Generative AI - Short & Sweet
Posts
The two most important AI news that have happened this year!

The two most important AI news that have happened this year!

Martin Musiol
July 30, 2024 • Estimated Reading Time: 14 minutes

Two of the tech advancements this year are underappreciated.

The unexpected: Google DeepMind’s Math System. I’m sure they’ll find a better name for that soon.
The foreseeable: Everyone can code today! ✨

With five months left in the year, another major AI breakthrough is likely to challenge our beliefs.

More about Elon, coding languages, and cancer detection is below.

Reading time is 5:21 min; Vamos!

The Unexpected: Google DeepMind's (GDM) Math System 🧮

GDM has built the most capable AI system for solving mathematical problems.

Background - AlphaGeometry

GDM released AlphaGeometry 1 in January 2024. It’s designed to solve complex geometry problems comparable to those tackled by a human gold medalist in the International Mathematical Olympiad (IMO).

What is the IMO? 
It's the most prestigious mathematical competition in the world, and very difficult. There are 6 problems that can earn 7 points each, a total of 42.

AlphaGeometry combines a symbolic deduction engine with a language model trained on synthetic data only.

Now, they have silently launched AlphaGeometry 2.

AlphaGeometry 2 solves 83% of historical IMO geometry problems from the last 25 years, outperforming its predecessor's 53% success rate.

In geometry, that's gold-medalist-level.

AlphaProof - the second piece

GDM has combined AlphaGeometry 2 with AlphaProof (AP). AP solves math problems with step-by-step proofs.

See an example of how Lean the formal language for proofs work.

Lean (formal language) for proofs.

First, AP translates problems from the informal natural language into the formal language Lean, which is used to verify mathematical proofs.

Second, AP utilises reinforcement learning (RL) to solve the problems, re-using here AlphaZero.

you can see that informal problems are formalized by the network, up to 100M problems have been formalized. Then there is a solver network that search for the solution.

Source: https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/

Putting it to test

The math system achieved a final score of 28 points.

Not easy - this year’s gold-medal threshold started at 29 points and was achieved by 58 of 609 pro contestants this year.

Noone has thought this is possible. See this manifold bet.

The fact that the program can come up with a non-obvious construction like this is very impressive and well beyond what I thought was state-of-the-art.

Prof. Sir Timothy Gowers, IMO Gold Medalist and Fields Medal Winner

In comparision, top-notch LLMs e.g., GPT-4.0, and Claude 3.5, struggle with the dumbest math. (E.g. counting letters in words.)

Why you might ask. Because of architectural differences incl. tokenisation, and lack of structural data (a whole topic by itself; LMK if you want to know more).

Future AI systems (or AGI systems) will probably require multiple subsystems, analog to how different brain areas function.

1. **Cerebrum**: The largest part of the brain, responsible for higher brain functions such as thought, memory, and decision-making. It includes: - **Frontal Lobe**: Controls reasoning, planning, movement, and emotions. - **Parietal Lobe**: Processes sensory information, including touch, temperature, and pain. - **Occipital Lobe**: Handles visual processing. - **Temporal Lobe**: Involved in hearing, memory, and speech. 2. **Cerebellum**: Located under the cerebrum, it coordinates muscle movements and maintains posture and balance. 3. **Brainstem**: Connects the brain to the spinal cord and controls vital life functions such as heartbeat, breathing, and blood pressure. It includes: - **Midbrain**: Involved in vision, hearing, and motor control. - **Pons**: Assists in regulating breathing and serves as a relay station between different parts of the brain. - **Medulla Oblongata**: Manages autonomic functions like heartbeat and breathing. 4. **Limbic System**: A set of structures involved in emotions, memory, and motivation. Key components include: - **Hippocampus**: Critical for memory formation. - **Amygdala**: Processes emotions such as fear and pleasure. 5. **Hypothalamus**: Regulates body temperature, hunger, thirst, and circadian rhythms. It also controls the endocrine system via the pituitary gland. These areas work together to manage a wide range of functions necessary for survival and interaction with the environment.

