SP-Life

SP Lover's notes

SSPAI Morning Brief: OpenAI Launches GPT-5.5 Instant With Improved Accuracy and Faster Responses

,

Morning Brief

  1. OnePlus and realme merge
  2. Mac mini with M4 chip (256GB version) discontinued
  3. Doubao confirms upcoming paid version
  4. OpenAI launches GPT-5.5 Instant
  5. Linux exposed to “Copy Fail” vulnerability
  6. Daemon Tools hit by supply chain attack
  7. 2026 LG Gram laptop lineup goes on sale
  8. Bose unveils Lifestyle Collection speakers
  9. U.S. Department of Defense reaches AI agreements with multiple tech companies
  10. Google rolls out source preference settings globally
  11. AMD brings HDMI 2.1 compatibility update to Linux
  12. Spotify introduces real-user verification badge
  13. Meta news roundup
  14. Text messages I just can’t seem to delete

OnePlus and realme merge

On April 29, OPPO announced internally the establishment of a new sub-series business unit, integrating the marketing and sales systems of OnePlus and realme. The unit will be led by OPPO Senior Vice President Sky Li (Li Bingzhong), with former realme marketing president Xu Qi serving as head of marketing. On the product side, OPPO has created a sub-series product center with separate domestic and overseas divisions, both overseen by Li Jie, who reports to Pete Lau. Former realme vice president Wang Wei (Derek) will serve as deputy general manager of the product center, reporting to Li Jie. In terms of R&D, realme’s engineering teams have been reintegrated into the group, with imaging, hardware, and other departments merged into OPPO’s existing hardware divisions.

According to an exclusive report by Leifeng.com, OPPO had already begun relocating realme teams to its Binhai Bay headquarters in batches since late March. OPPO employees indicated that following the integration of realme and OnePlus, the company will place greater emphasis on product line reuse. Source

Mac mini with M4 chip (256GB version) discontinued

On May 1, Apple discontinued the 256GB storage version of the Mac mini with the M4 chip globally. Including China, the base configuration on the Apple Store now starts at M4 chip, 16GB memory, and 512GB storage, priced at RMB 5,999. The 256GB version is no longer available for purchase.

Earlier, CEO Tim Cook stated during an earnings call that supply for Mac mini and Mac Studio remains constrained, and that demand driven by AI and agent-based tools has exceeded expectations. He noted that it may take several months to rebalance supply and demand. Apple had already discontinued the 512GB memory version of the Mac Studio in March, and amid a global memory chip shortage, Cook expects memory costs to rise significantly this quarter. Source

Doubao confirms upcoming paid version

On May 4, Doubao trended on Weibo after updating its App Store page with details of upcoming subscription plans. According to the listing, Doubao will introduce three tiers: Standard at RMB 68/month (RMB 688/year), Advanced at RMB 200/month (RMB 2,048/year), and Pro at RMB 500/month (RMB 5,088/year). Specific feature differences between tiers have not yet been disclosed.

In response, Doubao stated that it will continue offering free services, while exploring premium features to meet more diverse user needs. Details are still under testing and will be officially announced later. Reports suggest that paid features will primarily target complex tasks and productivity scenarios. Source

OpenAI launches GPT-5.5 Instant

On May 5, OpenAI announced the release of GPT-5.5 Instant, replacing GPT-5.3 Instant as the default model in ChatGPT and made available via API under the name “chat-latest.”

GPT-5.5 Instant improves factual accuracy, image analysis capabilities, STEM question answering, and web search decision-making. In internal evaluations across high-risk domains such as medicine, law, and finance, hallucinations were reduced by 52.5% compared to GPT-5.3 Instant. In difficult conversations flagged by users for factual errors, inaccuracies dropped by 37.3%. Output length was also reduced, with average word count decreasing by 30.2% and line count by 29.2%, alongside optimizations to minimize unnecessary follow-up questions, excessive formatting, and irrelevant emoji usage. The model also enhances personalization by leveraging past chats, uploaded files, and connected Gmail accounts (with user permission), reducing the need for repeated context input.

