AI: DeepSeek-R1@ Amazon Bedrock & SageMaker

 DeepSeek-R1 model now available in Amazon Bedrock Marketplace and Amazon SageMaker JumpStart | AWS Machine Learning Blog

"DeepSeek-R1 is a large language model (LLM) developed by DeepSeek-AI that uses reinforcement learning to enhance reasoning capabilities through a multi-stage training process from a DeepSeek-V3-Base foundation. A key distinguishing feature is its reinforcement learning (RL) step, which was used to refine the model’s responses beyond the standard pre-training and fine-tuning process.

DeepSeek-R1 uses a Mixture of Experts (MoE) architecture and is 671 billion parameters in size. The MoE architecture allows activation of 37 billion parameters, enabling efficient inference by routing queries to the most relevant expert “clusters.” This approach allows the model to specialize in different problem domains while maintaining overall efficiency. DeepSeek-R1 requires at least 800 GB of HBM memory in FP8 format for inference. In this post, we will use an ml.p5e.48xlarge instance to deploy the model. ml.p5e.48xlarge comes with 8 Nvidia H200 GPUs providing 1128 GB of GPU memory."

Quantum processor: Microsoft Majorana 1

Microsoft’s Majorana 1 chip carves new path for quantum computing - Source

Microsoft unveils Majorana 1, the world’s first quantum processor powered by topological qubits - Microsoft Azure Quantum Blog

"the world’s first Quantum Processing Unit (QPU) powered by a Topological Core, 

designed to scale to a million qubits on a single chip."

Majorana 1 Explained: The Path to a Million Qubits - YouTube

AI/ML: Overfitting vs Underfitting

Model Fit: Underfitting vs. Overfitting - Amazon Machine Learning

What is Overfitting? - Overfitting in Machine Learning Explained - AWS

Underfit models experience high bias—they give inaccurate results for both the training data and test set. 

On the other hand, overfit models experience high variance—they give accurate results for the training set but not for the test set. More model training results in less bias but variance can increase. 

Data scientists aim to find the sweet spot between underfitting and overfitting when fitting a model. A well-fitted model can quickly establish the dominant trend for seen and unseen data sets.

xAI: Grok3 Launch

XAI LAUNCHES GROK 3 [FULL REPLAY] - YouTube (Wes Roth)

xAI’s Mind Blowing Grok 3 Demo w/Elon Musk & Team (full replay) - YouTube

Elon's Grok-3 Just Beat EVERYONE?! - YouTube

𝗘𝗹𝗼𝗻 𝗠𝘂𝘀𝗸 𝗷𝘂𝘀𝘁 𝗹𝗮𝘂𝗻𝗰𝗵𝗲𝗱 𝗚𝗿𝗼𝗸 3 — 𝗖𝗮𝗹𝗹𝗶𝗻𝗴 𝗜𝘁 "𝗧𝗵𝗲 𝗦𝗺𝗮𝗿𝘁𝗲𝘀𝘁 𝗔𝗜 𝗼𝗻 𝗘𝗮𝗿𝘁𝗵." | LinkedIn

Grok3 Launch / X

Musk's xAI Debuts Grok-3 AI Model - YouTube

Elon Musk’s xAI releases its latest flagship model, Grok 3 | TechCrunch

SQLite in Rust: Turso Limbo + libSQL

 Turso is rewriting SQLite in Rust with Glauber Costa, CEO at Turso (Changelog Interviews #626)

Introducing Limbo: A Complete Rewrite of SQLite in Rust

Limbo on GitHub
Limbo is a work-in-progress, in-process OLTP database management system,
compatible with SQLite.


Antithesis


SQLite Copyright


Upcoming Changes to the Turso Platform and Roadmap


My Story by pthorpe92.dev

GitHub - tursodatabase/libsql: libSQL is a fork of SQLite that is both Open Source, and Open Contributions.

libSQL is an open source, open contribution fork of SQLite

• libSQL server for remote SQLite access, similar to PostgreSQL or MySQL

Google Gemini AI

Google AI is evolving; The new focus / marketing is on "Gemini" AI models
It is available as

• integrated with Google search; free, no control, search only
  
  
• Google One subscription: $20/month, 1 month free trial); that includes
• chatbot
• gmail + google docs integration
• NO API
  
  
• Google cloud subscription: 90 days, $300 free trial
• this include AI API access

While Google Gemini AI models have some advance capabilities (context size 2 M tokens)
for full access API is required, so Google Cloud is required.

