The First *REAL GOLD* Tesla Cybertruck - What did it Cost?! - YouTube
+$100K
Wednesday, July 31, 2024
Tuesday, July 30, 2024
node.js += TypeScript?
competition is good! with Deno using TypeScript natively, and having Sqlite embedded,
now even node.js plans to add that functionality.
Node Weekly Issue 542: July 30, 2024
"Node merged an experimental ability to transpile TypeScript into JavaScript, ultimately letting Node directly ‘run TypeScript’. However, no type checking is performed and, as Matt Pocock explains, experimental and TypeScript-only features are a no-no. Perhaps think of it as less ‘running TypeScript’ and more ‘able to tolerate most of TypeScript’?"Monday, July 29, 2024
Microsoft Dev Box: Dev Workstation @ Azure Cloud
Microsoft Dev Box – Dev Workstation in the Cloud | Microsoft Azure
"Accelerate project onboarding and maximize developer productivity with dev-optimized workstations in the cloud."To use Microsoft Dev Box, each user must be licensed for Windows 11 Enterprise or Windows 10 Enterprise, Microsoft Endpoint Manager, and Azure Active Directory P1.
from $1.5/h, Max Monthly Price: $138.20
8 vCPU, 32 GB RAM, 256 GB Storage
8 vCPU, 32 GB RAM, 256 GB Storage
Architecture: A-frame houses
An A-frame building is an architectural style[1] of building that features steeply-angled sides (roofline) that usually begin at or near the foundation line, and meet at the top in the shape of the letter A. An A-frame ceiling can be open to the top rafters.
Plan 80519 | Contemporary House Plan with Walk-Out Basement
Roberts A-Frame Log Cabin Near Red Lodge Mountain!, MT - Booking.com
Sunday, July 28, 2024
SearchGPT: AI Search from OpenAI
SearchGPT is a prototype of new AI search features | OpenAI
SearchGPT, a temporary prototype of new AI search features that give you fast and timely answers with clear and relevant sources.can join "waitlist", not public (yet)
OpenAI generates about $2 billion in annual revenue, while investors have valued the San Francisco startup at over $80 billion.
...it’s unlikely that the company will go bankrupt just yet. With backers like Microsoft and Sequoia, the company will likely secure funding during the next seeding round.
Open Source tool: SQL studio
frectonz/sql-studio: SQL Database Explorer [SQLite, libSQL, PostgreSQL, MySQL/MariaDB, DuckDB] @GitHub
Rust + TypeScript, MIT
Single binary, single command SQL database explorer. SQL studio supportsSQLite, libSQL, PostgreSQL, MySQL and DuckDB.
powershell -c "irm https://github.com/frectonz/sql-studio/releases/download/0.1.17/sql-studio-installer.ps1 | iex"sql-studio postgres [url]
Saturday, July 27, 2024
Safe nuclear power? vs. solar
Mag 7 sell-off, Wiz rejects Google, UBI, Kamala in, China's nuclear buildout, Sacks responds to PG - YouTube All-in Podcast
jump to nuclear story: China's nuclear buildout, Sacks responds to PG - YouTube
+ alternative view: China is mostly moving to solar & wind
China installs wind and solar equivalent of 5 nuclear power plants per week - YouTube
China installed 1GW of nuclear last year, compared to 300GW of solar and wind
today US has roughly one terawatt of total electricity production capacity
China today has about 3 TW of total electricity production capacity
and about you know 2 to 3% of that is nuclear today for China
and about you know 2 to 3% of that is nuclear today for China
by 2050 the US has projected to build out an additional terawatt to getting us to two terawatts of capacity so we're going to double our total electricity output by 2050
China meanwhile has a plan stated to increase electricity production to
8.7 TW so basically tripling between now and 2050 88% % of their power by 2050
cost of electricity China's about 7 to 9 cents a kilowatt hour
the US is 17 to 25 cents a kilowatt hour
and China is projected to drop their price to less than 6 cents due to the expansion of Renewables and nuclear power
the US cost is about triple what it is in China to build out new electricity capacity
it's cheaper to make electricity and it's cheaper to build new capacity and they're building at an accelerated rate and this really highlights the industrial challenge the United States is going to have in the decades ahead
Gen 4 is the next generation of nuclear power reactors and their next Generation
because they're meant to be much more safe where they cannot theoretically have a meltdown you can't have a nuclear meltdown like you had with Fukushima or with Three Mile Island and to be clear Fukushima was generation two those are the boiling water reactors
(in China test reactor) ... last summer and even when they turned everything off the system did not fail it did not have a meltdown it it maintained its ability to control its temperature and not have a meltdown
GoLang Channels
Three Ways To Think About (Go) Channels | DoltHub Blog
"One of Golang's steepest learning curves is how to use channels.At the practical level, channels are locked, buffered queues. Their API matches that of a queue, the implementation is a queue, and I find it easiest to write and read channel code with this abstraction.