Source: https://askabiologist.asu.edu/brain-regions

What’s next?

GDM plans to continue scaling its AI models, optimizing them for better performance and broader applicability.

What will be possible? Step-by-step AI math systems will help us solve the unsolvable. A list of unsolved math problems. (I briefly thought I could solve the kissing number problem, but after 30 minutes, I realized it wasn't happening tonight or ever. 😄)

AI will help us solve most of these problems.

For example, one of the Millennium Prize Problems (you can win a million $ when you solve one of these) is the Riemann Hypothesis.

Solving it would have deep implications for the distribution of prime numbers. We would understand quantum physics in greater depth, likely leading to new techniques and methods in math, and breakthroughs in other areas.

It would also break most of current encryption tech -> new problems! 🫠

AI math systems, and eventually AGI math systems, will invent new math.

The ability to process vast amounts of data can explore inaccessible mathematical areas, accelerating knowledge growth and revealing new truths.

The Foreseeable: Today, Everyone can Code ! ✨

… and most people don’t know it yet.

If you can articulate your idea clearly, you can build it with AI!

AI has torn down the walls of coding. Not being able to code is no longer a limitation for developing things.

The quality of your ideas matters most today!

Anyone can build software. You are at the helm; AI is the developer.

My team and I have developed a framework that has been battle-tested on more than 25 real-world projects across various industries. Now, we're ready to share our proven method with you.

We're opening up our first batch of the online course.

You'll learn how to build anything from web apps to mobile applications to AI-powered solutions.

And the best part? I'll be there to guide you personally. ❣️

It's limited to 80 seats, and it's half the regular price. Plus, you'll get direct access to me through office hours. I'll help you apply what you've learned to your projects.

Join our waitlist NOW for this exclusive first batch and transform your ideas into reality!

If you know your way around code, you, of course, have an advantage, but it is no longer a complete blocker for non-coders.

As I am finalizing the course, is there something specific you'd like me to cover?

Certain implementation, application, wishes ...?

(Source) How energy efficient are coding languages?

Python is the worst 😭 .

Energy Efficiency in Programming Languages: A Comparative Overview Selecting the right programming language can significantly influence energy consumption, an increasingly crucial consideration in the tech industry. Understanding which languages are energy-efficient and which are not helps developers optimize applications for performance and sustainability. Energy-Efficient Programming Languages: C and C++: Known for their close-to-the-metal operations, C and C++ are highly efficient, offering fine-grained control over hardware resources. Their energy efficiency comes from low-level memory management and minimal runtime overhead. Rust: Rust offers safety and concurrency without sacrificing performance, making it a modern choice for energy efficiency. Its memory safety features reduce the likelihood of costly runtime errors. Go: Go is designed for simplicity and efficiency, with a minimalistic runtime and fast compilation times. Its garbage collection, while present, is optimized for performance, aiding in energy-efficient operation. Swift: Swift combines modern language features with efficient memory management, particularly in iOS development. It balances safety and speed, making it a good choice for energy-sensitive applications. Less Energy-Efficient Programming Languages: Python: While popular for its ease of use and versatility, Python's energy efficiency is hindered by its high-level abstraction and interpreter-based execution. This results in higher energy consumption compared to compiled languages. Java: Although Java offers portability and robustness, its Just-In-Time (JIT) compilation and garbage collection processes can lead to higher energy consumption, especially in resource-intensive applications. Ruby: Ruby, similar to Python, is known for its developer-friendly syntax but suffers from lower energy efficiency. Its interpreted nature and less optimized runtime contribute to higher power usage. Key Considerations: Performance vs. Development Speed: Often, there's a trade-off between the ease of development and energy efficiency. Languages like Python and Java offer rapid development cycles but can be less energy-efficient. Application Context: The choice of programming language should align with the specific needs of the project. For instance, systems programming or applications where performance is critical may benefit from using more energy-efficient languages like C++ or Rust. Optimizing for energy efficiency in programming not only helps reduce operational costs but also contributes to environmental sustainability, aligning tech development with broader ecological goals. As the industry advances, balancing efficiency, ease of use, and performance remains a key challenge and opportunity for developers.