//gemini.google.com/

Long context  |  Generative AI  |  Google Cloud

Free cloud features and trial offer  |  Google Cloud Free Program

Gemini (chatbot) - Wikipedia

JSON Query language

JSON Query - a small, flexible, and expandable JSON query language

A small, flexible, and expandable JSON query language.

Documentation: https://github.com/jsonquerylang/jsonquery

EV for $5500 (by GM, in China)

GM massively improves the $5500 Wuling Hongguang EV hatchback - YouTube

Wuling Hongguang Mini EV - Wikipedia

sold 1.1m, best selling EV in China

AI: RAG tutorials

RAG Is (almost) All You Need! | LinkedIn

Any RAG solution requires, first to convert your data into embeddings, and then to store them on a specific data base for that purpose, aka vectorDB.  

✔️ Vanilla RAG: https://lnkd.in/dPf3x92e
✔️ Advanced RAG: https://lnkd.in/dPVa7enW
✔️ Multi modal RAG: https://lnkd.in/dkBkJqEt
✔️ Agentic RAG: https://lnkd.in/dtM9FMHA

my new AWS AI certification

last year I did a similar certification on Azure AI... now AWS AI

AWS Certified AI Practitioner Early Adopter - Credly

AWS Certified AI Practitioner Certification | AWS Certification | AWS

useful training classes:

[NEW] Ultimate AWS Certified AI Practitioner AIF-C01 | Udemy

[Practice Exams] AWS Certified AI Practitioner - AIF-C01 | Udemy

by Stephane Maarek

Mastering AWS Certified AI Practitioner AIF-C01 - Hands On! | Udemy

by Frank Kane

the same classes are available as a part of subscription, and for library members at

Udemy Business (Gale)

bun.js: 5x faster access to aws s3

 Bun 1.2 - YouTube

https://bun.sh/blog/bun-v1.2

food: Regenerative Organic Certified

Regenerative Organic Certified: Farm like the world depends on it

With regenerative organic agriculture, we can rehabilitate soil, respect animal welfare, and improve the lives of farmers. We can sequester carbon, build healthier communities, and reap more nutritious and abundant yields.

Regenerative agriculture - Wikipedia

data: countries

List of countries by GDP (nominal) - Wikipedia

To link from a Wikipedia page to Wikidata, you can use the "Wikidata item" link in the "Tools" side panel.

list of countries by GDP (nominal) - Wikidata (no data for some reason?)

List of countries by population (United Nations) - Wikipedia

list of countries by population - Wikidata

List of countries by GDP (PPP) per capita - Wikipedia

list of countries by GDP (PPP) per capita - Wikidata (empty?)

Powerful AI speech by JD Vance

VP JD Vance on the future of artificial intelligence - YouTube

US Vice President JD Vance is delivers a keynote speech for the final day of the Paris AI Summit, marking his first foreign trip since taking office as vice president. The Vice President warned global leaders and tech CEOs that “excessive regulation” would kill the rapidly growing AI industry.

course: GenAI & LLMs, by DeepLearning.AI & AWS @ Coursera

Generative AI with Large Language Models | Coursera

learn the fundamentals of how generative AI works, and how to deploy it in real-world applications.

• Generative AI use cases, project lifecycle, and model pre-training
• Fine-tuning and evaluating large language models
• Reinforcement learning and LLM-powered applications

DeepLearning.AI: Start or Advance Your Career in AI

AI: Artificial Synonyms

"AI" needs a better name :)

makes a bad marketing!

What If You Start Eating Honey Every Day For 30 Days? - YouTube

AI sometimes makes big mistakes in summaries

64 Synonyms & Antonyms for ARTIFICIAL | Thesaurus.com

Strongest matches:  man-made, synthetic, unnatural
Strong matches: counterfeit, ersatz, factitious, manufactured
Weak matches: faked, false, imitation, mock, not genuine, plastic, simulated,substitute

Zig Programming Language

Zig is a new simple system-level language, like modern C.
Rust, anther modern system-level language is more complex and powerful, like modern C++.

While Go is also a modern descendant of C (made by some same people who made C!)
there are some essential difference: Go is using "garbage collection" memory management and had "runtime". So Zig is simpler, smaller and more predictable. Maybe more fun.

Zig for Impatient Devs - YouTube 10 min

Why Ghostty is written in Zig (not Rust or Go) - YouTube 

from person who created Terraform, in Go, sold for $ billions 

Home ⚡ Zig Programming Language

Getting Started ⚡ Zig Programming Language

Zig Playground online with example code, like a tutorial

GitHub - ziglang/zig: General-purpose programming language and toolchain for maintaining robust, optimal, and reusable software.