But it is not enough to know how the channel API works. You also need to be comfortable with the channel ecosystem. Channels usually depend on errorgroups, goroutines, and other concurrency primitives..."
Friday, July 26, 2024
pgAdmin 4 Docker, for Postgres DB
dpage/pgadmin4 - Docker Image | Docker Hub
This is the official Docker distribution of pgAdmin 4.
pgAdmin 4 is written as a web application with Python(Flask) on the server side and ReactJS, HTML5 with CSS for the client side processing and UI.
Although developed using web technologies, pgAdmin 4 can be deployed either on a web server using a browser, or standalone on a workstation. The runtime/ subdirectory contains an NWjs based runtime application intended to allow this, which will execute the Python server and display the UI.
Thursday, July 25, 2024
Architecture: Octagon Dome House
Armour–Stiner House - Wikipedia
The Armour–Stiner House is an octagon-shaped and domed Victorian-style house located at 45 West Clinton Avenue in Irvington, in Westchester County, New York. It was designated a National Historic Landmark in 1976.[1][3] It is the only known fully domed octagonal residence. The house was modeled after Donato Bramante’s 1502 Tempietto in Rome, which in turn was based on a Tholos, a type of ancient classical temple..jpg/500px-Armour-Stiner_House_(1).jpg)
Wednesday, July 24, 2024
node:sqlite: node.js embedded sqlite
lib,src,test,doc: add node:sqlite module by cjihrig · Pull Request #53752 · nodejs/node
import { DatabaseSync } from 'node:sqlite'; const database = new DatabaseSync(':memory:'); // Execute SQL statements from strings. database.exec(` CREATE TABLE data( key INTEGER PRIMARY KEY, value TEXT ) STRICT `);
Node.js v22.5.0 (Current) Released – But Don't Install It! — A notable release for three reasons: first, the WebSocket feature in node:http is now exposed. Second, remember how Node is embedding SQLite? node:sqlite is now included and ready to play with. Third, however, v22.5.0 introduced a nasty bug, so you'll want to try v22.5.1 instead.
Unix explained by its creators
Most of computers now run or use some variant of Unix.
Here is clear explanation why it was crated 50+ years ago
Ken Thompson and Dennis Ritchie Explain UNIX (Bell Labs) - YouTube
Pushing the Limits of Technology: The Ken Thompson and Dennis Ritchie Story - YouTube
Tuesday, July 23, 2024
AI: Llama 3.1 better than GPT-4o and Claude 3.5 Sonnet?
