- AI
- A
Echo Digest: from Russian laptops to remote surgery
Overview of the latest achievements in technology: Russian laptops on Elbrus, breakthroughs in telemedicine, innovations in ADHD treatment, and AI development in various industries.
Hardware News
Russia
The author of the Telegram channel "Azhal" was shown a prototype laptop on Elbrus 2s3 from the company Promobit (Bitblaze).
In Russia, the design of domestic equipment for electronic engineering has begun.
The company "Mars Technologies" plans to launch the production of equipment for creating private cellular networks in the SEZ "Innopolis" in 2025, aimed at large corporate customers, in particular, enterprises of the oil and gas chemical complex and the mining industry.
The holding "Roselectronics" (part of the Rostec State Corporation) has begun supplying domestic transformer substation parameter meters "Coefficient-M-2022C" for mobile electrical laboratories, which will replace previously used foreign analogues and will be able to operate in harsh climatic conditions.
World
Surgeons from Switzerland successfully performed a remote endoscopy on a pig in Hong Kong at a distance of 9300 km using a magnetic endoscope controlled through a game controller, opening up prospects for remote operations on humans in hard-to-reach regions.
Australian startup Neurode has developed an innovative wearable headband that uses mild electrical stimulation of the prefrontal cortex to treat and track ADHD symptoms, offering an alternative to traditional stimulants and potentially expanding the technology's application to other cognitive disorders.
Former Samsung Electronics employees have been arrested in South Korea on suspicion of stealing and transferring DRAM memory production technologies to China, causing the company an estimated damage of $3.2 billion.
German semiconductor manufacturer Infineon has announced a technological breakthrough in the production of gallium nitride (GaN) chips on 300-millimeter wafers, which will allow the company to capture a significant share of the growing GaN chip market, which is expected to reach several billion dollars by the end of the decade.
India is actively developing its own semiconductor industry, attracting major electronics manufacturers (in particular, Apple) and investments, with the aim of becoming an alternative to China in chip production, planning to increase the industry's turnover from the current $155 billion to $500 billion by the end of the decade.
AI News
Russia
MTS AI has developed a new version of the large language model Cotype Lite, capable of processing and analyzing texts in the Tatar language, which will speed up document processing in various organizations in Tatarstan and support the development of the Tatar language in the digital age.
Norilsk Nickel is developing a "chief metallurgist's assistant" service based on GPT technology, which will contain all company regulations and scientific articles for quick provision of action algorithms in unforeseen situations.
Yadro, in collaboration with the Higher School of Digital Culture ITMO, is launching a one-year educational program Yadro AI School for third-year undergraduate students and above, aimed at in-depth study of artificial intelligence, including courses in Python, statistics, natural language processing, and deep learning, with the possibility of subsequent internships at the company for the best graduates.
Scientists from the St. Petersburg Federal Research Center of the Russian Academy of Sciences have developed a service using artificial intelligence and digital twins to model and prevent cyberattacks on critical digital infrastructure objects, allowing vulnerabilities in systems to be identified at early stages and significantly increasing their security.
MTS Web Services (MWS), a subsidiary of PJSC MTS, is investing more than 1 billion rubles by the end of 2025 in the creation of an AI cloud - a new direction that will provide companies with infrastructure with graphics accelerators (GPU) and a range of ready-made services to simplify and accelerate the implementation of artificial intelligence in business processes.
World
OpenAI has introduced a new artificial intelligence model o1 (code name Strawberry), which is capable of solving complex problems through step-by-step reasoning.
The study showed that ideas generated by large language models (LLMs) are rated by experts as more innovative than those of a human expert, although slightly less feasible.
According to a survey by the Federal Reserve Bank of New York, the introduction of artificial intelligence in companies has not led to a significant reduction in jobs, but on the contrary, may contribute to employment growth in the near future.
The lead researcher at OpenAI for GPT-4o/GPT-5 is leaving to start his own company.
The author proposes an effective four-step approach to working with AI (specifically Sonnet 3.5) for software development, which includes sequential interaction with AI in the roles of software engineer, product manager, technical lead, and developer, significantly improving the quality of the generated code.
Act as a software engineer:
Ask the AI to act as a software engineer.
Provide all relevant files and ask the AI to explain the code.
This helps the AI understand the context and existing codebase.
Act as a product manager:
Ask the AI to act as a product manager.
Clearly describe the new feature you want to develop.
If necessary, ask to create a product requirements document.
Act as a technical lead:
For complex features, write pseudocode first.
This gives the AI a chance to review the codebase and your requirements again.
Helps determine which files and sections of code need to be changed.
Act as a developer:
Ask the AI to write actual code based on the pseudocode and previous steps.
After completing the coding, ask the AI to create a git commit message.
Models from Hugging Face
OpenVLA is a set of open multimodal models with 7 billion parameters that can control robots by taking language instructions and visual data as input.
Mini-Omni, a new open-source multimodal language model developed based on the Any Model Can Talk method, offers real-time speech processing capabilities, including speech-to-speech conversion without additional ASR or TTS models, simultaneous text and audio generation, and streaming playback.
Raspberry Pi
An enthusiast created a "smart" shelf for his vinyl record collection using a Raspberry Pi 4 and several ESP32s to control LED strips that illuminate the location of the selected record.
A student-developer created Mosaico — a platform powered by Raspberry Pi that allows Python scripts to come to life on an LED matrix, with the ability to test in a simulator without the need for real hardware.
Running Phi-3/Mistral 7B LLMs on Raspberry Pi 5: A Step-by-Step Guide
Python
Transcribing Russian and English speech using Whisper and Python
LightAPI — a new lightweight framework for quickly creating API endpoints in Python, using native libraries and providing simplicity, performance, and asynchronicity through SQLAlchemy and aiohttp.
PyJSX — a library that allows writing JSX-like code directly in Python.
A developer has open-sourced his web scraping tool Stealth-Requests, which mimics browser behavior when sending requests, making the process of collecting data from websites more efficient and less noticeable.
A library called Dict Hash has been developed, allowing efficient hashing of dictionaries, Pandas and Polars DataFrames, NumPy arrays, and Numba objects.
EasySubber automates the creation of subtitles for video files using OpenAI's Whisper speech recognition model and FFmpeg for audio processing.
Meatie automatically generates code for working with REST APIs based on type annotations.
SyncStar — an application for creating bootable USB drives with the selected operating system.
Linux
KDE Plasma 6.1.5 released
Beauty
Analog drum machine with digital control.
What to watch
Waiting for GPT-5: a journey to the heart of the AI revolution
Diving into the world of artificial intelligence through the lens of travels across the USA and Japan, attending the Hot Chips conference, and reflecting on the future of AI. From the rise of Waymo to the anticipation of GPT-5 — what is the real economy of self-driving cars and what awaits the industry in the coming years?
Write comment