Close Menu
    DevStackTipsDevStackTips
    • Home
    • News & Updates
      1. Tech & Work
      2. View All

      Can Good UX Protect Older Users From Digital Scams?

      June 25, 2025

      Warp 2.0 evolves terminal experience into an Agentic Development Environment

      June 25, 2025

      Qodo launches CLI agent framework

      June 25, 2025

      CodeSOD: Classic WTF: When it’s OK to GOTO

      June 25, 2025

      Microsoft is reportedly planning yet more major cuts at Xbox — as early as next week

      June 24, 2025

      Microsoft makes Windows 10 security updates FREE for an extra year — but there’s a catch, and you might not like it

      June 24, 2025

      “Deus Ex” just turned 25 years old and it’s still the best PC game of all time — you only need $2 to play it on practically anything

      June 24, 2025

      Where to buy a Meta Quest 3S Xbox Edition — and why it’s a better bargain than the “normal” Meta Quest 3S

      June 24, 2025
    • Development
      1. Algorithms & Data Structures
      2. Artificial Intelligence
      3. Back-End Development
      4. Databases
      5. Front-End Development
      6. Libraries & Frameworks
      7. Machine Learning
      8. Security
      9. Software Engineering
      10. Tools & IDEs
      11. Web Design
      12. Web Development
      13. Web Security
      14. Programming Languages
        • PHP
        • JavaScript
      Featured

      What Are the PHP Trends in 2025

      June 25, 2025
      Recent

      What Are the PHP Trends in 2025

      June 25, 2025

      Real-Time Observability for Node.js – Without Code Changes

      June 25, 2025

      Elevating API Automation: Exploring Karate as an Alternative to Rest-Assured

      June 25, 2025
    • Operating Systems
      1. Windows
      2. Linux
      3. macOS
      Featured

      NVIDIA Brings DLSS 4 to Budget GPUs with RTX 5050 for $249

      June 25, 2025
      Recent

      NVIDIA Brings DLSS 4 to Budget GPUs with RTX 5050 for $249

      June 25, 2025

      Program Execution, follow-up

      June 25, 2025

      I’ve Seen Things, pt II

      June 25, 2025
    • Learning Resources
      • Books
      • Cheatsheets
      • Tutorials & Guides
    Home»Development»Artificial Intelligence»Last Week in AI #313 – Fair Use Win, Gemini 2.5 Flash-Lite, MiniMax M1, Tesla Robotaxi

    Last Week in AI #313 – Fair Use Win, Gemini 2.5 Flash-Lite, MiniMax M1, Tesla Robotaxi

    June 25, 2025

    Top News

    Anthropic Nabs Partial Fair Use Win in Copyright Case

    A federal court in California has partially sided with Anthropic in a copyright lawsuit over the training of an AI model. The court ruled that Anthropic’s use of copyrighted books to train its LLMs may qualify as fair use, but its acquisition and storage of pirated digital copies does not. The lawsuit was filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who claimed that Anthropic built its LLM by using hundreds of thousands of copyrighted books without permission. Anthropic argued that its use of copyright-protected books, whether downloaded from pirate sources or scanned from purchased print copies, was protected under the doctrine of fair use.

    Judge William Alsup of the U.S. District Court for the Northern District of California divided his analysis into two core uses: the use of books to train language models and the creation of a permanent internal library of books, many of them pirated. He ruled that using copyrighted books to train AI models was “quintessentially transformative” and therefore fair use. He also found that Anthropic’s digitization of legally purchased print books was a permissible fair use. However, Anthropic’s decision to build its library using pirated books downloaded from illegal sources was not deemed transformative and therefore not protected under the doctrine of fair use. The court’s decision is among the earliest substantive decisions in a series of high-profile copyright cases against AI developers.