Meta moves to dethrone ChatGPT | LinkedIn
Meta took a big swing at its artificial intelligence rivals on Tuesday, releasing the 3.1 version of its Llama AI model alongside some bold claims. Llama 3.1 is the “most capable” version of Meta’s AI model to date — but the company has benchmarks beyond its own performance. It claims the updated Llama outperforms OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks, including math and multilingual skills. CEO Mark Zuckerberg ratcheted the competition up even further, saying he expects Llama 3.1 to overtake ChatGPT as the most widely used AI assistant by the end of the year./2024/07/23/1721737589608.gif?w=1920)
Polypane: web browser for developers
a tool for fast testing web apps in multiple sizes (desktop, tablet, mobile, etc.)
based on Electron, with Chromium embedded, React for UI
subscription based, free trial
Polypane, The browser for ambitious web developers
Like using DevTools with cheats enabled.A stand-alone browser and devtool for ambitious web developers that want to build more responsive, accessible and performant sites in less time. Save hours on each project, love every minute of it.
podcast interview with the creator of the tool
Polypane-demonium with Kilian Valkhof, creator of Polypane (JS Party #327)
Monday, July 22, 2024
ClippyGPT: OpenAI Doc Search for Supabase
ClippyGPT - How I Built Supabase’s OpenAI Doc Search (Embeddings) - YouTube
ask our old friend Clippy anything you want about Supabase, and it will answer it using natural language. Powered by OpenAI + prompt engineering.
ask our old friend Clippy anything you want about Supabase, and it will answer it using natural language. Powered by OpenAI + prompt engineering.
- Prompt engineering and best practices
- Working with a custom knowledge base via context injection + OpenAI embeddings
- How to store embeddings in Postgres using pgvector
Supabase blog post: https://supabase.com/blog/chatgpt-sup...
pgvector extension: https://github.com/pgvector/pgvector
source code
Generate embeddings implementation: https://github.com/supabase/supabase/...
Clippy edge function implementation: https://github.com/supabase/supabase/...
Clippy frontend implementation: https://github.com/supabase/supabase/...
Prompt engineering: https://prmpts.ai/blog/what-is-prompt...
X.AI: the most powerful AI cluster: 100,000 Nvidia H100 GPUs
Supermicro provided much of the hardware, and the company's CEO, Charles Liang, also commented on Musk's thread, touting the team's execution. This follows Liang's recent glowing words for Musk's liquid-cooled AI data centers.
Sunday, July 21, 2024
AI: thispersondoesnotexist.com StyleGAN2 and StyleGAN3
"paint with AI"
AI generated "fake" images of people!
too realistic :( how to "trust" any image now?
thispersondoesnotexist.com (1024×1024)
from NVidia labs NVIDIA Research Projects
Synthesizing High-Resolution Images with StyleGAN2 | NVIDIA Technical Blog
Synthesizing High-Resolution Images with StyleGAN2 - YouTube
Generating Faces with StyleGAN3 (7.2) - YouTube
NVlabs/stylegan3: Official PyTorch implementation of StyleGAN3 @GitHub
Generative adversarial network - Wikipedia
mentioned in
alternatives
Create Art or Modify Images with AI with Stable Diffusion model
Stable Diffusion AI Art from Stability AI
Stable Diffusion 2-1 - a Hugging Face Space by stabilityai
Samsung SSDs: QVO vs. EVO vs. PRO
Samsung QVO vs. EVO: SSD differences explained | Digital Trends
The Pro SSDs remain the most expensive that Samsung has on offer. In comparison, both the EVO and the QVO are more reasonably priced. Of the three of them, the new QVO drives are by far the cheapest.The Samsung 870 EVO is an excellent drive and is usually priced betweeen $120 and $170 on Amazon for 1TB of storage. The cheaper equivalent of that drive, the Samsung 870 QVO, usually comes with a price tag of around $100 to $130.
Longevity
Samsung 870 EVO and the Samsung 870 QVO:
- 1TB disk size: 600 TBW for the EVO, 360 TBW for the QVO
- 2TB disk size: 1,200 TBW for the EVO, 720 TBW for the QVO
- 4TB disk size: 2,400 TBW for the EVO, 1,440 TBW for the QVO
price history / tracking
Samsung Electronics 870 EVO 2TB 2.5 Inch SATA III Internal SSD (MZ-77E2T0B/AM) camelcamelcamel.com
SAMSUNG 990 PRO SSD 2TB PCIe 4.0 M.2 2280 Internal Solid State Hard Drive, | camelcamelcamel.com
SAMSUNG 990 PRO SSD 2TB PCIe 4.0 M.2 2280 Internal Solid State Hard Drive, camelcamelcamel.com
related
Samsung's 2nd-gen SmartSSD processes data right on the drive | Digital Trends
Normally, the CPU would need to scan the hard drive for the information it needed and then pull that info out of the drive before processing it or sending it to the GPU. But the SmartSSD allows the hard drive to handle most of that process.