Enhanced personalization features are now rolling out on the web for Plus and Pro users, with mobile support coming soon and broader availability planned for Free, Go, Business, and Enterprise tiers in the coming weeks, subject to regional differences. GPT-5.3 Instant will remain accessible to paid users for three months before being fully retired. Source

Linux exposed to “Copy Fail” vulnerability

On April 29, the Xint Code Research Team disclosed a Linux kernel vulnerability named “Copy Fail” (CVE-2026-31431). The flaw allows a local low-privilege user to trigger a deterministic and controllable 4-byte write into the page cache of any readable file on the system. The researchers stated that a Python 3.10+ script of only 732 bytes is sufficient to achieve root privilege escalation on mainstream Linux distributions. Because the kernel does not mark the corrupted pages as dirty, the disk files remain unchanged, making detection through conventional disk integrity checks extremely difficult.

The vulnerability path has existed since 2017. The team confirmed successful exploitation on multiple systems, including Ubuntu 24.04 LTS (6.17.0-1007-aws), Amazon Linux 2023 (6.18.8-9.213.amzn2023), RHEL 10.1 (6.12.0-124.45.1.el10_1), and SUSE 16 (6.12.0-160000.9-default). A patch (a664bf3d603d) was merged into the mainline kernel on April 1, 2026. The researchers recommend upgrading the kernel or applying temporary mitigations such as blocking AF_ALG socket creation via seccomp or disabling the algif_aead module. Source

On May 4, the U.S. cybersecurity agency CISA also issued a warning, stating that Copy Fail is already being exploited in the wild and requiring U.S. federal civilian agencies to patch affected systems by May 15. Source

Daemon Tools hit by supply chain attack

On May 5, Kaspersky revealed that the well-known disk imaging tool Daemon Tools had been compromised in a supply chain attack lasting approximately one month. The attack began on April 8 and was still ongoing at the time of reporting. Affected versions include 12.5.0.2421 through 12.5.0.2434. The malicious installers were signed with the developer’s official digital certificate and distributed via the official website. Technical analysis suggests that the infected versions are likely limited to Windows. The compromised builds modify the Daemon Tools executable and execute malicious code at system startup. In the initial stage, the malware collects information such as MAC address, hostname, DNS domain, running processes, installed software, and system locale, then transmits it to attacker-controlled servers.

Kaspersky reported that thousands of devices across more than 100 countries were affected, with around 10% belonging to enterprises or organizations. Among them, approximately 12 devices in sectors such as retail, research, government, and manufacturing were subjected to follow-up attacks, including remote command execution, file downloads, in-memory shellcode execution via a lightweight backdoor, and even deployment of QUIC RAT on a device in a Russian educational institution. This suggests selective targeting by the attackers. Kaspersky recommends that users perform full system antivirus scans and closely monitor executable files launched from directories such as Temp, AppData, and Public for suspicious code injection into legitimate system processes. Source

2026 LG Gram laptop lineup goes on sale

On May 1, LG announced the release of its 2026 LG Gram laptop lineup. Compared to previous models, the new devices adopt LG’s proprietary aluminum-magnesium alloy “Aerominum,” further reducing weight while improving durability and scratch resistance. The 16-inch LG Gram Pro weighs roughly the same as a 13-inch MacBook Air, and most models meet military-grade durability standards (except the LG Gram Book).

The LG Gram Book targets students and everyday users, with the 15.6-inch version starting at $1,150, equipped with an Intel Core Ultra 5 processor, 16GB memory, and a 1TB SSD. The 16-inch version starts at $1,500, featuring a Ryzen AI 7 processor, 32GB memory, and a 1TB SSD.

The standard 15.6-inch LG Gram starts at $1,600 with a Ryzen AI 7 450, 32GB memory, and 1TB SSD. The 17-inch LG Gram Pro starts at $1,650 with an Intel Core Ultra 7, 16GB memory, and 512GB SSD. The 16-inch LG Gram Pro starts at $2,400 with a Ryzen AI 7 450, 32GB memory, and 2TB SSD. The 16-inch LG Gram Pro 2-in-1 starts at $2,200, featuring a Core Ultra 7, 32GB memory, 1TB SSD, and a 360-degree hinge.