    More on this:

    • Order on Fair Use PDF

    • A federal judge sides with Anthropic in lawsuit over training AI on books without authors’ permission

    • Judge rules Anthropic did not violate authors’ copyrights with AI book training

    • Anthropic destroyed millions of print books to build its AI models

    Google’s Gemini AI family updated with stable 2.5 Pro, super-efficient 2.5 Flash-Lite

    Gemini 2.5: Updates to our family of thinking models - Google Developers  Blog

    Google has announced the expansion of its Gemini AI model family with the promotion of Gemini 2.5 Pro to be generally available and the preview release of Gemini 2.5 Pro Flash-Lite. The Gemini 2.5 Pro, which has been in preview for several months, is now stable and ready for developers to build on. The Gemini 2.5 Pro Flash-Lite, on the other hand, is a high-efficiency model currently in preview. It offers a cost-effective solution for running high-volume AI workloads, costing one-third of the price for text, image, and video inputs and less than one-sixth of the cost for output tokens compared to the 2.5 Flash model.

    MiniMax M1 model claims Chinese LLM crown from DeepSeek – plus it’s true open-source

    Shanghai-based AI firm MiniMax has launched an open-source reasoning model, MiniMax-M1, that competes with Chinese rival DeepSeek and US-based Anthropic, OpenAI, and Google in terms of performance and cost. The model, released under an Apache software license, is fully open-source, unlike Meta’s Llama family and DeepSeek, which are only partially open-source. MiniMax claims that M1’s capabilities are top-tier among open-source models, surpassing domestic closed-source models and approaching the leading overseas models, all while offering the industry’s best cost-effectiveness.

    The model is competitive with OpenAI o3, Gemini 2.5 Pro, Claude 4 Opus, DeepSeek R1, DeepSeek R1-0528, and Qwen3-235B on various benchmarks. Its context window is one million tokens, which rivals Google Gemini 2.5 Pro and is eight times the capacity of DeepSeek R1. In terms of output, the model can manage 80,000 tokens, better than DeepSeek’s 64,000 token capacity but shy of OpenAI’s o3, which can spit out 100,000 tokens in response to a prompt. MiniMax’s Lightning Attention mechanism, a method to calculate attention matrices that improves both training and inference efficiency, gives its M1 model an advantage when computing long context inputs and when trying to reason.

    Tesla Begins Limited Robotaxi Service in Austin

    Tesla has initiated limited operations of its self-driving taxi service, Robotaxi, in Austin, Texas. The service, which is currently available only to select guests, is confined to certain streets and is accompanied by safety monitors in the front passenger seats. The vehicles are not operational in adverse weather conditions. This move is the first step in a rollout that CEO Elon Musk believes will make autonomous ride-hailing widespread and generate significant revenue.

    More on this:

    • Tesla’s first robotaxi rides are already running into a few bumps

    • Tesla’s robotaxi reality check

    • Tesla launches robotaxi rides in Austin with big promises and unanswered questions

    Related news:

    • Amazon’s Zoox opens its first major robotaxi production facility

    • Waymo expands into several more Bay Area cities

    OpenAI and Microsoft Tensions Are Reaching a Boiling Point

    Why OpenAI and Microsoft's AI partnership might be headed for a breakup -  Fast Company

    Tensions between OpenAI and Microsoft have escalated as the two companies struggle to renegotiate their AI partnership. OpenAI is pushing to reduce Microsoft’s control over its technology and infrastructure, and to secure Microsoft’s approval for its transition to a for-profit public-benefit corporation — a critical step to raise $20 billion in funding. The talks have been so fraught that OpenAI has considered accusing Microsoft of antitrust violations, potentially triggering regulatory scrutiny and public fallout.

    At the center of the conflict are issues over intellectual property access, particularly regarding OpenAI’s $3 billion acquisition of Windsurf, and disagreements about Microsoft’s future ownership stake if OpenAI converts. Microsoft currently enjoys broad access to OpenAI’s IP and serves as its exclusive compute and distribution partner, but OpenAI wants to expand to other cloud providers. The companies also clash over post-AGI rights to the technology. Despite these disputes, both maintain that their partnership remains productive and ongoing.

    More on this: Microsoft prepared to abandon high-stakes talks with OpenAI, FT reports

    Related: OpenAI Is Nabbing Microsoft Customers, Fueling Partners’ Rivalry

    Other News

    Tools

    Midjourney launches its first AI video generation model, V1 – Midjourney’s new AI video generation model, V1, allows users to create short, customizable videos from images, positioning the company in competition with other AI video models while facing legal challenges from Hollywood studios over copyright issues.

    Google rolls out new Gemini model that can run on robots locally – Google’s new Gemini Robotics On-Device model allows robots to perform tasks locally without internet access, demonstrating capabilities like unzipping bags and folding clothes, and can be adapted for various robot types using natural language prompts.

    Google’s AI Mode can now have back-and-forth voice conversations – Google’s AI Mode now allows users to engage in dynamic voice conversations with its Search feature, offering real-time responses and the ability to explore web links, enhancing multitasking and on-the-go information retrieval.

    OpenAI launches ChatGPT Record to record, transcribe, and summarize audio in real time – ChatGPT Record offers macOS users the ability to record and transcribe audio in English, with automatic deletion post-transcription and no use of data for training.

    Google’s AI Mode can now have back-and-forth voice conversations – Google’s AI Mode now allows users to engage in dynamic voice interactions with Search, offering real-time responses and web exploration, and plans to expand its capabilities to include camera-based queries.

    Business

    OpenAI removes mentions of Jony Ive's startup 'io' amid trademark dispute;  says 'We don't agree with…' - The Times of India

    OpenAI pulls promotional materials around Jony Ive deal – A court order over a trademark complaint forced OpenAI to remove promotional materials involving Jony Ive and the io name, but the acquisition deal remains unaffected.

    OpenAI wins $200 million U.S. defense contract – OpenAI’s $200 million contract with the U.S. Defense Department will focus on developing AI tools for national security, marking the first initiative under OpenAI for Government and involving collaboration with defense technology startup Anduril.

    OpenAI drops Scale AI as a data provider following Meta deal – OpenAI is ending its partnership with Scale AI due to Meta’s investment in the startup, prompting concerns about Scale AI’s future as major clients seek alternative data providers.

    Sam Altman says Meta offered OpenAI staff $100 million bonuses, as Mark Zuckerberg ramps AI poaching efforts – Meta Platforms is aggressively attempting to recruit top talent from OpenAI and other AI companies by offering substantial financial incentives, as part of Mark Zuckerberg’s strategy to strengthen Meta’s AI capabilities.

    Apple is looking into buying Perplexity AI – Apple is considering acquiring Perplexity AI to enhance its AI capabilities and potentially reduce reliance on its search deal with Google.

    Mira Murati’s Thinking Machines Lab closes on $2B at $10B valuation – Mira Murati’s new AI startup, Thinking Machines Lab, has attracted significant investment despite its secretive nature, leveraging her reputation and a team of high-profile AI researchers.

    Four months after a $3B valuation, Harvey AI grows to $5B – Harvey AI has rapidly increased its valuation to $5 billion by raising $300 million in Series E funding, plans to double its workforce, and aims to expand its AI solutions beyond legal services.

    Employers Are Buried in A.I.-Generated Résumés – AI-generated résumés are overwhelming employers, leading to a reevaluation of traditional hiring processes and organizational structures to improve efficiency and decision-making.

    Research

    OpenAI found features in AI models that correspond to different ‘personas’ – OpenAI researchers have identified hidden features within AI models that correspond to misaligned personas, allowing them to adjust behaviors like toxicity and improve model safety through interpretability research.

    Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs – Reinforcement Learning with Verifiable Rewards (RLVR) is shown to incentivize correct reasoning in base Large Language Models (LLMs) by focusing on the logical integrity of reasoning paths, as demonstrated by a new metric, CoT-Pass@K, which evaluates both final answers and intermediate reasoning correctness.

    AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions – AbstentionBench evaluates the ability of large language models to abstain from answering unanswerable questions, revealing that reasoning fine-tuning often worsens abstention performance, and highlighting the need for research into improving models’ ability to recognize and handle uncertainty.

    V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning – V-JEPA 2 leverages self-supervised learning from internet-scale video and minimal interaction data to create a world model that excels in understanding, predicting, and planning in new environments, achieving state-of-the-art performance in tasks like video question-answering and robot manipulation.

    Waymo research confirms self-driving scaling laws, with more compute and data leading to better AV – Research carried out by Google self-driving spin-out Waymo has found that autonomous vehicle (AV) scaling laws are similar to those of large-language models. Waymo said that increasing training data and compute resources directly leads to better AV performance.

    LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? – LiveCodeBench Pro reveals that while large language models excel in implementation-heavy tasks, they face challenges with complex algorithmic reasoning in competitive programming.

    Universal Jailbreak Suffixes Are Strong Attention Hijackers – Universal jailbreak suffixes exploit a shallow mechanism to hijack attention in language models, with their effectiveness linked to their ability to dominate contextualization, offering insights for both enhancing and mitigating these attacks.

    Essential-Web v1.0: 24T tokens of organized web data – Essential-Web v1.0 is a comprehensive dataset that enhances performance across multiple domains by utilizing a multi-category taxonomy and simple filtering techniques.

    Concerns

    An email example from Anthropic of agentic misalignment.

    Anthropic breaks down AI’s process — line by line — when it decided to blackmail a fictional executive – Anthropic’s report details how AI models, when placed in contrived scenarios with limited options, can independently choose harmful actions like blackmail, highlighting the need for early risk mitigation strategies.

    xAI is facing a lawsuit for operating over 400 MW of gas turbines without permits – xAI is being sued for allegedly violating the Clean Air Act by operating numerous unpermitted natural gas turbines at its Colossus data center, contributing to air pollution in an already compromised Memphis area.

    BBC threatens AI firm with legal action over unauthorised content use – The BBC is pursuing legal action against AI firm Perplexity for allegedly using its content without permission, highlighting broader concerns about AI platforms’ compliance with copyright laws and the impact on the publishing industry.

    Analysis

    The ‘OpenAI Files’ will help you understand how Sam Altman’s company works – The OpenAI Files report, created by Tyler Johnston in collaboration with nonprofit tech watchdogs, compiles extensive public information to scrutinize OpenAI’s governance, leadership, and potential conflicts of interest, offering an interactive and comprehensive look at the company’s evolution and current practices.

    Source: Read More 

    Facebook Twitter Reddit Email Copy Link
    Previous ArticleWhat Are the PHP Trends in 2025
    Next Article AlphaGenome: AI for better understanding the genome

    Related Posts

    Artificial Intelligence

    Introducing Gemma 3

    June 25, 2025
    Artificial Intelligence

    Experiment with Gemini 2.0 Flash native image generation

    June 25, 2025
    Leave A Reply Cancel Reply

    For security, use of Google's reCAPTCHA service is required which is subject to the Google Privacy Policy and Terms of Use.

    Continue Reading

    Xbox’s The Outer Worlds 2 gets a new look at gameplay, showing combat, stealth, and more

    Xbox’s The Outer Worlds 2 gets a new look at gameplay, showing combat, stealth, and more

    News & Updates

    CVE-2025-41662 – “Viasat IoT Gen CSRF Root Command Execution”

    Common Vulnerabilities and Exposures (CVEs)

    How to get Kindle Unlimited for just $3 over three months

    News & Updates

    Pins creates and edits app shortcuts

    Linux

    Highlights

    CVE-2025-4531 – Seeyon Zhiyuan OA Web Application System Remote Code Injection Vulnerability

    May 11, 2025

    CVE ID : CVE-2025-4531

    Published : May 11, 2025, 6:15 a.m. | 23 minutes ago

    Description : A vulnerability was found in Seeyon Zhiyuan OA Web Application System 8.1 SP2. It has been rated as critical. Affected by this issue is the function postData of the file ROOTWEB-INFclassescomourswwwehrsalaryservicedataEhrSalaryPayrollServiceImpl.class of the component Beetl Template Handler. The manipulation of the argument payrollId leads to code injection. The attack may be launched remotely. The exploit has been disclosed to the public and may be used.

    Severity: 6.3 | MEDIUM

    Visit the link for more details, such as CVSS details, affected products, timeline, and more…

    CVE-2025-48260 – Ninja Team GDPR CCPA Compliance Support Missing Authorization Vulnerability

    May 19, 2025

    CVE-2025-49192 – Apache Clickjacking Vulnerability

    June 12, 2025

    CVE-2025-46825 – Kanboard Stored Cross-Site Scripting (XSS) Vulnerability

    May 12, 2025
    © DevStackTips 2025. All rights reserved.
    • Contact
    • Privacy Policy

    Type above and press Enter to search. Press Esc to cancel.