Bun — A fast all-in-one JavaScript runtime (written in Zig, too)

Zig (programming language) - Wikipedia

Zig (Also known as Ziglang)[12] is an imperative, general-purpose, statically typed, compiled system programming language designed by Andrew Kelley.[13] It is free and open-source software, released under an MIT License.[14]

A major goal of the language is to improve on the C language,[15][16] (also taking inspiration from Rust[17][18]), with the intent of being even smaller and simpler to program in, while offering more functionality.[19] The improvements in language simplicity relate to flow control, function calls, library imports, variable declaration and Unicode support. Further, the language makes no use of macros or preprocessor instructions.

3 reasons to choose ZIGLANG over RUST for Programming #zig - YouTube

The Zig and Go Programming Showdown! - by Erik Engheim

SW => AI Agents?

 Microsoft CEO’s STUNNING Reply: "Agents Will REALLY Replace All Software” - YouTube

Microsoft CEO: AI Agents Will Transform SaaS as We Know It - CX Today

"Speaking in an interview with Bill Gurley and Brad Gerstner on their B2G podcast, Nadella suggested that the “notion that business applications exist” could “collapse” in the agentic AI era.

The comments came in response to a question concerning Microsoft’s copilot-first approach and whether it will render some existing infrastructure obsolete.

The Microsoft CEO expounded on his point, suggesting that at a fundamental level, SaaS/business applications are essentially CRUD (create, read, update, delete) databases with business logic.

It is the business logic or ‘rules’ aspect that Nadella believes will be taken over by AI agents."

EV: 2026 Tesla Model Y, design & engineering

 Exclusive First Drive of the 2026 Tesla Model Y | Jay Leno's Garage - YouTube

Join Jay Leno in this exclusive episode of Jay Leno's Garage as we take a first drive and an in-depth tour of the revolutionary 2026 Tesla Model Y! Packed with cutting-edge features, including matrix headlights, improved aerodynamics, and a luxurious, all-new interior, this is Tesla's most advanced SUV yet. Learn directly from Tesla's lead designers and engineers about the innovations that make this Model Y a game-changer.

2026 Tesla Model Y Prices, Reviews, and Photos - MotorTrend

AI: book about LLMs

[2501.09223] Foundations of Large Language Models

This is a book about large language models. As indicated by the title, it primarily focuses on foundational concepts rather than comprehensive coverage of all cutting-edge technologies. The book is structured into four main chapters, each exploring a key area: pre-training, generative models, prompting techniques, and alignment methods. It is intended for college students, professionals, and practitioners in natural language processing and related fields, and can serve as a reference for anyone interested in large language models.

Foundations of Large Language Models (PDF)

Tong Xiao and Jingbo Zhu

NLP Lab, Northeastern University & NiuTrans Research

AI: Deep Dive by Andrej Karpathy

Deep Dive into LLMs like ChatGPT - YouTube

This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their "psychology", and how to get the best use them in practical applications. 

Instructor Andrej was a founding member at OpenAI (2015) and then Sr. Director of AI at Tesla (2017-2022), and is now a founder at Eureka Labs, which is building an AI-native school. 

His goal in this video is to raise knowledge and understanding of the state of the art in AI, and empower people to effectively use the latest and greatest in their work. 

Andrej Karpathy - YouTube

Andrej Karpathy.ai

Andrej Karpathy (@karpathy) / X

previous videos

Let's build GPT: from scratch, in code, spelled out. - YouTube

GitHub - karpathy/ng-video-lecture

Code created in the Neural Networks: Zero To Hero video lecture series, specifically on the first lecture on nanoGPT. Publishing here as a Github repo so people can easily hack it, walk through the git log history of it, etc.

GitHub - karpathy/llm.c: LLM training in simple, raw C/CUDA

Let's reproduce GPT-2 (124M) - YouTube

podcast: Science, AI: Into The Impossible, by Brian Keating, Ph.D

Into The Impossible Podcast Episodes

Dr Brian Keating - YouTube

Dr. Brian Keating: Charting the Architecture of the Universe & Human Life - Huberman Lab

Brian Keating, Ph.D., is a cosmologist, a professor of physics at the University of California, San Diego, an author and a public science educator.

Brian Keating - Wikipedia

Brian Gregory Keating is an American cosmologist. He works on observations of the cosmic microwave background, leading the POLARBEAR2 and Simons Array experiments. He also conceived the first BICEP experiment. He received his PhD in 2000, and is a distinguished professor of physics at University of California, San Diego, since 2019. He is the author of two books, Losing The Nobel Prize and Into the Impossible.

How to train ChatGPT to serve you | AI Legend Terry Sejnowski - YouTube

Why Become a Scientist? Brian Keating on Lex Fridman - YouTube

Yann LeCun: AI Doomsday Fears Are Overblown [Ep. 473] - YouTube

Max Tegmark: Will AI Surpass Human Intelligence? [Ep. 469] - YouTube

Most Influential Papers in Computer Science

The 7 Most Influential Papers in Computer Science History – Terrible Software

“On Computable Numbers, with an Application to the Entscheidungsproblem” (1936)
Author: Alan Turing

=> defined computing

“A Mathematical Theory of Communication” (1948)
Author: Claude Shannon

=> defined information

“A Relational Model of Data for Large Shared Data Banks” (1970)
Author: Edgar F. Codd

=> defined databases

“The Complexity of Theorem-Proving Procedures” (1971)
Author: Stephen A. Cook

“A Protocol for Packet Network Intercommunication” (1974)
Authors: Vinton G. Cerf and Robert E. Kahn
=> internet

“Information Management: A Proposal” (1989)
Author: Tim Berners-Lee

=> web

“The Anatomy of a Large-Scale Hypertextual Web Search Engine” (1998)
Authors: Sergey Brin and Larry Page

=> search

“Recursive Functions of Symbolic Expressions and Their Computation by Machine” (1960) –

John McCarthy

“Go To Statement Considered Harmful” (1968)

Edsger Dijkstra

"Time, Clocks, and the Ordering of Events in a Distributed System” (1978)

Leslie Lamport

“No Silver Bullet—Essence and Accident in Software Engineering” (1986)

Fred Brooks

“Attention Is All You Need” (2017)

Vaswani et al.
=> LLMs AI

AI: $50 reasoning model

 New AI Reasoning Model Rivaling OpenAI Trained on Less Than $50 in Compute

S1 model: researchers at Stanford and the University of Washington trained the “reasoning” model using less than $50 in cloud compute credits.

S1 is a direct competitor to OpenAI’s o1, which is called a reasoning model because it produces answers to prompts by “thinking” through related questions that might help it check its work.

According to TechCrunch, S1 is based on an off-the-shelf language model, which was taught to reason by studying questions and answers from a Google model, Gemini 2.0 Flashing Thinking Experimental

a relatively small amount of training data—1,000 curated questions, along with the answers—and teach it to mimic Gemini’s thinking process.

story: Eisenhower roads to future

Transcontinental Motor Convoy - Wikipedia

1919 Transcontinental Motor Convoy | Eisenhower Presidential Library

In the summer of 1919, a young Lieutenant Colonel named Dwight D. Eisenhower participated in the first Army transcontinental motor convoy. The expedition consisted of 81 motorized Army vehicles that crossed the United States from Washington, DC, to San Francisco, a venture covering a distance of 3,251 miles in 62 days. The expedition was manned by 24 officers and 258 enlisted men. The convoy was to test the mobility of the military during wartime conditions. As an observer for the War Department, Lt. Col. Eisenhower learned first-hand of the difficulties faced in traveling great distances on roads that were impassable and resulted in frequent breakdowns of the military vehicles. These early experiences influenced his later decisions concerning the building of the interstate highway system during his presidential administration.

His administration undertook the development and construction of the Interstate Highway System, which remains the largest construction of roadways in American history. 

In 1957, following the Soviet launch of Sputnik, Eisenhower led the American response which included the creation of NASA and the establishment of a stronger, science-based education via the National Defense Education Act. 

This Is Why the U.S. Can’t Use the Oil It Produces - YouTube

EV: Samsung Solid State batteries

How much range electric cars with Samsung's Solid State batteries will have - YouTube

Samsung to Mass-Produce Solid-State Batteries for 'Super Premium' EVs by 2027 | PCMag

Samsung's latest solid-state battery technology will power up premium EVs first, giving them up to 621 miles of range

The new batteries—which promise to improve vehicle range, decrease charging times, and eliminate risk of battery fires—could go into mass production as soon as 2027. Multiple automakers have been reportedly testing samples. Samsung did not list any by name but it's worked with Hyundai, Stellantis, and General Motors, among others.

AI: Gemini 2.0 best performance & price?!

Apparently, the latest Google AI models are not only very competitive in performance,
but also have by far lowest price, even better than DeepSeek.

That makes sense, since Google has custom designed and optimized HW (TPU processors),
software, people, and the most data of all companies. 

The real AI competition!

Gemini 2.0 blew me away - YouTube by Theo-t3.gg

Gemini 2.0 model updates: 2.0 Flash, Flash-Lite, Pro Experimental

The Gemini app is now powered by Gemini 2.0 Flash.