All models are now available for order on LG’s official website and will roll out to retail partners. Select models come with a $1-per-year Premium Care service plan if purchased before May 10. Source

Bose unveils Lifestyle Collection speakers

On May 5, Bose introduced the Lifestyle Collection home audio lineup, including the Lifestyle Ultra Speaker, Lifestyle Ultra Soundbar, and Lifestyle Ultra Subwoofer. All three products support Google Cast and AirPlay, and can be combined into a full home theater system or used for multi-room audio playback.

The Lifestyle Ultra Speaker is a compact unit with a knitted fabric grille and touch-and-slide controls on the top. It features three drivers (two front-facing and one upward-firing), supports Alexa+, stereo pairing, Bluetooth, Wi-Fi, and 3.5mm input, and uses TrueSpatial processing to enhance spatial audio, along with CleanBass and QuietPort technology for improved low-frequency performance. The Lifestyle Ultra Soundbar is a 5.2-channel system with six full-range drivers, including two upward-firing units, a dedicated center tweeter for dialogue clarity, and two PhaseGuide drivers for directional sound. Adding two Ultra Speakers or a Subwoofer upgrades it to a 7.1.4 system. It features a glass top design, gesture controls, and technologies such as CustomTune room calibration, SpeechClarity AI dialogue enhancement, TrueSpatial, and CleanBass.

The Ultra Speaker is priced at $299 (black/white) or $349 (limited brown edition), the Soundbar at $1,099, and the Subwoofer at $899. Pre-orders are open now, with availability starting May 15. Source

U.S. Department of Defense reaches AI agreements with multiple tech companies

On May 1, the United States Department of Defense announced agreements with NVIDIA, Microsoft, Amazon, and Reflection AI, allowing the deployment of AI technologies and models within classified networks for “legitimate strategic purposes.” Previously, similar agreements had been signed with Google, SpaceX, and OpenAI.

These AI systems will operate in Impact Level 6 (IL6) and Impact Level 7 (IL7) environments, supporting data integration, situational awareness, and strategic decision-making. The Department of Defense stated that the goal is to avoid reliance on a single AI vendor and provide greater technological flexibility.

This move follows a dispute between the Department of Defense and Anthropic over AI usage terms. The Pentagon had sought unrestricted access to Anthropic’s AI tools, while Anthropic pushed for safeguards to prevent domestic mass surveillance and autonomous weapons use. The dispute is currently in litigation. Source

At the same time, The New York Times reported that the White House is considering establishing a new AI oversight task force. Unlike its previously more relaxed stance outlined in the AI Action Plan, the proposed group may require federal review before the public release of new AI models, though details remain undecided. Source

Google rolls out source preference settings globally

On April 30, Google announced the global rollout of the “Source Preferences” option in its search feature. First introduced in the second half of 2025 and initially limited to select regions and languages, the feature expanded to English users worldwide by the end of 2025 and is now fully available. “Source Preferences” allows users to customize which media outlets or websites are prioritized in news search and Top Stories, rather than relying entirely on algorithmic ranking. However, it does not block other sources. Users can manage their preferences on the relevant settings page. Google stated that the tool now covers more than 200,000 independent websites1. Source

AMD brings HDMI 2.1 compatibility update to Linux

In the amdgpu driver patch notes released on May 1, AMD confirmed the addition of HDMI Fixed Rate Link (FRL) support on Linux, enabling near-complete HDMI 2.1 compatibility for AMD GPUs on the platform.

Compared to the TMDS standard used in HDMI 2.0 and earlier versions, FRL offers higher bandwidth and resolution, along with support for dynamic HDR and variable refresh rate (VRR). Developer Harry Wentland noted that Display Stream Compression (DSC) will be merged after testing, potentially enabling up to 10K resolution at 100 Hz. Full HDMI 2.1 compliance testing is also reportedly in progress.

This improvement benefits Linux gaming users, including Valve’s ecosystem, where previous bandwidth limitations required workarounds such as chroma subsampling and AMD FreeSync. Source

Spotify introduces real-user verification badge

On April 30, Spotify launched the “Verified by Spotify” badge to help users identify real human artists, addressing the growing presence of AI-generated artists, tracks, and impersonation content on the platform.

