📼 Will AI take embedded software jobs?

In the article by Jacob Beningo, the author delves into the evolving role of generative AI in embedded software development. While embedded software is often considered a specialized domain requiring extensive hardware knowledge and coding skills, Beningo points out that generative AI models, such as ChatGPT, are increasingly capable of handling tasks like code optimization, hardware abstraction, and low-level driver generation. This challenges the perception that embedded software is too unique to be impacted by AI.

However, Beningo also emphasizes that generative AI should be viewed as a valuable tool rather than a replacement for experienced developers. He draws a parallel to a nail gun, stating that AI can significantly accelerate development when wielded by knowledgeable engineers. AI, like a nail gun, enhances the efficiency and quality of work but does not diminish the value of human expertise. It is crucial to recognize that while AI can assist in coding, it requires human guidance and expertise to produce the desired results effectively.

The author concludes that generative AI is unlikely to replace embedded software jobs. Instead, he sees it as a tool that can help bridge the growing demand for embedded software engineers by making them more efficient and enabling them to work faster. In a society driven by technological advancement, generative AI is poised to enhance the skills of embedded software engineers rather than replace them, ultimately benefiting the field and its practitioners.

⛓ More information...

#articles #artificial_intelligence

📺 Writing an Arduino driver with OpenAI ChatGPT and PDF parsing

Witness how AI can simplify the process of turning complex datasheet information into actual code, providing valuable insights into the future of microcontroller programming. While it may not be a speedier alternative to manual coding, the video offers a fascinating glimpse into AI-driven development and how it can potentially benefit the development community. Adafruit has generously provided comprehensive documentation and ChatGPT logs for your reference, making it a valuable resource for enthusiasts and developers alike.

⛓ https://youtu.be/mQwZTo6c_yU

#video #arduino #artificial_intelligence

📼 Andrej Karpathy’s Baby Llama Runs on Samsung Galaxy Watch 4

Andrej Karpathy, from OpenAI, developed Baby Llama as a weekend project to demonstrate the possibility of running the Llama 2 model on edge devices. The project, inspired by Georgi Gerganov's llama.cpp, involves training the Llama 2 model using PyTorch and then saving the weights in a binary file. Karpathy created a compact C file ('run.c') for loading and running the model with single-precision calculations, ensuring a low-memory footprint and independence from external libraries.

The model runs efficiently even on a single M1 laptop without GPUs, with performance enhancements achieved through various compilation flags for code optimization. A Twitter user successfully demonstrated Baby Llama running on a Samsung Galaxy Watch 4.

Karpathy's repository provides the pre-trained model and code, allowing anyone to experiment with Baby Llama on their devices. It's important to note that this project is experimental and not meant for production use, aiming to showcase the viability of running advanced models on low-powered devices with C, a language typically not associated with machine learning due to its GPU incompatibility.

⛓ More information...

#news #artificial_intelligence

⛓ 3 trends for 2024: edge intelligence, RISC-V, & chiplets

This article discusses three major technology trends expected to shape 2024:

➡️ Edge Intelligence Enhancement: The article highlights advancements in deploying large language models (LLMs) on devices with limited memory, particularly focusing on a research paper by Apple. This paper introduces techniques for efficient LLMs operation on such devices, indicating a broader trend of increased machine learning and inference capabilities at the edge of networks.

➡️ RISC-V Architecture Adoption: The formation of Quintauris, a company backed by major tech players, is set to accelerate the commercialization of RISC-V hardware globally. This development suggests a growing interest in RISC-V architecture, especially in automotive, mobile, and IoT applications. The industry is moving towards heterogeneous computing architectures, where RISC-V plays a crucial role in various functions.

➡️ Rise of Chiplets: Chiplets are emerging as a solution to the high demands of machine learning-intensive products. The trend is compared to the evolution of the intellectual property business two decades ago. Startups like Zero ASIC are democratizing chip-making with platforms for chiplet-based design and emulation. The increasing use of chiplets will rely on standardization efforts like the UCIe specification.

⛓ More information...

#news #artificial_intelligence

📺 Running Advanced AI on Raspberry Pi: A Journey with GPT-4 and Beyond

In an innovative exploration of AI's reach beyond high-end hardware, this video delves into the feasibility of deploying advanced language models on accessible technology like the Raspberry Pi 5. The presenter, a seasoned enthusiast in small-tech advancements, embarks on a journey to test a variety of language models including Orca, Fi, and the groundbreaking GPT-4.

With a focus on practicality, the video demonstrates how these models, typically associated with powerful computing resources, can operate on a device as compact and affordable as the Raspberry Pi 5. This approach not only democratizes access to cutting-edge AI but also opens up a realm of possibilities for tech enthusiasts and hobbyists.

The video highlights the Raspberry Pi's capacity to handle models with up to 1.7 trillion parameters, a feat that traditionally requires substantial computing power. The presenter's experiments with local, private AI deployments raise intriguing questions about the potential of small-scale, private AI systems.

⛓ https://youtu.be/Y2ldwg8xsgE

#video #raspberry #artificial_intelligence

🧠 BitNetMCU - Implement neural networks even on the cheapest microchips!

BitNetMCU presents an exciting journey into the fusion of machine learning and embedded systems, offering valuable insights into the implementation of neural networks even on the most modest piece of hardware.

Features of the project:
➡️Utilizes quantization aware training and fine-tuning techniques to achieve over 99% accuracy on tasks like MNIST.
➡️Operates efficiently within strict memory constraints, with algorithms designed to run on any microcontroller.
➡️Employs a 🐍 PyTorch-based training pipeline for versatility and ease of use.
➡️Implements an 👩‍💻 Ansi-C inference engine for seamless portability across different microcontroller platforms.