Source: https://x.com/MarioBarbatti/status/1569939318478880769

(Source) Tesla and xAI

Elon Musk wants Tesla to invest in xAI (after a public poll).

Elon Musk has recently proposed that Tesla invest $5 billion into his artificial intelligence startup, xAI. This proposal has garnered significant attention and mixed reactions from various stakeholders. ### Background and Proposal Elon Musk, the CEO of Tesla, conducted a poll on the social media platform X (formerly known as Twitter), asking whether Tesla should invest $5 billion in xAI. The poll received nearly one million responses, with over two-thirds of participants supporting the idea[1][2][3]. Musk stated that he would discuss the potential investment with Tesla's board of directors and indicated that any such investment would require shareholder approval[3][7]. ### Rationale for Investment Musk has highlighted several potential benefits of investing in xAI for Tesla: - **Advancement in Self-Driving Technology**: Musk mentioned that xAI could help advance Tesla's self-driving technology, which is a critical area of focus for the electric vehicle manufacturer[2][3][7]. - **Integration of AI Tools**: The integration of xAI's chatbot, Grok, into Tesla's software could enhance user experience and operational efficiency[2][7]. - **Data Center Development**: xAI's expertise and resources could contribute to the development of Tesla's new data center[3]. ### Financial and Strategic Implications xAI has already raised $6 billion in a recent funding round, valuing the startup at approximately $24 billion[4][8][9]. The proposed $5 billion investment from Tesla would be a significant financial commitment, especially considering Tesla's recent financial performance, which included a fourth consecutive quarter of disappointing profits[1][7]. ### Concerns and Criticisms Several concerns have been raised regarding the proposed investment: - **Conflict of Interest**: Critics argue that Musk's dual roles in Tesla and xAI could lead to conflicts of interest. This is reminiscent of past controversies, such as Tesla's acquisition of SolarCity, which was also founded by Musk[2][6][10]. - **Resource Allocation**: There are worries that Tesla's resources, including AI talent and hardware, might be diverted to xAI, potentially undermining Tesla's core business[6][10]. - **Shareholder Impact**: Some analysts and shareholders are skeptical about the potential benefits of the investment, suggesting that the funds could be better utilized within Tesla to enhance its existing projects and technologies[6][10][13]. ### Next Steps Musk has indicated that the investment proposal will be discussed with Tesla's board of directors and will require shareholder approval. The outcome of these discussions and the subsequent vote will determine whether Tesla proceeds with the $5 billion investment in xAI[3][7][11]. In summary, while the proposed investment in xAI could offer strategic benefits to Tesla, it also raises significant concerns about conflicts of interest and the optimal use of Tesla's financial resources. The decision will ultimately hinge on the approval of Tesla's board and shareholders.

Photo: Apu Gomes (Getty Images)

P.S.: EM mentioned in an interview with Jordan Petersen that Grok (xAI’s AI) hasn't been trained with Tesla's driving video data. What could an AI learn from processing PetaBytes of such data?

(Source) AI detects cancer via sugar analysis

The AI, called Candycrunch, quickly analyzes complex cell sugar structures to uncover cancer indicators more accurately than current methods.

Scientists to focus on interpreting results and developing new hypotheses.

Overview of cancer markers for sugar chains.

Source: https://www.sciencedirect.com

That’s it for now.

I’m in Switzerland rn, and can’t recommend that country enough. Especially, Valle Verzasca. Look at this:

Ponte dei Salti, a historic Roman bridge over the clear waters of Vale Verzasca, Switzerland, with scenic mountain views.

Please feel free to spread the word about the course.

Martin

Everyone can code! ✨ —> Landing page.
Want to AI-upgrade your customer service? Contact us.
Spread the word, get the perk! Referral program.
Would you like to sponsor a post?
My book - Generative AI: Navigating the Course to AGI.
My webpage has changed (a lot is planned 😉).