To qualify, artists must demonstrate a verifiable presence both on and off the platform, including live performances, merchandise, and linked social accounts, as well as sustained listener engagement. Artists whose primary output is AI-generated or who rely on AI personas are not eligible. Spotify also prioritizes artists with active fan interest and cultural contribution over purely functional music creators. At launch, more than 99% of actively searched artists are expected to receive verification, most of them independent musicians across various genres, stages, and regions. The badge—displayed as “Verified by Spotify” with a green checkmark—will appear on artist profiles and in search results in the coming weeks. Source

Meta news roundup

Meta Introduces AI-Based Minor Detection Using Visual Signals

On May 5, Meta announced a new AI system that scans photos and videos on Facebook and Instagram to detect users under 13 based on visual cues such as height and skeletal structure, removing accounts that violate policies. The system also estimates user age using visual themes, text, and interaction signals, including contextual clues like birthdays and school grade references. It is currently deployed in select countries and will expand further, including to Instagram Live and Facebook Groups. Source

Meta Ends Partnership with Data Labeling Firm Sama

According to BBC, Meta has terminated its contract with Sama, a Kenya-based data labeling company that previously provided video, image, and voice annotation services for the Ray-Ban Meta AI system. Sama stated that 1,108 employees are affected. Meta said the decision was due to Sama not meeting its standards, emphasizing that user privacy is protected and that human review requires explicit user consent. Sama denied any prior notice of non-compliance.

Earlier reports in February 2026 alleged that Sama employees had access to sensitive user footage captured by Ray-Ban Meta smart glasses, including private moments such as changing clothes or using the restroom, some potentially recorded without user awareness. Following the reports, a class-action lawsuit was filed in March in a U.S. federal court, and the UK Information Commissioner’s Office also raised concerns about transparency and user control in smart glasses handling personal data. Source

Meta Faces Copyright Lawsuit Over AI Training Data

Publishers Hachette, Macmillan, McGraw Hill, Elsevier, and Cengage, along with author Scott Turow, have filed a joint lawsuit against Meta and CEO Mark Zuckerberg, alleging unauthorized use of copyrighted works to train the Llama generative AI platform. The complaint claims Meta “copied and distributed millions of copyrighted works” without compensation, and that Zuckerberg personally authorized and encouraged the infringement despite knowing it violated copyright law. Source

Text messages I just can’t seem to delete

  • On May 5, Google announced that Chrome for Android now supports sharing “approximate location” with websites, allowing users to access services like weather and local news without revealing precise location data. The feature will expand to desktop Chrome in the coming months, along with a developer API.
  • On April 30, Microsoft open-sourced some of the earliest DOS source code, including the 86-DOS 1.00 kernel, PC-DOS 1.00 development snapshots, and tools like CHKDSK. The materials were originally preserved as printed documents and digitized by the DOS Disassembly Group. Microsoft had previously open-sourced MS-DOS 1.25, 2.0, and 4.0.
  • Bloomberg reported that Apple is in early discussions with Samsung and Intel to diversify its chip supply chain and reduce reliance on TSMC. CEO Tim Cook recently noted that Apple’s supply chain flexibility is currently below normal due to chip shortages and rising AI-related demand.
  • Mark Gurman reported that Apple plans to allow users to select third-party generative AI models in iOS 27, iPadOS 27, and macOS 27 via an “Extensions” mechanism, enabling integration with features like Siri and Writing Tools.
  • SamMobile reported that Samsung is developing a new Galaxy Book laptop running One UI 9 based on Android 17, with multiple configurations and a lightweight design, potentially integrating Google’s Aluminium system and an updated DeX experience.
  • On May 4, Apple released the iOS 26.5 Release Candidate, adding support for end-to-end encrypted RCS messaging between iPhone and Android devices based on GSMA standards.
  • Mystic Leaks reports that the Google Pixel 11 lineup will not include the “Project Toscana” facial recognition hardware, as the solution “did not meet release standards.” The project was previously rumored to improve the existing 2D camera-based face unlock on Tensor-powered devices, using infrared sensors to enable operation in low-light conditions and achieve speeds comparable to Face ID. It was also said to involve under-display IR hardware. In addition, there are reports that Google may remove the temperature sensor that has been included in the Pixel lineup for the past three years.
  1. Google didn’t say that. ↩︎
Tags: , ,

Leave a Reply