AI, Software and Business Update

#AI

• Google explains Gemini’s ‘embarrassing’ AI pictures of diverse Nazis

The controversy surrounding Google's Gemini AI model, which was criticized for its "woke" image generation bias, underscores the burgeoning cultural and political debates over AI's underlying values. Google's attempt to promote diversity in its AI algorithms led to unintended outcomes, such as depicting women and people of color in historically inaccurate contexts, leading to public apologies and internal memos from the CEO acknowledging the issue.

Why it is important:

This incident reveals the complex challenges of balancing ethical considerations, historical accuracy, and societal expectations in AI development. The backlash from conservative circles, highlighting perceived liberal biases in AI responses, further complicates the discourse. Efforts to correct biases in AI, often seen as quick fixes, fail to address deeper systemic issues, raising concerns about the technology's future direction. As AI technology advances and plays a more significant role in decision-making, the debate over its values, biases, and the impact on society is set to intensify, making it a critical issue worth understanding.

• Microsoft partners with OpenAI-rival Mistral for AI models, drawing EU scrutiny

Microsoft announced a collaboration with Mistral, a French AI company, integrating Mistral's AI models into its Azure platform, accompanied by a 15 million euro non-equity investment. This partnership has caught the attention of EU regulators due to concerns over Microsoft's potential equity in Mistral and the broader implications for competition and regulatory compliance in Europe. The deal reflects Microsoft's strategy to enhance its AI offerings and has sparked scrutiny amidst its dominant cloud computing market position and Mistral's lobbying for relaxed AI regulations.

Why it is important:

This partnership highlights the growing concern over American Big Tech's influence on European technological innovation and market competition. EU regulators' scrutiny of the Microsoft-Mistral deal underscores the delicate balance between promoting innovation and preventing market dominance by non-European giants. It raises critical questions about safeguarding a competitive digital economy in Europe, emphasizing the need for regulatory measures that support both competition and innovation without succumbing to external influences. This situation exemplifies the complexities of global tech competition and the strategic importance of fostering a diverse and competitive AI ecosystem within Europe.

#Software Craftsmanship

• White House urges developers to dump C and C++

The Biden administration is pushing for a change in how software is made, suggesting that developers stop using programming languages like C and C++ because they are not safe from cyberattacks. Instead, they want programmers to use languages that are better at protecting against bugs and hacks, such as Rust.

This advice comes from a report that says using safer programming languages could prevent a lot of the common security problems that make software vulnerable. The idea is to make the internet and digital tools safer for everyone by making these changes at the level of big companies and government, rather than expecting individual people or small businesses to handle these cybersecurity risks. Experts agree that moving away from C and C++ is a good idea because of the risks, but they also say it will take time and effort to make this big change.

• Decision Archeology: Using ADRs with Existing Products

Developers and organizations often face challenges with decisions made before the implementation of Architecture Decision Records (ADRs). The question arises whether it's worth creating ADRs for these historical decisions. The answer is affirmative due to the significant benefits, such as enhancing architectural decision knowledge and allowing for the revision of past decisions with new insights. This process not only aids in better understanding and preserving historical decisions but also prevents the inadvertent alteration of these decisions based on outdated information.

To address this challenge, a structured method for documenting historical ADRs is recommended, utilizing all available documentation and insights from team members to reconstruct past decision-making processes. This collaborative and iterative approach ensures that assumptions and gaps in information are clearly identified and validated, contributing to a more comprehensive and accurate documentation. By systematically recording historical decisions, organizations can maintain a cohesive understanding of their architectural evolution, facilitating informed decision-making and minimizing the risk of unnecessary revisions in the future.

• How Reddit Scaled Its Authentication to Millions of Decisions Per Second

Reddit faced a significant scaling challenge with its advertising platform's authorization system, needing to support millions of decisions per second while adapting to dynamic user demands and maintaining system performance. This situation was complicated by the lack of an existing generalized authorization service that could handle Reddit's unique requirements, such as a dynamic user hierarchy and customizable permissions.

To overcome these challenges, Reddit decoupled policy from code, inspired by Google's Zanzibar, allowing for independent management of authorization policies and application logic. This approach was implemented using Open Policy Agent (OPA) for policy evaluation, enhancing scalability and flexibility. Additionally, Reddit explored OPAL (Open Policy Administration Layer), an open-source solution that provides real-time policy and data synchronization with policy engines like OPA. OPAL uses Git repositories for rule storage, offering benefits such as version control, rollback capabilities, collaboration, and automated deployments.

Are you facing similar challenges with scaling and optimizing authorization systems as described in the Reddit case?

If so, VirtusLab has experience in addressing such issues for our clients. We specialize in developing advanced solutions that help our clients achieve operational goals while ensuring system availability, security, and flexibility.

We invite you to explore our case study in the retail sector, which showcases how we effectively tackle complex technological problems: Building a Risk Detection Engine with Kafka & Stream.

#Business

• Apple's Core Technology Fee explained - Who'll be paying what (and is this even legal?)

Apple has updated its app installation policies for iPhones in Europe in response to the European Union's Digital Markets Act (DMA). The DMA aims to prevent tech giants from monopolizing the market, prompting Apple to allow third-party app stores on its devices for the first time. This is a significant shift from Apple's previous policy, where it had exclusive control over app installations. The move is designed to comply with EU regulations aimed at opening up competition and giving users more choices.

The specific changes introduced by Apple include allowing users to download app store apps directly from websites, which can then be set as the default app store on their devices. Although the initial app store app downloaded will undergo Apple's security checks, subsequent apps installed from these third-party stores won't be scrutinized by Apple to the same extent. To continue earning revenue, Apple has introduced a new fee structure: a 17% commission for apps using alternative payment systems (reduced to 10% for small businesses), a 3% processing fee, and a Core Technology Fee for apps exceeding one million annual installs in the EU. This approach aims to maintain Apple's profitability while adapting to the new regulatory environment.

Created by Artur Skowroński