Samsung once again partnered with AMD to create the second-generation SmartSSD. The drive uses an AMD-based SOC heavily optimized for Samsung SSD architecture.
Saturday, July 20, 2024
Architecture: Home Around a Boulder
interesting design! house, like a pearl shell
...Catskill Mountains in upstate New York ... the Rock’n’House, a remarkable family home built around a 12,000 year-old-boulder. Some might have found themselves between a rock and a hard place after finding a prehistoric glacial erratic in the middle of their land, but for architect Christian Wassmann it was just what he was looking for. The boulder became the central feature of Wassmann’s design and is always present in his family’s impressive home–the perfect example of how humans and nature can coexist as one.
Step Inside a New York House Built Around a Boulder | Architectural Digest
clever detail: "Polaris", a handrail that double functions as naked-eye observatory for north star!
AI: ONNX Runtime (Open Neural Network Exchange)
ONNX Runtime is a cross-platform machine-learning model accelerator, with a flexible interface to integrate hardware-specific libraries. ONNX Runtime can be used with models from PyTorch, Tensorflow/Keras, TFLite, scikit-learn, and other frameworks
The Open Neural Network Exchange (ONNX) [ˈɒnɪks][2] is an open-source artificial intelligence ecosystem[3] of technology companies and research organizations that establish open standards for representing machine learning algorithms and software tools to promote innovation and collaboration in the AI sector. ONNX is available on GitHub.
Each computation dataflow graph is a list of nodes that form an acyclic graph. Nodes have inputs and outputs. Each node is a call to an operator. Metadata documents the graph. Built-in operators are to be available on each ONNX-supporting framework.[6]
ONNX Runtime | Getting-started
Production-grade AI engine to speed up training and inferencing in your existing technology stack.Friday, July 19, 2024
in-security: CrowdStrike
Microsoft outage hits Windows users; CrowdStrike deploys fix: Live updates - The Washington Post
"On July 19, 2024, a faulty CrowdStrike update caused Blue Screens of Death on Microsoft Windows machines, affecting millions of Windows computers worldwide.[52] Affected machines were forced into a bootloop, making them unusable. The downtime caused widespread global impact, including grounding of commercial airline flights, temporarily taking Sky News offline, and impacting 911 emergency call centers."xkcd: CrowdStrike
time to switch to Linux for critical infrastructure?
Virtual Machines—Linux | Microsoft Azure
More than 60 percent of customer cores in Azure run Linux workloads.
AI: Transformers.js, with WASM
Transformers.js uses ONNX Runtime to run models in the browser (and node.js)
The best part about it, is that you can easily convert your pretrained PyTorch, TensorFlow, or JAX models to ONNX using 🤗 Optimum.
Official GitHub Repository — https://github.com/xenova/transformers.js/
HuggingFace Guide — https://huggingface.co/docs/transformers.js/index
The best part about it, is that you can easily convert your pretrained PyTorch, TensorFlow, or JAX models to ONNX using 🤗 Optimum.
For more information, check out the full documentation.
Transformers.js is designed to be functionally equivalent to Hugging Face’s transformers python library, meaning you can run the same pretrained models using a very similar API. These models support common tasks in different modalities, such as:
Although Transformers.js was originally designed to be used in the browser, it’s also able to run inference on the server. In this tutorial, we will design a simple Node.js API that uses Transformers.js for sentiment analysis.
