at the way trans persons are being demonized and oppressed in your country. I know HN is meant to be an apolitical space, but I hope that the mods here have the sense to allow a certain amount of push back again this fascist nonsense.
angusturner··on Fastest rate of natural carbon dioxide rise over the last 50k years
I think the tools to solve the challenges of waste, environmental damage etc. already exist within the framework of capitalism. Mostly they are just unpopular and seen by many as a government overreach.

1. taxes that force corporations and individuals to pay for the negative externalities / social costs of their actions 2. regulation (e.g. stop allowing planned obsolence, mandate the right to repair etc.) 3. government spending into R&D, incentives and subsidies for renewables etc.

Anyway, my point is that the issue is basically one of co-ordination and political will. It obviously doesn't help that many Americans (and Australians too for that matter, where I live) don't accept the basic facts of the situation (before we can even discuss solutions).

angusturner··on Why I Don't Invest in Real Estate
Not sure why this is down-voted so heavily.

In Australia, the property market has been artificially inflated for years with favorable tax incentives that encourage investors to hoard properties and sit on them, while writing off any losses as tax deductions.

Meanwhile, renters are being squeezed because landlords are trying to pass on the cost of rising interest rates to their tenants, who are forced to choose between groceries and making rent.

And yet, I would bet my kidney that when interest rates fall we won't see landlords pass on the savings to their tenants.

As the OP's point, the landlord renter system HAS "divided our society into haves and have nots", and it causing serious bitterness in division in our country. And the issue is politically intractable, because the propertied class are not going to accept any reform that lowers the value of their house.

angusturner··on Level of Gaussians: Real-Time View Synthesis for Millions of Square Meters
Fascinating - thanks for the detailed reply. I can’t believe I failed to think of lighting but this makes so much sense.

It’s almost like for a pure ML solution you would need a nerf (or similar) which is conditioned on the entire (dynamically changing) scene geometry and lighting positions?

Graphics isn’t my area of ML though, so I’m sure there’s a lot of nuance that I don’t appreciate.

angusturner··on Level of Gaussians: Real-Time View Synthesis for Millions of Square Meters
Can anyone familiar with 3d graphics speculate what would be required to implement this into a game engine?

I'm guessing that adding physics, collision-detection etc. on top of this is non-trivial compared to using a mesh?

But I feel like for stuff like tree foliage (where maybe you don't care about collisions?), this would be really awesome, given the limitations of polygons. + also just any like background scenery, stuff out of the player's reach.

angusturner··on A vegan cheese beat dairy in a big competition. Then the plot curdled
Are beyond and impossible really failed? In Australia they seem quite popular. Just basing this off the fact that the supermarket section for plant based meat/dairy alternatives seems to be exploding in recent years. Also a few popular burger chains here offer impossible meat as one of the main vege options.
angusturner··on h5ai – modern HTTP web server index
Its worse than that. I've seen some people conflating the term "AI" with "LLM".
angusturner··on Making Truth Social Comply with the AGPL
Great write up. The Github mirror is also quite convenient.
angusturner··on Arraymancer – Deep learning Nim library
I would love for a non-python based deep learning framework to gain traction.

My initial impression though is that the scope is very broad. Trying to be both sci-kit learn and numpy and torch seems like a recipe for doing none of these things very well.

Its interesting to contrast this with the visions/aspirations of other new-ish deep learning frameworks. Starting with my favorite, Jax offers "composable function transformations + autodiff". Obviously there is still a tonne of work to do this well, support multiple accelerators etc. etc. But notably I think they made the right call to leave high level abstractions (like fully-fledged NN libraries or optimisation libraries) out of the Jax core. It does what it says on the box. And it does it really really well.

TinyGrad seems like another interesting case study, in the sense that it is aggressively pushing to reduce complexity and LOC while still providing the relevant abstractions to do ML on multiple accelerators. It is quite young still, and I have my doubts about how much traction it will gain. Still a cool project though, and I like to see people pushing in this direction.

PyTorch obviously still has a tonne of mind-share (and I love it), but it is interesting to see the complexity of that project grow beyond what it is arguably necessary. (e.g. having a "MultiHeadAttention" implementation in PyTorch is a mistake in my opinion).

angusturner··on Ask HN: What things are happening in ML that we can't hear over the din of LLMs?
One area that I would dive into (if I had more time) is "geometric deep learning". i.e) How to design models in a principled way to respect known symmetries in the data. ConvNets are the famous/obvious example for their translation equivariance, but there are many recent examples that extend the same logic to other symmetry groups. And then there is also a question of whether certain symmetries can be discovered or identified automatically.
angusturner··on MTA board votes to approve new $15 toll to drive into Manhattan
Sorry, but I cannot get past the fact that so many Americans openly refer to undocumented migrants as "illegals".

