Good Readings 2021

Hello all! This is the continuation of topic Good Readings 2020, for the year 2021. It is for sharing deep learning papers that are important, interesting, or otherwise worth reading.

9 Likes

image

4 Likes

Mixup without Hesitation:


2 Likes

AutoFreeze: Automatically Freezing Model Blocks to Accelerate Fine-tuning

code: https://github.com/uw-mad-dash/AutoFreeze.

@muellerzr

5 Likes

This looks very interesting indeed. :smiley: :smiley:

1 Like

This is very cool!

1 Like

Zero-Shot Text-to-Image Generation (DALL-E paper)

Partial code:

1 Like

This essay by Sam Altman of OpenAI explores the immense social consequences of the work we are doing here. AI will inevitably replace labor, displace jobs, and create vast amounts of wealth. To whom will these benefits go? In our current system, to corporations, shareholders, and billionaires - to the owners, not to all of society. How can we address this issue and have our actions help create the world we wish to live in?

Note: The essay is USA-centric, but its points will apply to to many societies.

4 Likes

Hi Pomo hope all is well!

Can you think of any governments that are likely to or are currently taking any of the routes suggested in the essay?

I’ll keep hoping for the best.

Cheers mrfabulous1 :smiley: :smiley:

Hi @mrfabulous1,

I do appreciate your positive and encouraging posts across these forums.

You ask a very provocative question. I am not sure what I believe and therefore how to answer.

With a critical, evaluative eye, I do not see any government taking the suggested routes. The USA government is already captured by corporations and billionaires via their money and propaganda. Money begets more power and power begets more money. I do not see those interests voluntarily giving up the increased profits and power offered by AI, even if to save from collapse the whole system that supports them. What I see historically is that when income, assets, and power get very out of balance, there is inevitably a revolution that tears down the whole sick system. Lots of people get hurt, lots of people die, we start over.

At the same time, like you I hope for the best. Bill Gates has proposed taxing the equivalent wages of the worker replaced by a robot. It’s a more palatable step than the big ones in the essay. Maybe the AI Revolution will all settle out in a few decades like the Industrial Revolution, with benefits spread across society. Maybe there will be a revolution in human consciousness. Maybe AIs will turn out to be the next step in our evolution. Maybe a meteor or a pandemic…

In any case, we who are learning and inventing here are right in the middle of the coming changes. That implies having some responsibility for the outcomes, IMO.

Cheers to you, Mr Fab! :slightly_smiling_face:

P.S. Can anyone think of ways that AI/machine learning can help shift power to ordinary people rather than to corporations and governments? Some kind of collective action perhaps, price negotiation with Amazon, measurement of external costs of policies, etc.?

1 Like

You’re not alone! :smiley:

Sadly this is similar to the conclusion I drew when I read the article and it is this that caused me to raise the question.

This is I think is a good Idea, ultimately coupled with some sort of universal wage if very few jobs are available.

Now this I would like to see in a book I recently read called the “The Economic Singularity Artificial Intelligence and the Death of Capitalism”.
The author talks about various possible scenarios. My personal oppinion is that “With great power comes great responsibilitty” more than ever, now is the time for all humans to live and be ethical as we can, or the machines we create will not be ethical.

Now this idea also appeals to me, if evolution means we connect to AI as individuals either mentally or physically like this cyborg,
more detail and we have the freedom to augment ourselves to become super beings i am ready. :smiley:

One thing that could lead us towards the above, is for more iniatives like Fastai which aims to domocratize AI, but in my oppinion fastai will need to be abstracted, so you do not even need one years worth of python programming eperience to create a model.

Seldom do ordinary people control power, as you stated earlier they have it for a moment during phases of revolution. Then just like in the [French revolution]( French Revolution - Wikipedia, the Russian revoluttion and George Orwells Animal Farm The revolutionaries morph into goverments and corporations and the cycle starts again ethical benevolent leaders looks like the best option.

Cheers mrfabulous1 :smiley: :smiley: