انتقل إلى المحتوى الرئيسي

2023-04-01

Twitter's Recommendation Algorithm Now Open-Sourced​

Twitter's algorithm has faced scrutiny for potentially favoring right-wing American political content but Twitter denies nefarious intent. Documents revealed Twitter employees with a leftist ideological bias spending time blacklisting right-wing politicians. Twitter CEO requested transparency but only a fraction of the documents have been released. Elon Musk requested Twitter to remove a feature that tracked engagement metrics for important accounts, but its existence and potential reasons for creation is debated. Twitter released its natural language processing algorithm, but there is limited transparency, making it difficult to evaluate. Twitter open-sourced its recommendation algorithm, "Home Mixer," consisting of three-stage pipelines to increase user engagement. Twitter's "For You" timeline uses in-network and out-of-network sources, a neural network, and heuristics to provide a diverse and balanced feed to users.

LLaMa.cpp 30B Model Achieves Significant Performance Improvements with Reduced RAM Usage​

LLaMa.cpp 30B model runs with only 6GB of RAM making it easy to integrate into other applications, sparking discussions on distributed training ideas in the open-source community. Despite some hardware and quantization issues, the progress made in optimizing large language models is impressive, and the potential for running models such as this on local hardware is exciting. A new PR for llama.cpp project aims to make loading of model weights 10-100x faster, allowing for faster inference and concurrent inference processes. The llama.cpp project introduces a new file format and a migration tool, praised for great work by reviewers, but some users encountered issues migrating their models. Discussions on the ethical and legal implications of using and sharing models like LLAma are ongoing among users.

Developers discuss pros & cons of AI-enhanced development tools​

Developers discuss the advantages and limitations of AI-enhanced development tools such as ChatGPT, their potential in accelerating software development, and their impact on crucial software development fields. Mixed opinions exist about ChatGPT's usefulness in creating code, with some finding it helpful for suggesting high-level designs and others cautioning about blindly relying on its output. Codex and other AI-generated code tools are seen by some as a massive step forward, though some argue that they are only useful for simple tasks. The potential implications of AI-generated code on software reliability are being discussed as it has the potential to increase productivity, but also amplify bad practices if not guided correctly. The Tech Times editor highlighted a project developed using AI-enhanced development tools, demonstrating the potential of these tools to enable developers to tackle complex projects with more ease, reducing the effort needed significantly.

Experts discuss the pros & cons of using Postgres as a graph database​

Experts discuss advantages and limitations of using Postgres as a graph database, citing difficulty with recursive queries and scalability issues with large edge tables. However, Postgres can be a suitable option for simple use cases, with some suggesting specialized extensions like Apache Age or native graph databases like Memgraph or EdgeDB. Recursive queries with CTEs and materializing paths enable efficient traversal of graphs, avoiding the need for repetitive table joins. Postgres is becoming more competitive with its evolving capabilities, including potential Worst-Case Optimal Joins.

Italian Privacy Regulator Bans OpenAI's ChatGPT Over Privacy Concerns​

Italian privacy regulator bans OpenAI's ChatGPT over alleged privacy violations, particularly for exposing minors to unsuitable content without verifying age. Critics argue that the ban is unnecessary and could limit innovation. The status of privacy regulations in different countries, like GDPR, is questioned. The comments section features debates on finding a balance between innovation and regulation and the need for technology ministers to have technical expertise. The ban highlights the challenge of enforcing GDPR's extraterritorial jurisdiction worldwide. The Tech Times covers news and analysis on various policy areas but is not the correct platform for this content.

NYPD non-compliance with surveillance laws raises concerns in tech community​

The NYPD's non-compliance with surveillance technology laws in New York City is causing concern in the tech community. Users on Hacker News discuss the prevalence of surveillance, challenges of surveillance footage as evidence, and concerns about government surveillance leading to a surveillance state. There is debate over whether it is the responsibility of law enforcement to decide the constitutionality of laws. Police accountability and oversight are crucial, but there is minimal accountability when it comes to the NYPD. The NYPD is also accused of not disclosing information about its surveillance technologies. The lack of police accountability and transparency highlights the need for police reform.

Feasibility & Future of Building Low-Cost Language Models for Web Browsers​

Building low-cost language models for web browsers is being explored by tech experts with a focus on using fine tuning techniques against the LLaMA model to emulate GPT-3. Train-on-idle-GPU frameworks are being proposed to reduce the cost of training models, but challenges in parallelizing the training process and inter-node latency remain. Some suggest that creating an open-source, non-profit AI clone of OpenAI could be a solution. Projects using lightweight models like Alpaca from Stanford are attracting attention for their accessibility and low hardware requirements. New models like LLaMA and Alpaca could face increasing competition from openly licensed models in the future. Advancements in WebAssembly and WebGPU are enabling the stable diffusion image model and the Hugging Face Transformer library of models to run entirely in the browser. Safe web browser capabilities are crucial, and the new H100 Tensor Core GPU from Nvidia provides 30x faster performance over their previous model.

Facial recognition tech misuse raises concerns for civil liberties​

A wrongful arrest of an innocent man using facial recognition software highlights concerns about the technology's reliability and infringement on civil liberties. The case underscores the need for stricter regulations and responsible management of emerging technologies. E-warrants and automated matching also raise concerns about false positives and rubber-stamping by judges. The combination of AI technology with human social factors has resulted in complex damages. The lack of laws and regulations to control the use of such technologies is cause for caution. The misuse of technology should not go unchecked, and the importance of ethical and legal standards must be emphasized.

