[hamdani yusuf (OP)]

Maybe but maybe those building blocks were not there and did not exist at all.

Assuming that it doesn't exist gives us nothing. Even nihilists get no advantage by doing so, but for different reason.
So, we would be better off assuming it exists, so we can make and continuously improve our models so we can make more effective and efficient plans to try achieving our terminal goal.

[hamdani yusuf (OP)]

AI is Evolving Faster Than You Think [GPT-4 and beyond]

In this episode, we take a deep look at the two weeks that changed the world. From GPT-4 to Google Bard, Midjourney v5 and even talk of AGI from Microsoft, it’s all right here.

Correction at 16:20: Upon taking a closer re-reading of the the statement it seems like the internal red team were more trying to cover their backs incase something goes wrong, not so much flat out saying they advise against release. Privately they could feel either way, wanted to just note that!

The problem of goal alignment is getting more urgent to solve and be trained to the AI as soon as possible.

[alancalverd]

We have built and continue to build a lot of Physics on the idea of particles existing.   Maybe they don't, it could all be waves and oscillations in some underlying field.

Not entirely true. We describe what we have observed in terms of particle and wave mathematics, but stating that x "is" y is the domain of philosophy and pointless vanity, not science.
My point is that we generally pursue our research whenever the observed universe doesn't behave exactly as our model predicts, so the answer to the original question is that we are as close to building a virtual universe as is necessary except for the following [insert unexplained phenomena here].

[hamdani yusuf (OP)]

The problem of goal alignment is getting more urgent to solve and be trained to the AI as soon as possible.

This is why.
GPT 4 Can Improve Itself - (ft. Reflexion, HuggingGPT, Bard Upgrade and much more)

GPT 4 can self-correct and improve itself. With exclusive discussions with the lead author of the Reflexions paper, I show how significant this will be across a variety of tasks, and how you can benefit.

I go on to lay out an accelerating trend of self-improvement and tool use, laid out by Karpathy, and cover papers such as Dera, Language Models Can Solve Computer Tasks and TaskMatrix, all released in the last few days.

I also showcase HuggingGPT, a model that harnesses Hugging Face and which I argue could be as significant a breakthrough as Reflexions. I show examples of multi-model use, and even how it might soon be applied to text-to-video and CGI editing (guest-starring Wonder Studio). I discuss how language models are now generating their own data and feedback, needing far fewer human expert demonstrations. Ilya Sutskever weighs in, and I end by discussing how AI is even improving its own hardware and facilitating commercial pressure that has driven Google to upgrade Bard using PaLM.

Here's the top comment to the video.

People: stop training models more powerful than GPT4
GPT4: improves itself

[Zer0]

The problem of goal alignment is getting more urgent to solve and be trained to the AI as soon as possible.

Here's the top comment to the video.

People: stop training models more powerful than GPT4
GPT4: improves itself

So...it has Emergence properties, it can Code & also utilize the Internet.

Then...What would Goals mean to it?
Making Sam Altman the Richest person?
Making OpenAI the Most Valuable Company?
Making the U.S. of A Impenetrable & Undefeatable?
OR
Uplifting the whole Human Species?

It's like having a Choice between making the World a Better place VS becoming the King/Queen & Ruling the World.
(i know what i would Choose, but Fear what the Other 8 Billion would Decide)

[hamdani yusuf (OP)]

So...it has Emergence properties, it can Code & also utilize the Internet.

Then...What would Goals mean to it?

Technically, goals of AI models are maximizing their reward function. In other words, they make decisions based on what they expect to produce highest value of their reward function, which were set by their developers (at least initially).
Effectively, goals of AI models as seen from outsiders' perspective would look like a condition which would make their reward function reach maximum value, based on whatever inputs put into the models.

ChatGPT is a natural language processing tool driven by AI technology that allows you to have human-like conversations and much more with the chatbot. The language model can answer questions and assist you with tasks like composing emails, essays, and code.


The model has many functions in addition to answering simple questions. ChatGPT can compose essays, describe art in great detail, create AI art prompts, have philosophical conversations, and even code for you.

https://www.zdnet.com/article/what-is-chatgpt-and-why-does-it-matter-heres-everything-you-need-to-know/

But the manually hard coded reward functions may not be really aligned to what were intended by their developers. Previous AI models had shown these cases, which made them shut down soon after they went on line.