To sum up, this project provides a gateway to deeper understanding in electronics, AI, and machine learning, empowering individuals to create innovative solutions for real-world challenges in IoT, edge computing, and beyond.

More information:
🔗 The main article
😹 The GitHub repository

#projects #libraries #artificial_intelligence #machine_learning

🍓 Raspberry foundation has released an affordable AI Kit

The Raspberry Pi Foundation has launched the Raspberry Pi AI Kit, priced at $70, in collaboration with Hailo. This kit, featuring an M.2 HAT+ board preassembled with a Hailo-8L AI accelerator module, is designed to enhance the Raspberry Pi 5's capabilities in neural networks (NN), artificial intelligence (AI), and machine learning.

Key features of the kit:
➡️13 tera-operations per second (TOPS)
➡️Single-lane PCIe 3.0 connection running at 8Gbps
➡️Full integration with Raspberry Pi's software ecosystem
➡️Support for multiple cameras and NN configurations

The kit supports a range of AI applications, such as object detection and pose estimation, by running complex AI models on the Hailo-8L co-processor, thus freeing up the Raspberry Pi's CPU for other tasks.

⛓ More information...

#news #raspberry #artificial_intelligence

🍏 Apple Intelligence: Apple’s New Generative AI Unveiled

At WWDC 2024 which took place today, Apple introduced Apple Intelligence, a generative AI integrated across its ecosystem. Apple Intelligence is designed to offer highly personalized experiences while prioritizing user privacy.

It’s personal intelligence, and it’s the next big step for Apple
Tim Cook

This announcement underscores the integration of Apple Intelligence into iOS, macOS, and VisionOS. SVP Craig Federighi highlighted that much of the processing occurs locally on Apple silicon, with additional support from Private Cloud Compute for tasks needing cloud resources.

The key moments of the presentation:
➡️Enhanced Siri: Deeply integrated with new capabilities, including text-based queries and expanded App Intents for first and third-party apps.
➡️App Integration: Helps compose messages in Mail and supports Smart Replies.
➡️Image Playground: On-device image generation for apps like Messages and Keynote.
➡️Image Wand: A tool for Apple Pencil to create images from text.

Apple Intelligence will be available on iPhone 15 Pro and M1 Mac and iPad devices with upcoming OS updates. Additionally, Apple announced a partnership with OpenAI to integrate ChatGPT into Siri, offering advanced image and text generation features.

⛓ More information...

#news #artificial_intelligence

🧠 American math genius checked out the latest OpenAI's language model

Terence Tao is widely regarded as a math genius. He scored a 760 on the SAT math section at age 8, and became the youngest gold medalist in the International Mathematical Olympiad at age 13. Tao earned his PhD from Princeton at just 20.

Recently on Mastodon he shared his experience with OpenAI's GPT-o1 model, noting its advancements in handling mathematical tasks. In his experiments, the model successfully identified and applied Cramer's theorem, showing clear improvement over past versions. However, when tackling more complex problems, it still needed guidance, resembling the work of a capable but inexperienced graduate student.

Tao believes that with further development and integration of tools like computer algebra systems, AI could soon assist in advanced research. While current models require effort to prompt and verify, he sees potential for broader use in mathematics as they continue to improve.

⛓ More information...

#news #artificial_intelligence

🤖 Insecure Deebot robots collect photos and audio to train AI

Ecovacs Deebot vacuums have been caught gathering photos, videos, and voice recordings inside users' homes to train AI models. Despite privacy concerns, the company claims users willingly participate in a data collection program aimed at improving product functionality.

The program involves collecting 2D/3D house maps, microphone recordings, and camera images, some of which may be retained even if deleted. While Ecovacs says the data helps enhance AI capabilities, the company is under scrutiny due to cybersecurity flaws that could expose sensitive user information.

A similar issue arose with iRobot when intimate images taken by its vacuums were leaked online by contractors analyzing the footage, raising broader concerns about how companies handle user data. These incidents highlight the growing need for stronger safeguards to protect consumer privacy in smart home devices.

⛓ More information...

#news #robots #artificial_intelligence #security

🤖 Chinese Company Unitree Robotics Pushes Boundaries in AI-Powered Motion Control

Unitree Robotics, a company based in 🇨🇳 Hangzhou, China, has made impressive strides in motion control technology using artificial intelligence. In a recent video, the company showcased its ability to teach robots human-like walking within just two days through reinforcement learning.

The video opens with 🤖 Unitree’s humanoid robot, the G1 (priced at over $16,000), walking fluidly across a room. In another segment, 🐕 Unitree’s robot dog Go2, demonstrates remarkable balance on two legs—even as it’s nudged and struck with a stick by a human.

This development highlights the potential of AI-powered robotics to adapt and acquire new skills quickly, a crucial capability as robots become more integral to our daily lives.

⛓ More information...

#news #robotics #artificial_intelligence

🐟 How to Run DeepSeek R1 on Raspberry Pi 5

DeepSeek R1 can run on the 🍓 Raspberry Pi 5, enabling local AI experimentation without cloud reliance. Setting it up is simple, making it accessible even for beginners. However, running the model on the Pi 5 comes with limitations, requiring creative workarounds for better performance.

Features of this approach:
➡️Optimized distilled LLM for Raspberry Pi 5’s 8GB RAM.
➡️No AI acceleration; external GPU possible via PCIe.
➡️Simple setup using ollama’s installation script.
➡️Much slower than desktop GPUs, with noticeable latency.

This guide is perfect for AI enthusiasts and developers exploring LLMs on low-cost hardware. While the Pi 5 can handle basic tasks, serious AI work requires more powerful setups. Still, it’s an interesting experiment for those curious about running AI on compact devices.

⛓ More information...

#deepseek #tutorials #raspberry #artificial_intelligence