AI training Archives

Tech Optimizer

July 21, 2025

Scaling AI Inference at the Edge With Distributed PostgreSQL

Data is crucial for artificial intelligence, especially for inference workloads used in real-time decision-making across various platforms. Traditional centralized cloud-based AI inference struggles with demands for low latency and high availability, particularly in applications like autonomous vehicles and healthcare. Shifting AI inference to the edge reduces latency, enhances data privacy, and lowers bandwidth costs. Antony Pegg emphasizes the need for a multi-master active-active architecture that allows read and write operations at any node, ensuring data synchronization and high availability. Misconceptions about edge AI include beliefs that edge hardware can't handle AI workloads, that edge inference is limited to low-stakes use cases, and that centralized systems are necessary for data integrity. The shift to distributed inference can lead to reduced latency, faster insights, and lower costs, while supporting data compliance with regulations. Companies are adopting distributed PostgreSQL solutions with multi-master architecture for low latency and high availability. Enquire AI is an example of a company that improved performance and compliance by transitioning to pgEdge Cloud. This architecture allows for consistent data availability and supports scalable AI solutions at the edge.

AppWizard

July 14, 2025

How to disable Gemini AI on Android and keep control of your apps

Google is integrating its AI, Gemini, with Android systems, allowing access to applications like WhatsApp, Messages, and Phone, starting July 7, 2025. This integration may override existing privacy settings unless users disable Gemini. Users can restrict Gemini's access to apps through the Google Gemini app settings or uninstall it. If Gemini is not pre-installed, it will not be automatically installed on devices. Disabling Gemini maintains previous privacy settings, but conversations may still be reviewed by human reviewers for AI training, with data retention lasting up to three years. Users are advised to avoid sharing sensitive information in Gemini chats.

AppWizard

July 7, 2025

Google Gemini app gains automatic Android access to phone and messaging data

On July 7, 2025, Google Gemini app users experienced automatic activation of permissions for accessing phone calls, messaging services, WhatsApp, and system utilities on Android devices, following a privacy policy update on June 11, 2025. This change affects millions of users globally and allows user data to be reviewed by humans for AI training. Users can disable these permissions through the app's settings, but the option to opt-out is criticized for being difficult to find. Legal experts have raised concerns about compliance with GDPR regulations due to the lack of explicit user consent for these changes. The data collected supports machine-learning development and may be retained for up to three years, even after users delete their activity.

Winsage

July 4, 2025

Microsoft’s enshittification of Xbox, Surface, and even Windows itself, are by design — with greed and corner-cutting replacing innovation

Microsoft announced 9,000 layoffs, bringing the total for the year to over 15,300, including cuts within Xbox. The layoffs are attributed to unrealistic fiscal expectations set by CFO Amy Hood, creating an environment of fear and uncertainty. There is a perception that these layoffs are part of a broader shift towards artificial intelligence, with Microsoft executives and shareholders reportedly resenting the investment in Activision-Blizzard. The company plans to invest billions in new AI training centers globally, raising concerns about the impact on its core gaming and consumer products. Microsoft’s AI products have faced criticism, and the company has been accused of losing sight of the human element in its consumer businesses. The recent layoffs have left employees feeling uncertain about their futures, and there is a growing sentiment that Microsoft lacks a coherent mission or ethos, prioritizing short-term gains over long-term innovation.

AppWizard

June 28, 2025

Why You Should Turn Off This Gemini Setting on Your Android Phone

Beginning July 7, Google’s Gemini will allow users to manage their privacy settings more effectively. Previously, nearly all interactions with Gemini on Android devices, including chat histories and voice commands, were shared with Google by default for users aged 18 and older. Users will soon have the option to use Gemini as a phone assistant without enabling the “Gemini Apps Activity” feature, which means their interactions will not contribute to Google’s AI training database or be reviewed by humans. However, even with this feature disabled, Google will retain interactions for up to 72 hours for processing. Users can find the toggle for Gemini Apps Activity in the app settings and can review or delete past activities on their Google Account. Some data may still remain on Google’s servers if it has been used for training or review.

AppWizard

February 28, 2025

French Apex Legends voice cast goes to war over AI

French voice actors of Apex Legends, led by Pascale Chemin, have rejected an agreement that would allow their vocal performances to be used for training artificial intelligence. Chemin and 31 other voice actors have collectively declined to sign the controversial clause, which is believed to be introduced by Electronic Arts. They expressed their refusal in a letter to the publisher, emphasizing the need for a clause that protects their voices in contracts. Chemin highlighted the precarious financial situation many voice actors face and the broader implications of AI on their profession, stating that agreeing to such terms would undermine their expertise and support the technology that could replace them.

TrendTechie

February 7, 2025

Торрент-империя Meta: Цукерберг подарил миру теневую библиотеку с 160 ТБ книг

Meta is alleged to have unlawfully used pirated books for training its AI models, downloading at least 81.7 terabytes of data from torrent sources, including 35.7 terabytes from Z-Library. Internal communications reveal employee concerns about the legality of downloading pirated content, while leadership reportedly obscured their activities and operated in "stealth mode." Plaintiffs are seeking renewed interrogations of Meta personnel, claiming previous testimonies were misleading, and asserting that Meta's copyright infringements may involve distribution of pirated books. Meta defends its actions as falling under "fair use" and intends to challenge the allegations. Additionally, Meta and its products are prohibited in the Russian Federation due to being designated as extremist.

TrendTechie

February 4, 2025

Meta* обвинили в использовании контента с торрентов для обучения своего ИИ

Meta is facing a lawsuit in 2023 regarding the training of its LLM model, Llama, with allegations of using pirated content from torrent trackers. A judge has ordered the release of original documents, revealing internal discussions about the appropriateness of using torrents for AI training. An engineer raised concerns about using torrents on a corporate laptop, confirming the use of pirated content. There are indications that Mark Zuckerberg may have approved the use of such materials. Among the sources of pirated content was LibGen, a repository of pirated books and articles. Meta is defending its actions by citing the legal doctrine of "fair use."

Winsage

December 21, 2024

How to get started with Windows Recall on Windows 11

Windows 11 version 24H2 introduces a feature called Windows Recall, which captures and remembers on-screen activities using AI. Users can configure settings for Windows Recall, including opting in or out, adjusting storage allocations (25GB for 256GB SSD, 75GB for 512GB SSD, and 150GB for 1TB SSD), deleting snapshots, and filtering specific apps and websites. The feature can filter sensitive information and allows users to pause and resume snapshots. Users can search for past activities using natural language queries and interact with snapshots to perform actions like copying text or erasing objects. Windows Recall operates only when the device is in use and does not upload data to the cloud. The data remains on the user's device, with encryption for security. The feature supports multiple languages, including English, Chinese, French, German, Japanese, and Spanish.

Winsage

November 23, 2024

You can now try Microsoft’s Recall AI feature on a Copilot Plus PC

Windows Insiders in the Dev Channel are receiving a preview of the Recall AI feature for Copilot Plus PCs, which includes functionalities like Recall and Click to Do. Recall captures snapshots of user activities, allowing for easy retrieval through natural language queries and a scrollable timeline. It is an optional feature requiring user consent, initially available only on Qualcomm-powered Copilot Plus PCs, with plans to support Intel and AMD systems later. Users can control their snapshots, including deleting them and excluding specific apps from recording, while sensitive information is automatically protected. Microsoft assures that no snapshots are sent to the cloud or used for AI training, emphasizing user privacy and security. The Click to Do feature allows users to perform actions with text and images from captured snapshots and will eventually enable broader interactions with screen content. Recall's introduction was delayed due to security concerns, but Microsoft has enhanced its security framework, allowing users to opt-in or uninstall the feature.