What is really dystopian is focusing blame onto immigrants instead of the economic and political elite.

angusturner··on Is the reversal curse in LLMs real?
I think this “wisdom” is not really agreed upon, and in fact I’ve seen a few works successfully leverage synthetic data (including a cool one from Sergey Levine et al recently).
angusturner··on If we want a shift to walking we need to prioritize dignity
I’m an Aussie, and after living 3 months in LA I think it was the most poorly designed city I’ve ever been to for this exact reason - I felt unable to walk practically anywhere!

It felt like my options were drive or taxi. And we know what LA traffic is like.

I can’t speak to other US cities, and it is possible that certain areas of LA are less terrible than where I stayed. (But I will say, I was in quite an affluent area which had no business being unwalkable and without public transport).

But it really opened my eyes to how good we have it in Australian cities (which themselves are still far behind many European cities).

angusturner··on YouTube removed dislike counts, so this guy made Rotten Tomatoes for YouTube
HN still seems to value constructive criticism though, to an extent that is simply not possible on YouTube.

YouTube comment section is basically a cesspool of toxic positivity now, with absolutely no informational value at all.

I feel like this has been the case for a couple of years now, where nothing even slightly critical will be surfaced by whatever automated algorithm they are using to filter/boost content.

angusturner··on NaturalSpeech 2: Zero-shot speech and singing synthesizers
Anyone else think its an abuse of terminology to refer to speaker conditioning as "in-context learning" and "prompting" now?

Like, using a reference encoder to condition on an unseen target speaker has been around for 4 or 5 years now. These results are already cool without mystifying the results by calling this "prompting" or "icl"

Unless there is some subtle difference that warrants the new terminology?

angusturner··on Sparks of Artificial General Intelligence: Early Experiments with GPT-4
I guess the challenge here is that the human mind is not a blank slate, and has been optimized first by billions of years of evolution.

If it takes all the data on the internet (or more) to bootstrap AGI, but that system is then capable of leveraging its knowledge to solve new out-of-distribution tasks, that seems like a fair test to me.

I agree with the article that we see "sparks" of this generality with GPT4.

angusturner··on High-res image reconstruction with latent diffusion models from human brain
Largely agree with this, although I think it would be interesting to formulate in terms of: "what is the mutual information between the fMRI scan and the stimulus".

i.e) is there actually more information than a few bits encoding a crude object category, which stable diffusion then hallucinates the rest (/ uses to regurgitate an over-fit image)?

Or are there many bits, corresponding spatially to different regions of the stimulus - allowing for some meaningful degree of generalization.

angusturner··on We’ve filed a law­suit chal­leng­ing Sta­ble Dif­fu­sion
There is a strong, well-understood connection between deep-latent variable models (e.g. VAEs, diffusion models), and compression.

Many state-of-the-art compression algorithms are in fact based on generative models. But the thing is, the model weights themselves are not the compressed representation.

The trained model is the compression algorithm (or more technically, a component of it... as it needs to be combined with some kind of entropy coding).

You could use Stable Diffusion to compress and store the training data if you wanted, but nobody is doing that.

angusturner··on We’ve filed a law­suit chal­leng­ing Sta­ble Dif­fu­sion
Oh my god, I'm glad someone else noticed this because it was driving me nuts. Deeply troubling that their case hinges on such a fundamental misunderstanding of the technology.

I bet a proper analysis of that toy experiment would conclude that none of the original data points are perfectly recovered: Only the underlying distribution / manifold is recovered, which really doesn't lend well to their argument at all.

angusturner··on We’ve filed a law­suit chal­leng­ing Sta­ble Dif­fu­sion
Stable diffusion (or any likelihood-based generative model) is a learned compression algorithm. It is not the "container of compressed bytes". You can use a trained generative model to compress images, by combining it with some kind of entropy coding / arithmetic coding.

In this sense, stable diffusion is more analogous to the JPEG algorithm than it is to a specific collection of JPEG files. As it stands, the originals trainng data is not stored, even in a compressed way.

angusturner··on This Voice Doesn't Exist – Generative Voice AI
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 team’s singing work: https://youtu.be/LPy20zSWhZA)

angusturner··on GPT Takes the Bar Exam
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 architectures, which make few assumptions about the domain. They are basically SOTA on every domain for which learning-based methods are applicable.