Gemini Flash - Google DeepMind

Google’s Gemini app adds access to ‘thinking’ AI models | The Verge

AI art with three.js

GitHub - eduardfossas/codrops-generative-artwork-three: Codebase for the tutorial created on Codrops: "Creating a Generative Artwork with Three.js" @GitHub

Manoela Ilic | LinkedIn

 Creating a Generative Artwork with Three.js

"Develop a dynamic, generative artwork using Three.js and grid systems,
inspired by Lygia Clark’s minimalist geometric designs."

AI: Reasoning LLMs: DeepSeek R1, OpenAI o1, o2

very detailed technical description of some of AI training techniques

 Understanding Reasoning LLMs - by Sebastian Raschka, PhD

"reasoning": the process of answering questions that require complex, multi-step generation with intermediate steps.

AI Czar David Sacks Explains the DeepSeek Freak Out - YouTube (All-in Podcast)

DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459 - YouTube

DeepSeek-R1 from A-to-Z | newline

Chess engines: JS, GoLang

Chess Engines: A Zero to One Guide

This a guide on how to build a Chess engine from scratch, updated periodically as I further explore the space. This assumes some basic knowledge of Chess, search algorithms, and coding.

code: 

GitHub - 0hq/starter_chess_engine

chessboardjs.com » Homepage

GitHub - jhlywa/chess.js: A TypeScript chess library for chess move generation/validation, piece placement/movement, and check/checkmate/draw detection

GitHub - 0hq/antikythera:
An instructional parallelized chess engine written in Golang.
Starts simple, ends complex.

Antikythera

A parallelized Chess engine written in Golang.

This engine is mainly written with educational use in mind: ...start with an engine that is as simple as possible and add more complex features one-by-one. You can read through each one of the engine versions in the /engines/ folder, each with different levels of sophistication.

Chess opening book (computers) - Wikipedia

Computer chess - Wikipedia

EV: Toyota bZ3X: $13,800 ?

Toyota's new bZ3X EV gets 520 km range - price starts at $13,800 USD - YouTube

2025 Toyota bZ3X: What We Know So Far @CarAndDriver

Toyota bZ3X - Wikipedia

The Toyota bZ3X is a battery electric compact crossover SUV produced by Toyota through its joint venture GAC Toyota in China.

extract PDF data

JS libs/modules

pdf.js-extract - npm

"extracts text from PDF files
This is just a library packaged out of the examples for usage of pdf.js with nodejs.
It reads a pdf file and exports all pages & texts with coordinates. This can be e.g. used to extract structured table data.
This package includes a build of pdf.js. why? pdfs-dist installs not needed dependencies into production deployment.
Note: NO OCR!"

alternative libs:

Parsing PDFs in Node.js - LogRocket Blog

pdf-parse

pdf2json

pdfreader

AWS OCR API

Using AWS services with OCRs (Textract and Tesseract.js) in Node.js | by Noam | Medium

OCR Software, Data Extraction Tool - Amazon Textract - AWS

Amazon Textract Resources - Amazon Web Services

What is Amazon Textract? - Amazon Textract

story: TSMC (maker of most AI chips)

excellent interview and original story (podcasts)  @acquired.fm

TSMC Founder Morris Chang: The Complete History and Strategy

TSMC (Remastered): The Complete History and Strategy

TSMC - Wikipedia

tsmc stock - Google Search

#10 largest company in the world, almost $1T

All U.S. Listed Stocks, Ranked by Market Cap - Stock Analysis

No.	Symbol	Company Name	Market Cap	Stock Price	% Change	Revenue
1	AAPL	Apple Inc.	3,406.32B	226.52	-4.02%	395.76B
2	MSFT	Microsoft Corporation	3,058.04B	411.36	-0.89%	261.80B
3	NVDA	NVIDIA Corporation	2,863.74B	116.94	-2.61%	113.27B
4	AMZN	Amazon.com, Inc.	2,502.14B	237.96	0.12%	620.13B
5	GOOGL	Alphabet Inc.	2,493.49B	203.70	-0.16%	339.86B
6	META	Meta Platforms, Inc.	1,772.95B	699.76	1.54%	164.50B
7	TSLA	Tesla, Inc.	1,244.65B	386.96	-4.36%	97.69B
8	AVGO	Broadcom Inc.	1,035.18B	220.85	-0.19%	51.57B
9	BRK.B	Berkshire Hathaway Inc.	1,010.43B	465.19	-0.74%	369.89B
10	TSM	Taiwan Semiconductor Manufacturing Company Limited	843.76B	202.50	-3.26%	88.29B