Random AI/Robot Thoughts and News

[ygolo]

I have wondered if quantum processors will lead to AI sentience. While we don’t understand it in humans, I’ve heard speculative theoretical physicists musing it could be a function of quantum physics.

My understanding is that consciousness is a property of matter that is expressed when the conditions are right. I’ve wondered if that property of matter is what drives evolutionary processes. I find it absurd to assume it’s an accident or anomaly. I don’t see a reason those conditions could not be intentionally created. I don’t see it as spiritual in the magical sense but like an inherent property or force common throughout the universe like any other property of matter.

“We are the universe observing itself”

That's Chalmers "Hard Problem" view. I don't place much credence in it.

I might suggest Anil Seth's book, Being You. It's an extremely fact based deconstruction of consciousness. You'll actually see similarities to what an LLM does.

The issue is that LLMs, even naive multi-modal models with the same architecture are limited. Even those who went hard(many open AI and Anthropic researchers) on the pre-trained scaling "laws" of the transformer architecture are admitting seeing diminishing returns from scaling pre-trained models. They're turning to inference-time compute, tool use(agents), world models, and improved product design.

 

[ygolo]

Related events:

https://thehill.com/policy/technology/4997998-government-commission-proposes-manhattan-project-style-ai-funding-plan/

DeepSeek’s first reasoning model R1-Lite-Preview turns heads, beating OpenAI o1 performance

The company’s published results highlight its ability to handle a wide range of tasks, from complex mathematics to logic-based scenarios.

venturebeat.com

 

[Lark]

What AI revolutions are you anticipating that you personally would be looking forward to?

I would love it if there were a scheduler revolution, like if you could provide a sophisticated AI your schedule information which you have, work, leisure, interests, things you are curious about which have not developed an interest in yet but want to, and it would create a perfect schedule for you pulling in all the information of the web, a lot of the juggling and stress associated with it would disappear and you'd never miss a deadline for a car service, dental appointment or reach the end of the working year with additional annual leave to carry over or anything like that.

It would be cool to see how it would suggest what you could cut and keep in a schedule too in order to optimize different things.

I think until this kind of thing would work on an individual or personal level it wont work on a structural or service management level because it requires a degree of familiarity on the part of both provider and consumer/user for it to work and I think that'd only work if some sort of primary socialisation / familarity with the systems existed already.

 

[ygolo]

@Lark I think what you want could be done with a recommendation system with proper inputs from you on constraints and preferences.

Part of what I hate about most recommendation systems are that they're controlled by companies and not people. Imagine if the data on you was owned and controlled by you.

GitHub - grahamjenson/list_of_recommender_systems: A List of Recommender Systems and Resources

A List of Recommender Systems and Resources. Contribute to grahamjenson/list_of_recommender_systems development by creating an account on GitHub.

github.com

 

[ygolo]

This juxtaposition is why regulations need to be application and domain-specific.

In science, the difference between finding information and publication is huge.

AI-generated images threaten science — here’s how researchers hope to spot them

Generative-AI technologies can create convincing scientific data with ease — publishers and integrity specialists fear a torrent of faked science.

www.nature.com

OpenEvidence

The world's leading AI-powered medical search platform.

www.openevidence.com

 

[Lark]

@Lark I think what you want could be done with a recommendation system with proper inputs from you on constraints and preferences.

Part of what I hate about most recommendation systems are that they're controlled by companies and not people. Imagine if the data on you was owned and controlled by you.

GitHub - grahamjenson/list_of_recommender_systems: A List of Recommender Systems and Resources

A List of Recommender Systems and Resources. Contribute to grahamjenson/list_of_recommender_systems development by creating an account on GitHub.

github.com

Indeed, I used to talk to a few friends about a sort of alternative to facebook, much more detailed but its within the user discretion what the complete and what is visible or not and then they have a "sell my data" option, I dont know how it would be funded if it wasnt automatically for sale, maybe ads, I dont know but the point of it would be that the "sale" would create revenue for the user, a percentage would be taken for the site in its role as "host", but the idea being that it would be a major data collection source, less of a social network as a life logger to which you own and control, you can use the metrics entirely yourself or you can choose to "share" or "sell" if as you see fit.

It was something I thought of as a good idea to create as a kind of private UBI type thing, everyone creates data these days, just by living and experiencing stress, disease, epigenetics etc.

The benefits to the individual and also socially would be immense.