Transformers.js @Hugging Face
State-of-the-art Machine Learning for the web. Run 🤗 Transformers directly in your browser, with no need for a server!Transformers.js is designed to be functionally equivalent to Hugging Face’s transformers python library, meaning you can run the same pretrained models using a very similar API. These models support common tasks in different modalities, such as:
📝 Natural Language Processing: text classification, named entity recognition, question answering, language modeling, summarization, translation, multiple choice, and text generation.
🖼️ Computer Vision: image classification, object detection, and segmentation.
🗣️ Audio: automatic speech recognition and audio classification.
🐙 Multimodal: zero-shot image classification.
npm init -y npm i @xenova/transformers
HuggingFace Guide — https://huggingface.co/docs/transformers.js/index
$0 Embeddings (OpenAI vs. free & open source) - YouTube (transformes.js)
Thursday, July 18, 2024
AI for Architecture: ChatGPT-4o Landscape Design
Architecture and Landscape Design with ChatGPT-4o - YouTube
able to analyze uploaded aerial photos, count structures,
generate python code to vector trace to SVG... impressive
LandArch | Landscape Architecture Design | Architecture Community
vector databases tutorial
What is a Vector Database & How Does it Work? Use Cases + Examples | Pinecone
A vector database indexes and stores vector embeddings for fast retrieval and similarity search, with capabilities like CRUD operations, metadata filtering, horizontal scaling, and serverless
Wednesday, July 17, 2024
AI embeddings without OpenAI, $0; tokenization @ Hugging Face API & libs
Hugging Face, Inc. is a French-American company that develops computation tools for building applications using machine learning. It is most notable for its transformers library built for natural language processing applications and its platform that allows users to share machine learning models and datasets and showcase their work.
$0 Embeddings (OpenAI vs. free & open source) - YouTube
tutorial by Rabbit Hole Syndrome
Source code: //github.com/rabbit-hole-syndrome/open-source-embeddings
tutorial by Rabbit Hole Syndrome
Source code: //github.com/rabbit-hole-syndrome/open-source-embeddings
Sentence Embeddings (SBERT): //sbert.net/ MTEB Leaderboard:
MTEB: Massive Text Embedding Benchmark
MTEB Leaderboard - a Hugging Face Space by mteb
mteb/docs/adding_a_model.md at main · embeddings-benchmark/mteb @GitHub
Byte-Pair Encoding tokenization - Hugging Face NLP Course
WordPiece tokenization - Hugging Face NLP Course
Hugging Face JS libraries (calling API)
Brilliant | Learn interactively (embeddings)
Generate LLM Embeddings On Your Local Machine - YouTube
Let's build GPT: from scratch, in code, spelled out. - YouTube
data format: OData Web Services
used with Dynamics 365; based on XML
OData documentation - OData | Microsoft Learn
OData (Open Data Protocol) is an ISO/IEC approved, OASIS standard that defines a set of best practices for building and consuming REST APIs. It enables creation of REST-based services which allow resources identified using Uniform Resource Locators (URLs) and defined in a data model, to be published and edited by Web clients using simple HTTP messages.OData Web Services - Business Central | Microsoft Learn
The Open Data Protocol (OData) is a web protocol that is designed for querying tabular data and provides you with an alternative to SOAP-based web services. OData builds on web technologies such as HTTP and JavaScript Object Notation (JSON) to provide access to information from different applications, services, and stores. OData uses URIs for resource identification and commits to an HTTP-based, uniform interface for interacting with resources. This commitment to core Web principles allows for OData to enable a new level of data integration and interoperability across a broad range of clients, servers, services, and tools.The protocol enables the creation and consumption of HTTP-based Web APIs, which allow Web clients to publish and edit resources, identified using URLs and defined in a data model, using simple HTTP messages. OData shares some similarities with JDBC and with ODBC; like ODBC, OData is not limited to relational databases.
MIT, JS/TS
AI TDD code generator: micro-agent
BuilderIO/micro-agent: An AI agent that writes (actually useful) code for you @GitHub
TypeScript, MIT license
Just run micro-agent, give it a prompt, and it'll generate a test
and then iterate on code until all test cases pass.
