AI Progress // Ex-OpenAi Engineer Document

[CARDOZZO]

Are you serious about AI?

You should read it https://situational-awareness.ai/ 

In this document, we have a roadmap to AI on the next 10 years.

It is alarming/amazing to read.

We don't understand AI. I assume only 1000 people really understand AI on the world.

It will change EVERYTHING.

[CARDOZZO]

PDF

https://situational-awareness.ai/wp-content/uploads/2024/06/situationalawareness.pdf

Edited June 15 by CARDOZZO

[Rafael Thundercat]

Wow, I will print this and read well

[Leo Gura]

I posted a video summarizing it. It's on my blog.

Edited June 16 by Leo Gura

[undeather]

It's interesting, but I would be very catious in taking those aspergeresque bay-area-prophecies at face value.
If we have learned anything from past predictions, things will usually turn out unpredictably different.

I think Hossenfelder made a good video about his claims:

 

[Girzo]

This is terrible just from the beginning of the text. No fucking way US regions were this tech is being developed have enough water to cool all those GPUs.

[CARDOZZO]

8 hours ago, Leo Gura said:

I posted a video summarizing it. It's on my blog.

Wow, i see that later, thanks Leo!

[CARDOZZO]

We will have inferiority complex related to AI/Robots.

The path is leading to cyborgism & transhumanism.

Do you want a taste of the future?

Read "Diaspora" & "Permutation City" by Greg Egan. (sci-fi books)

[Rafael Thundercat]

Some insigths from the page LessWrong

https://www.lesswrong.com/posts/q8uNoJBgcpAe3bSBp/my-ai-model-delta-compared-to-yudkowsky

[M89]

On 2024. 06. 15. at 2:33 PM, CARDOZZO said:

Are you serious about AI?

You should read it https://situational-awareness.ai/ 

In this document, we have a roadmap to AI on the next 10 years.

It is alarming/amazing to read.

We don't understand AI. I assume only 1000 people really understand AI on the world.

It will change EVERYTHING.

Sorry, but this essay is BS from the get go.

The future visioned here won't happen (at least not in 10 years) because even the assessment of the present situation is massively wrong.

It supposes that we basically solved "intelligence" with the LLM architecture and we just need to scale it up and do some algorithmic refinement. It supposes that GPT-4 is on par with a smart high schooler regarding intelligence. This is far from the truth.

GPT-4 has very little intelligence, if at all. It is a relatively simple process based on stochastic data-retrieving and shallow pattern matching. But the lack of intelligence is more or less compensated with the gigantic amount of data it was trained on. Given certain boundary conditions it looks like intelligence but it's not. And scaling it up will not lead to AGI, that is to "real" intelligence that humans have.

"We don't understand AI. I assume only 1000 people really understand AI on the world." - Hence the hype.

Actually many people understand how these AI models work, it's not a mystery. What they don't understand is how their minds work and what intelligence is. And they don't understand their own projections on AI, either.

"It is alarming/amazing to read." - Yes, like a sci-fi, but I cannot take it seriously.

Instead, I would suggest Francois Chollet's writing 'On the Measure of Intelligence' (https://arxiv.org/pdf/1911.01547). He clarifies some basic distinctions about intelligence.

 

 

[Ero]

Courtesy of xkcd. Mira Murati (current CTO of OpenAI - not fired after a year and something like the guy who wrote this 165 page treatise) recently said they don't have anything in development that is much ahead of what is already open sourced. I suggest you guys familiarize yourself with the position of Yann LeCun, a heavyweight compared to this kid, and you will see why the majority of people in the field are not taking this guy seriously.  

[CARDOZZO]

Of course, it is just a vision.

I don't think we will achieve AGI with LLMs.

I think it is more reasonable to achieve AGI through Robotics (Embodied AGI).

[Rafael Thundercat]

 

[Rafael Thundercat]

Funny Truths

[CARDOZZO]

17 minutes ago, Rafael Thundercat said:

Funny Truths

😂

[OBEler]

What do you think about the conclusions of this document about upcoming super intelligence?

How will you prepare for that ?

[Jannes]

On 21.6.2024 at 1:41 PM, Rafael Thundercat said:

Funny Truths

But that works both ways. If you have accomplished things so far you have built an ego around that and see AI through a bias lens. Your bias would favor humans over AI.

[CARDOZZO]

36 minutes ago, OBEler said:

What do you think about the conclusions of this document about upcoming super intelligence?

How will you prepare for that ?

I believe on futures (yes, plural)

We'll have cyborgs, transhumanists, normal humans, robots, embedded AIs, cyberpunks.

You'll have to choose your own paths.

Our skill to co-create with AI will be limitless.

If a human tell you that they can precisely predict the next 3 years, he's completely wrong.

You can search for future signals, technology potentials and paradigms to trace possible scenarios.

Edited June 23 by CARDOZZO

[CARDOZZO]

Pioneers, Visionaries, Inventors, Concept-Makers, Zero Principles Thinker, First Principles Thinker, AI Co-Creators.

Our skill to give instructions to AI will be one of the most valuable skills.

Precise, Concise, Clear Thinking/Communication.

Superintelligent AI + Robots will have the power to manifest human visions at a deep scale extremely fast.

We can send AI humanoid robots as drones to explore different planets.

[Rafael Thundercat]

27 minutes ago, CARDOZZO said:

We can send AI humanoid robots as drones to explore different planets.

We can send Marçal as well hehe. Sorry, Side Joke.