 

[ygolo]

https://www.sfgate.com/tech/article/stanford-professor-lying-and-technology-19937258.php

The irony

 

[ygolo]

Global AI Power Rankings: Stanford HAI Tool Ranks 36 Countries in AI

From patents to private funding, Stanford’s Global Vibrancy Tool aggregates 42 indicators to reveal which countries lead in AI – and why the U.S. holds a significant edge over other nations.

hai.stanford.edu

 

[ygolo]

PR Puppet Sora - a Hugging Face Space by PR-Puppets

Discover amazing ML apps made by the community

huggingface.co

 

[The Cat]

Fever Dreams of electric sheep.
​

 

[ygolo]

Train Your Brain to Work Creatively with Gen AI

When most people prompt generative AI, they do so within the paradigm of how they think about what could or should come next. For example, when searching Google, users may ask a question, search for the best Thai restaurant “near me,” or insert specific criteria based on filtered output, such as...

hbr.org

 

[ygolo]

Fever Dreams of electric sheep.
​

Have you tried using Runway?

ImagineArt Video Studio

Simply describe your vision with words, and our advanced AI will transform your text prompts into engaging videos. It's text-to-video made easy!

www.imagine.art

I have friends interested in generative art and video. I have used it on occasion, but I haven't been impressed with its ability to give me what I want. The tools are helpful but they still don't fully impress me yet(maybe I haven't experimented with them enough).

I have more use for LLMs, and segmentation models. I customize many of them for my purposes.

Edit:
Here is a video I just made of a cat floating in space:

A_cat_floating_around_in_space._1f6f2daf-3710-4689-8d4e-1c7fb400cf0f

This is "A_cat_floating_around_in_space._1f6f2daf-3710-4689-8d4e-1c7fb400cf0f" by Renrael Trepxe on Vimeo, the home for high quality videos and the people…

vimeo.com

 

[ygolo]

4 Ways To Reap The Rewards Of AI Agents While Avoiding The Risks

We can deliberately manage AI agents to get the most from them instead of letting them take control of us. Here are 4 ways to start.

www.forbes.com

 

[ygolo]

The Open-Source Toolkit for Building AI Agents

Curated frameworks, tools, and libraries every developer needs to build functional and efficient AI agents

www.aitidbits.ai

AI created a Minecraft AI village with up to 1,000 inhabitants — Project Sid sees AI bots implement a taxation system and spread Pastafarianism religion

This includes "players" dynamically picking village-appropriate roles and spreading the word of "Pastafarianism"

www.tomshardware.com

 

[ygolo]

 

[ygolo]

Welcome to LLMflation - LLM inference cost is going down fast ⬇️ | Andreessen Horowitz

For LLM of equivalent performance, the inference cost is decreasing by 10x every year. What cost $60/million tokens in 2021 costs $.06/million tokens today.

a16z.com

^^^The scaling law that actually matters.

I have held a belief that the pre-trained emergent capabilities scaling law(a belief held by AGI companies) was always going to cap out due to the entropy of the available training data (especially if you believe the theoretical underpinnings being the existence of lower dimensional manifolds for concepts being more available as dimensions increase)

 

[ygolo]

I am most interested in world models lately.

World Labs

World Labs is a spatial intelligence AI company building Large World Models (LWMs) to perceive, generate, and interact with the 3D world.

www.worldlabs.ai

I think a lower level of quality can be achieved using good segmentation models.

SAMURAI: Adapting Segment Anything Model for Zero-Shot Visual Tracking with Motion-Aware Memory

Humans (and other animals) use things like parallax and combining vision with sound to really understand our environment.

 

[ygolo]

Quiet surge: China's AI innovators doing more with less - Asia Times

This trend is not new, but a surge in announcements over the last few weeks reconfirms how the Chinese models are not only keeping pace with the rest but

asiatimes.com

 

[ygolo]

I think that:

Pre-trained auto-regressive Large Language Models will top about at some point because of the inherent entropy of language.

However, I believe that the transformer architecture (and various efficiency gains on top of it) could be enough to create human-level intelligence. However, multi-modality will be key.

Guessing the next token in vision, touch, movement, sound, and language is, in some sense, the "Cortical Algorithm" proposed by Jeff Hawkins in On Intelligence almost two decades ago.

Only one Tech Mogul has the computing power and data to train such a model. My question to people: "Do you want him to control this?"

Almost every person trying to build such a thing says something like: "I will need to create it because others who build it will misuse it and make it bad."

I prefer different people to control various aspects of the new intelligence. If we are using all of human knowledge to create a new intelligence, I feel all humans should have some say.

Openness needed for AI Safety, the battle for the future of AI

For the smaller groups, communities, and individuals who would not want cronyism to take root in a technology that could be "more transformative than electricity": https://open.mozilla.org/letter/

www.typologycentral.com

 

[ygolo]

https://web-assets.bcg.com/8c/26/b80dfaa64b1d92bed7b64d2e19dd/ai-at-work-what-people-are-saying.pdf