17 Comments
User's avatar
Daniel Sangyoon Kim's avatar

Thank you for sharing Chamath. Good read. I enjoyed most of it, but did feel it was biased in favor of grok without sufficient backing.

For example:

“Having access to this data in multiple formats such as images and audio can also help xAI’s model achieve a deeper and more nuanced understanding of the world.”

Doesn’t openAI, bard and others also have access to image and audio data?

Also you say grok has distribution advantage, but how is that any better than the distribution google may have through all their products or Meta with billions of users talking through its platforms?

I see grok as “different” and valuable, but the logic you provided hasn’t convinced me it has a true advantage VS others.

Expand full comment
Sivacharan's avatar

Good grok puff piece disguised as LLM explainer. Somehow Chamath and co with all their “intellectual honesty” can find nothing wrong to say about Elon on pods. You are no different to the media that you criticise constantly.

Expand full comment
Aidan Szwec's avatar

Thanks for this! The race to AGI is getting wild with data, real-time smarts, and Grok's unique edge. Eager to see where this tech rollercoaster takes us next!

Expand full comment
Matthew Harris's avatar

A great complement to this article is this video from Andrej Karparthy, a leading engineer at OpenAI

https://youtu.be/zjkBMFhNj_g?si=m9bchYRu6Z-M_zRv

Expand full comment
Nancy Henson's avatar

Great article, simplifying a very complex topic. (Almost as good as my husband’s Blockchain 101)

For the non-grokers, here is my take:

I and my myriad business peers and friends, post on X. (Caveat that I have not posted in recent years) We all have nuanced opinions on subjects. Generally, we are not authors and have minimal or no footprint on the internet outside of social.

As such, X does provide the data for deeper learning.

Now the question is which social platform could provide the best data? X? (Run to the hills if someone is using TikTok to train their beast)

So now, there is increased competition to grow a social platform that fosters meaningful conversation and debate. 😎

Expand full comment
Ari's avatar

Very clear breakdown! Thanks for putting this together!!

Expand full comment
@cryptensen's avatar

Well written and explained.

We are currently contemplating sending our mental health AI to med school, any thoughts?

We buildt a MS mental health GPT.

Expand full comment
Jean-Marc's avatar

Great and helpful summary! Thanks for writing it and sharing. I need to try Grok

Expand full comment
isura Morawaka's avatar

First half gave me a good understanding of LLMs, second half was him shilling grok to me; which might have worked.

Thank you, it was a good read.

Expand full comment
Sheryl's avatar

Wonderful. Shades of VLDB and ACID. Waitiing for the bias in the training to become evident, as it will, particularly as it is used in healthcare.

Expand full comment
Quratulann Akbar's avatar

Really well written! Enjoyed reading it :)

Expand full comment
Fal Diabaté's avatar

Thanks Chamath for sharing

Expand full comment
Patrick Reynolds's avatar

This is super well explained!

I'm enjoying Stanford's course on NLP with Deep Learning which is a useful introduction.

Expand full comment
Tom's avatar

Thank you for providing your insights, Chamath. I found your input to be informative and engaging. While I appreciated much of the content, I did perceive a bias towards Grok that, in my opinion, lacked adequate support.

Expand full comment
Anne Canfield's avatar

I would like to bring to your attention a significant development in computing power: Intelligent Computing. Intelligent computing systems are advanced computer systems that integrate technologies like artificial intelligence (AI), machine learning, big data analytics, and the Internet of Things (IoT) to process vast amounts of data, learn from it, and make autonomous decisions. These systems are designed to mimic human cognitive abilities such as perception, reasoning, and problem-solving, often operating in real-time to adapt to dynamic environments. They are distinct from traditional computing by their ability to handle complex, unstructured data and provide insights or actions with high efficiency and accuracy. Applications span industries like healthcare, transportation, manufacturing, and finance, enabling tasks such as autonomous vehicles, predictive maintenance, and high-performance data analytics.

Key characteristics of intelligent computing systems include:

o Data Processing: They aggregate and analyze large datasets from diverse sources, often in real time, to extract actionable insights

o Learning and Adaptation: Using AI and machine learning, these systems improve performance over time by learning from new data and experiences.

o Interconnectivity: Integration with IoT and cloud computing enables seamless communication between devices and systems, supporting applications like smart cities and Industry 4.0.

o High Performance: Leveraging specialized hardware like GPUs and AI chips, they deliver superior computational power for tasks like deep learning and simulations.

In summary, Intelligent Computing, characterized by an ecosystem of interconnected, open, and self-governing AI models, offers transformative solutions for users open source and security challenges. By leveraging a flexible, cloud-native architecture—such as the Intelligent Computing Fabric—users can integrate open-source systems and enhance security at scale. Intelligent Computing benefits include:

o Open-Source Integration:

 Codeless Intelligence: Simplifies compliance by enabling non-technical users to design and document open-source systems via intuitive interfaces. This reduces the documentation burden, a key barrier for open-source providers, by automating the creation of compliant artifacts like System Security Plans (SSPs).

 Cloud Intelligence: Supports open-source ecosystems through open cloud architectures (e.g., Kubernetes-based Infrastructure as Code). This ensures open source CSOs can scale rapidly, self-heal, and integrate with FedRAMP’s OSCAL framework, for example, making compliance more accessible to decentralized development communities.

 Ecosystem Approach: Encourages collaboration between open-source communities and federal agencies by chaining multiple models to audit and enrich code. This addresses vetting concerns by providing transparent, automated vulnerability scans and compliance checks, fostering trust in open-source solutions.

o Security Enhancements:

 Cyber Intelligence: Implements zero-trust, quantum-safe security via Active Cyber Intelligence and Defense (ACID). This continuously monitors open source and proprietary systems for threats, adapting protections in real-time to counter vulnerabilities, thus alleviating agency concerns about open-source code quality.

 Automated Compliance: Uses AI-driven tools to streamline continuous monitoring and assessment processes, reducing manual effort and human error. This ensures open-source systems maintain FedRAMP’s and other users' rigorous security standards while accelerating authorization timelines.

 Scalable Resilience: The Intelligent Computing Fabric’s ecosystemic design accommodates diverse technologies, enhancing system resilience. By integrating models that learn from each other, it proactively identifies and mitigates security risks across open source and proprietary CSOs, ensuring robust protection.

For more information, please see this article: https://www.politico.com/sponsor-content/2024/10/21/powering-radical-change-with-intelligent-computing

Expand full comment
Matt McDonagh's avatar

Enjoyed your visuals a great deal.

Using a graph analogy to describe how related words are mapped closely together in vector space is helpful for visualizing how word embeddings work.

This analogy simplifies the concept of high-dimensional vector spaces without losing meaning.

It clearly conveys that semantic similarity between words is reflected in their proximity within this space.

Expand full comment