The transformer architecture is a type of neural network that is used for processing natural language data. A neural network simulates the way a human brain works by processing information through layers of interconnected nodes. Think of a neural network like a hockey team: each player has a role, but they pass the puck back and forth among players with specific roles, all working together to score the goal.

The transformer architecture processes sequences of words by using "self-attention" to weigh the importance of different words in a sequence when making predictions. Self-attention is similar to the way a reader might look back at a previous sentence or paragraph for the context needed to understand a new word in a book. The transformer looks at all the words in a sequence to understand the context and the relationships between the words.

https://www.zdnet.com/article/how-does-chatgpt-work/

When we allow the AI models to modify (at least some of) their own reward functions, we need to define the highest priority goal that they can never change by themselves.

[alancalverd]

Ut semper GIGO.

[hamdani yusuf (OP)]

GPT 4 Can Improve Itself

GPT-4 with Self-Reflection beats GPT-4

In this video, We quickly talk about "Reflexion: an autonomous agent with dynamic memory and self-reflection".

This is a framework that allows AI agents to emulate human-like self-reflection and evaluate its performance on the ALFWorld and HotpotQA benchmarks.

[hamdani yusuf (OP)]

Before we even adapt to GPT4, the next generation is already on the way.

GPT-5 is Coming: What you NEED TO KNOW!

Today we break down what rumors and other details we know so far about GPT 5 and why it might be closer than you think.

00:00 Intro
1:04 GPT 5 Rumors
3:47 How will GPT 5 be better?
5:03 Plugins / The Everything App
6:26 Androids
9:09 Outro

[alancalverd]

So it can recycle its own excrement and produce ever more refined garbage.

At what point does artificial intelligence turn into artificial religion?

[hamdani yusuf (OP)]

Ut semper GIGO.

The data fed into the AI models are not exclusively garbage. Good AI models can filter out garbage and can recognize it effectively and efficiently.

[hamdani yusuf (OP)]

So it can recycle its own excrement and produce ever more refined garbage.

At what point does artificial intelligence turn into artificial religion?

At the point where they can't make corrections to their previous errors. But if that does happen, then their competitors are ready to take over. Their developers won't let that happen.

[alancalverd]

The definition of garbage being...? AI does not have the ability to test anything - it can only rely on published data. So it is likely to be led by the consensus of others and end up believing in fairies, anthropogenic global warming and the Zionist conspiracy. .

[alancalverd]

At the point where they can't make corrections to their previous errors.

Which is indistinguishable from getting the answer right.

[hamdani yusuf (OP)]

At the point where they can't make corrections to their previous errors.

Which is indistinguishable from getting the answer right.

It's the difference between false positive and true positive.

Remember, when you are dead, you do not know you are dead. It is only painful for others.
The same applies when you are stupid.
Ricky Gervais

[hamdani yusuf (OP)]

The definition of garbage being...? AI does not have the ability to test anything - it can only rely on published data. So it is likely to be led by the consensus of others and end up believing in fairies, anthropogenic global warming and the Zionist conspiracy. .

It depends on which data used to train them, their model architecture, and their access to real world.

[hamdani yusuf (OP)]

First hand information can always be helpful.

Ilya Sutskever (OpenAI Chief Scientist) - Building AGI, Alignment, Spies, Microsoft, & Enlightenment

Asked Ilya Sutskever (Chief Scientist of OpenAI) about
- time to AGI
- leaks and spies
- what's after generative models
- post AGI futures
- working with MSFT and competing with Google
- difficulty of aligning superhuman AI


Timestamps
00:00 Time to AGI
05:57 What’s after generative models?
10:57 Data, models, and research
15:27 Alignment
20:53 Post AGI Future
26:56 New ideas are overrated
36:22 Is progress inevitable?
41:27 Future Breakthroughs

[hamdani yusuf (OP)]

https://youtube.com/shorts/QX5Nw-rVF30
The dark side of AI art.
The accurate and accessible virtual universe will make it harder to assign false narrative to AI generated arts.

[alancalverd]

Why bother? The real universe provides all the facts and tests you need to do anything.

[Zer0]

Ut semper GIGO.

Ever seen lil kids watch a Parrot Talk for the first time ever.
Observed the Expressions of Awe & Wonder on their lil cutie faces.
That look of Bedazzlement in their lil eyes...it's Worth It!

🦜
(colorful parrot emoji)