Tuesday, July 16, 2024
DIY AI in Go: minigrad: Andrej Karpathy's Micrograd in Go
0verread/minigrad: Andrej Karpathy's Micrograd in Go @GitHub
A neural network library in golang on an autograd engine.Faster and easy to use like PyTorch.
This is Andrej Karpathy's Micrograd implementation in Go.
book: Structure and Interpretation of Computer Programs: JavaScript Edition (MIT, from Scheme)
a "classic reference" book, translated from Scheme to JavaScript!
functional programming!
- Publisher : The MIT Press (April 12, 2022)
- Paperback : 640 pages
video presentation:
Gerald Jay Sussman (MIT)
sarabander (Andres Raba) @GitHub
cccbook/sicpjs: Structure and Interpretation of Computer Programs — JavaScript Edition (Example Code)
Monday, July 15, 2024
Nomic Atlas: AI data clustering tool
Python bindings for working with Nomic Atlas, the world’s most powerful unstructured data interaction platform. Atlas supports datasets from hundreds to tens of millions of points, and supports data modalities ranging from text to image to audio to video.
With Nomic Atlas, you can:
with unstructured text, image and audio data.
Course: The Complete OPENAI JS APIs Course - Build 15 Projects | Udemy
With Nomic Atlas, you can:
- Generate, store and retrieve embeddings for your unstructured data.
- Find insights in your unstructured data and embeddings all from your web browser.
- Share and present your datasets and data findings to anyone.
with unstructured text, image and audio data.
AI: llm.c vs GPT-2, $672, 24h, by Andrej Karpathy
In this post we are reproducing GPT-2 in llm.c. This is "the GPT-2", the full, 1558M parameter version that was introduced in OpenAI's blog post Better Language Models and their Implications in February 14, 2019. llm.c does so directly in C/CUDA (total of ~5,000 lines of code), without the typical training stack that would involve the Python interpreter and a significantly more complex deep learning library like PyTorch/JAX, huggingface/transformers, or etc. In 2019, training GPT-2 was an involved project from an entire team and considered a big model run but, ~5 years later, due to improvements in compute (H100 GPUs), software (CUDA, cuBLAS, cuDNN, FlashAttention) and data (e.g. the FineWeb-Edu dataset), we can reproduce this model on a single 8XH100 node in 24 hours, and for $672, which is quite incredible
Feed | LinkedIn by Aleksa Gordić | LinkedIn
Andrej Karpathy (born 23 October 1986[2]) is a Slovak-Canadian computer scientist who served as the director of artificial intelligence and Autopilot Vision at Tesla. He co-founded and formerly worked at OpenAI,[3][4][5] where he specialized in deep learning and computer vision
Sunday, July 14, 2024
Firbit API, no web dashboard anymore
Fitbit shut down its web dashboard on July 8, 2024, forcing users to rely on the Fitbit app instead.
Fitbit provides a set of public Web APIs that developers may use to retrieve Fitbit user data collected by the Fitbit trackers & smartwatches, Aria & Aria 2 scales, and manually entered log data. Anyone may use the Web APIs to build integrations with the Fitbit data services, so long as their application complies with the Fitbit Platform Terms of Service, the Fitbit User Data and Developer Policy, and the Fitbit user consents to share their data with the developer’s application.
Pixabay: free images, sounds, videos
4.5 million+ Stunning Free Images to Use Anywhere - Pixabay - Pixabay
"Stunning royalty-free images & royalty-free stock"
Saturday, July 13, 2024
AI Data Center: costs & revenue?
Most of recent stock market growth is based on investments in AI,
and that is mostly in data-centers powered by NVIDIA systems.
While that is good for stock market, to be sustainable would require some tangible revenue.
Will this come on time?
Biden chaos, Soft landing secured? AI sentiment turns bearish, French elections - YouTube
All-in podcast
Subscribe to:
Posts (Atom)