Valve's Steam Deck Gains Popularity Among Gamers & Developers​

Valve's Steam Deck, a handheld gaming PC with a Linux-based operating system and user-friendly compatibility layer, is gaining popularity among gamers and indie game developers due to its console-like experience, powerful performance, and support from passionate communities of gamers and hackers. Users find the controls of the Deck comfortable and a significant improvement over the previous Steam Controller. Protondb's user feedback and tweaks help users evaluate compatible games. Though some users report issues with certain games and hardware quality, the Steam Deck's popularity and success may potentially lead to a migration towards Linux for PC gaming. The device's convenience, smooth gameplay, and potential as a remote play companion are highly praised by users. Developers express interest in receiving a free/deeply discounted Steam Deck from Valve to capitalize on its increasing traction among gamers. Valve's testing metrics and Resident Evil 4 Remake's expected release point to a high percentage of supported games in the near future. BoilingSteam's reports on supported games and predictions for Linux gaming can also benefit readers.

CoScreen V5: New Screen Sharing App for Collaborative Projects​

CoScreen V5 is a new screen-sharing app for simultaneous and multi-directional sharing with audio and video chat, featuring high-definition code sharing, reduced CPU utilization, and latency. CoScreen has received much attention due to its unique features, which include flexibility for remote pair programming, brainstorming, and other collaborative projects. Currently available on macOS and Windows, support for Linux and a web client is in development. While the tool is currently free, plans for enterprise features and charging models are in the works. The latest version of CoScreen, V5, allows for individual mouse pointers, native window sharing, and a responsive video chat UI with advanced noise reduction and background blurring. It provides a seamless collaborative experience with optimized smoothness and low latency.

Janet Language: A New Way to Create Redistributable CLI Apps​

Janet Language, a LISP-inspired programming language, is being considered as an excellent option for creating redistributable CLI apps due to its ability to compile programs into statically-linked native binaries. It is easy to distribute even to non-technical users and has small binaries, making it convenient for making CLI or desktop apps, especially those targeting non-technical users. Janet's feature-rich ecosystem makes it a compelling choice, particularly for non-technical users. Its entire ecosystem is less than 1MB, and it boots up considerably faster than larger languages such as Python. Janet shares many similarities with Lua, making it an excellent choice for text-wrangling. A book on Janet, "Janet for Mortals" by Ian Henry, caters to those who already know how to program; the language is small, simple, usable on Windows, and has built-in concurrency, multithreading, and support for parsing expression grammars.

CDC team falls ill investigating hazardous Ohio train derailment​

A CDC team fell ill during their investigation of the East Palestine train derailment, raising concerns over downplayed incident severity and the adequacy of protection and coverage on the food supply chain. The incident's potential impact on human health remains unknown, and the state's technology and infrastructure may be questioned. Catastrophes involving chemicals or biological agents may have societal and engineering backgrounds, exacerbating their unpredictability. Seven U.S. government investigators also fell ill, further confirming potential health impacts of the toxic train derailment. While officials claim everything is safe, residents remain skeptical, and physical complaints may be driven by fear and anxiety.

Efficient Cosine Computation with Polynomial Approximation in C: Challenges, Trade-Offs, & Insights​

Efficient polynomial approximation of cosine in C takes precise derivatives and range reduction with modulo pi/2. It supports double-double precision and uses GPU/CPU hardware implementations. Discussion on Hacker News shares insights on Remez algorithm, Parks McClellan filter design, and decimal floating-point. The post provides a detailed account of sine & cosine algorithm implementations, trade-offs, and insights in numerical analysis. The context & requirement matter, expertise can differ, and source code for cosine function is a bonus. Programming in AI requires high-level math skills, while general programming & neural nets do not.

SVG's Potential as an Open Standard Version of Flash Discussed Amid Debate Over Flash's Decline & iPhone's Removal of Headphone Jack​

The potential for SVG to become an open standard version of Flash is discussed in a Hacker News post exploring SVG's network support for raw sockets. Flash was widely used in the early 2000s but declined for various reasons such as poor performance, security issues, and failure to adapt to mobile devices. Some mixed opinions around iPhone's removal of the headphone jack are discussed, along with debates around Bluetooth audio quality and charging. SVG was originally positioned as a rival to Flash and could potentially replace it as an authoring tool. However, SVG files can lose fidelity when zooming in on a pixel-oriented image but have potential for full applications with websockets and webrtc.

Heat pumps surpass gas furnaces in US homes despite challenges​

Heat pump sales have surpassed gas furnace sales in the US due to energy efficiency improvements, but decision-making depends on local factors such as cost and climate, with natural gas remaining cheaper in some regions. Multiple heating and cooling methods are popular but come at a cost. Landlords face little incentive to install cost-saving technologies, making retrofitting of existing buildings difficult. The article provides valuable insight into the realities and nuances of home heating and cooling systems, including the importance of proper insulation and maintenance to maximize efficiency and cost savings. The International Energy Agency's analysis reveals that key drivers for the global growth in heat pump sales include policy support and incentives targeting greenhouse gas emissions reduction.