Stopping the 'Digital God' Cult: If it's anthropomorphic, don't fund it

[ygolo]

More thoughts countering the "scale it and benefits will come" cult.

There are many forms of reasoning. But on one form of reasoning, deductive reasoning, we created extremely sharpened forms of quality control (usually involving symbols):
1) Mathematics - considered the science of patterns. The proof (a mechanism developed over centuries) is a means to create high quality deductive reasoning about patterns. There are attempts to automate it or formalize it called Theorem Provers, the most popular being Lean and Rocq.
Here is one video of Terence Tao (arguably the world's best living Mathematician) using LLMs with Lean (There are others):

2) Computers and Software - Initially just a split from Math but now used everywhere. We've gone from humans being computers, to tapes, to punch cards, to assemblers, to compilers, to code generators, no-code tools, and now AI. This is an amazing match for use with AI. It's an evolution of things that have been happening for a long time.

In both these cases, you really don't need or want to regenerate results in math or computer science. It is incredibly wasteful - you want a copy. There's an old joke about reducing these problems to already solved problems.

But when you see the most energy inefficient, and most used areas for these large scale AIs, it is computing and math. This, to me, suggests that building an appropriate "database" of deductive arguments (say using Lean or Rocq, and of code snippets for standard problems) that then gets served to the whole public is the most societally energy efficient way to build a system for this purpose.

 

[ygolo]

Some signs that some thought is being put into mitigate economic impacts and gain benefits:

https://openai.com/global-affairs/openais-economic-blueprint/

As job losses loom, Anthropic launches program to track AI's economic fallout | TechCrunch

As warnings mount about AI’s potential to displace millions of jobs, Anthropic on Friday launched a its Economic Futures Program, new initiative to support research and policy development that addresses AI’s economic impacts.

techcrunch.com

 

[ygolo]

Fine-tuning for particular goals, still seems to yield results:

Large language models to accelerate organic chemistry synthesis - Nature Machine Intelligence

Large language models (LLMs) can be useful tools for science, but they often lack expert understanding of complex domains that they were not trained on. Zhang and colleagues fine-tuned a LLaMA-2-7b-based LLM with questions on organic chemistry reactions.

www.nature.com

Accelerating scientific discovery with AI

FutureHouse, co-founded by Sam Rodriques PhD ’19, has developed AI agents to automate some of the most critical steps on the path toward scientific progress.

news.mit.edu

 

[chubber]

The ontological power of the elite is inevitable. I always interpreted Albert Einstein's rumoured interpretation of Isaac Newton, who was heavily into secret societies. Predicted end of the world at 2060, and I always interpreted that as the world as we know it, comes to an end. A new world will emerge, but I think these humans are not going to be humans like us, my off-the-cuff interpretation is, they will be bio engineered humanoids that complies and will be complacent of their environment, Similar to how the elite would play with mechanical humanoids at first but would long for biological instead, almost like how in the slave-world era, an upgrade in life style, society, cast, rank, was considered "up" if you moved from the plantation to working in the house.

Seems like the believers are simply preparing or mimicking themselves to be worthy of such a placement.

 

[ygolo]

The ontological power of the elite is inevitable. I always interpreted Albert Einstein's rumoured interpretation of Isaac Newton, who was heavily into secret societies. Predicted end of the world at 2060, and I always interpreted that as the world as we know it, comes to an end. A new world will emerge, but I think these humans are not going to be humans like us, my off-the-cuff interpretation is, they will be bio engineered humanoids that complies and will be complacent of their environment, Similar to how the elite would play with mechanical humanoids at first but would long for biological instead, almost like how in the slave-world era, an upgrade in life style, society, cast, rank, was considered "up" if you moved from the plantation to working in the house.

Seems like the believers are simply preparing or mimicking themselves to be worthy of such a placement.

We can still fight that power, however.

As nations build 'sovereign AI,' open-source models and cloud computing can help, experts say

Panelists at CNBC’s East Tech West 2025 in Thailand discussed how emerging nations can begin to build their own “sovereign AI."

www.cnbc.com

 

[ygolo]

^Someone with a take that may even be less fear-hypey than me.

 

[ygolo]

I hadn't checked on it for a while.

The definition of foundation models in CA SB 53 now seems much more reasonable, and less presumptive that only AGI happens in AI.

SB 53: CalCompute: foundation models: whistleblowers. | Digital Democracy

Digital Democracy overview of bill SB 53: CalCompute: foundation models: whistleblowers.

calmatters.digitaldemocracy.org

(e)Foundation model means an artificial intelligence model that is all of the following:

(1)Trained on a broad dataset.
(2)Designed for generality of output.
(3)Adaptable to a wide range of distinctive tasks.

Hopefully, this opens the path for people coming up with new ways to simulate biology, physics, or to do math --narrowly focused on important but difficult problems, can do so without worrying about lawyers and legal fees before proceeding.

 

[ygolo]

This gets to the core of the jaggedness of "Artificial Jagged Intelligence "

 

[ygolo]

A nuanced take on artificial intelligence, that's fairly convincing:

 

[ygolo]

A New Kind of AI Model Lets Data Owners Take Control

A novel approach from the Allen Institute for AI enables data to be removed from an artificial intelligence model even after it has already been used for training.

www.wired.com

 

[ygolo]

This gets to the core of the jaggedness of "Artificial Jagged Intelligence "

People have been fawning over the amazing results of Grok 4 (on benchmarks).

But this precisely the problem with the current paradigm. You can't test your way to high quality software. Engineers know this inherently. The prior assumption that makes thorough testing lead to higher quality and reliable results is the underlying understanding of the engineer doing the testing.

Just like a no hypothesis predictor ought to be considered post-hoc hypothesizing, a pure-predictor leading to world models suffers the same fate.

What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models

Foundation models are premised on the idea that sequence prediction can uncover deeper domain understanding, much like how Kepler's predictions of planetary motion later led to the discovery of Newtonian mechanics. However, evaluating whether these models truly capture deeper structure remains a...

arxiv.org

Attention is not all you need. You need curiosity (with proper extrapolation hypothesis creation, not just interpolation)