And secondly because ChatGPT can perform “in-context learning”. That is to say, purely by scaling next-token prediction, there is an emergent phenomena where GPT gains the ability to learn from the prompt it is given, and solve new tasks outside of the training distribution.

Hattie Zhou has a cool paper showing how you can teach GPT addition by prompting - a toy example, but it signifies something profound in my opinion. In the last 6-months I have shifted my own view from “scaling is a waste of compute” to a recognition that LLMs are a serious leap forward.

For something very hyped, I think it’s still under hyped and under appreciated.

angusturner··on Asking ChatGPT to write my security-sensitive code for me
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, such that they can solve new tasks outside of the training distribution.

See work by Hattie Zhou or Laura Ruiz for example.

It is clear to those following the research or using these models that they are not just copy-pasting… (Even if you can cherry pick examples were the LLM recalls highly occurring dataset items like fast square root or whatever).

angusturner··on Show HN: Whole Git repo was made with ChatGPT
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 about Searle's Chinese Room argument).

Reminds me of a joke about how the criterion for intelligence is simply whether or not it exists. That is, as soon we build it, it is deemed "not intelligent".

As the limitations of these models continues to recede, it feels like cold comfort to tell artists or programmers not to worry because GPT doesn't have "understanding" or whatever.

angusturner··on An unwilling illustrator found herself turned into an AI model
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 of patterns and regularity that generalise outside the training set. Most generated items will be novel, and most training items will not be reproducible.

It doesn’t seem radical to suggest that the copying issue will continue to recede as we get better models.

And there are areas of research specifically concerned with _provably_ showing that you cannot identify what items a model was trained was.

Lots of reasons to be optimistic in my view.

angusturner··on An unwilling illustrator found herself turned into an AI model
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 "copying" the training data, even if they can be coaxed into doing this for certain inputs (in the current iteration of the tools).

Ultimately, I think the problem of reproducing certain popular works or code snippets will be solved. One interesting direction here are the tools of information theory and differential privacy. e.g) proving that certain training inputs cannot be recovered from the weights, or that there is a threshold of how much information can be gleaned from a single input.

It is easy to imagine (because its nearly there already) a future version of StableDiffusion (or CoPilot) which provably compresses all training data beyond any possibility of recovery, and yet still produces extremely convincing results which disrupt the creative professions of art and programming.

Until we get to that point, it feels like the only consistent and sensible place to apply regulations is with the model end user. When I use CoPilot, I accept the small (honestly overblown) risk that _maybe_ I won't have the license to use some small snippet of code it spits out. But I'm happy to wear that responsibility, because the boost to productivity is so great that I dread a return to pre-CoPilot world. That is, a world where everyone keeps reinventing the same solution to simple problems over and over and over again.

angusturner··on GitHub Copi­lot inves­ti­ga­tion
the octopus would like a word with you
angusturner··on GitHub Copi­lot inves­ti­ga­tion
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 there are good reasons to suspect this is some part of what we are doing. (Even if its not implemented with back-prop).

e.g) consider the work of people like Anil Seth, who propose that our brain is basically a generative model of the world, which aims to minimize the likelihood of perceptual data. (see also: Karl Friston's free energy principle). What's up for debate is how it is structured, what priors are built in, what is the learning algorithm etc.

Anyway, for all their limitations, it seems clear that current artificial generative models can: 1. learn hierarchies of abstractions, which 2. explain the observed data in the fewest possible number of bits, and 3. generate new, novel data based on the patterns that have been learned

If you want to describe this as "just sophisticated pattern matching".. then sure I guess? But I think there's a clear qualitative difference between this and searching for code in a discrete database (which imo would not be okay).

angusturner··on GitHub Copi­lot inves­ti­ga­tion
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 tomorrow someone released a StableDiffusion, CoPilot etc with the same functionality, but respecting the provenance of the data (i.e. licensing etc), what concrete difference would this make? Programmers and other creative professionals would still (reasonably) be nervous about the implications for their livelihoods and communities.

At some point it will be possible to prompt a model for music in the style of <random artist>, and having never heard <random artist>, the model will generate a convincing emulation, based purely on statistical knowledge gleaned from millions of unrelated songs and text pairs. (I give it 5 years).

Now what? <random artist> should still be concerned (or not), but at least we're talking about the correct issue: How do we co-exist with generative models that massively disrupt/alter the process of doing creative or intellectual work?

← PreviousPage 3 of 4Next →