Alvaro Lopez Ortega / 2025-09-13 Briefing

Created Sat, 13 Sep 2025 22:01:29 +0000 Modified Tue, 14 Oct 2025 02:15:21 +0000
932 Words

OpenAI aims to boost its valuation by $50 billion through strategic revenue share reductions with Microsoft and partners, while a lawsuit against Google highlights ongoing copyright concerns over AI-generated content. Meanwhile, efforts to extend the lifespan of old Macs and advancements in AI model determinism underscore ongoing technological innovations today.

▶️ Open Source

OpenAI to Gain $50B by Cutting Revenue Shares with Microsoft and Partners

OpenAI plans to gain $50 billion by lowering revenue shares with Microsoft and partners, boosting its income through strategic revenue share reductions.

  • OpenAI will increase its revenue by approximately $50 billion by reducing its revenue share with Microsoft and partners
  • The change involves cutting the revenue share percentage, significantly boosting OpenAI’s income
  • The adjustment is part of a strategic shift to enhance OpenAI’s financial position and growth prospects

▶️ Management and Leadership

Companies, notably Instacart, are allegedly hiding U.S. job openings through deceptive practices, prompting legal actions and raising questions about immigration policy and fair employment.

  • Corporations, including Instacart, have engaged in practices to conceal job openings from U.S. citizens, such as unusual application instructions and limited advertising.
  • Companies like Meta paid $13 million in 2021 and Apple paid $25 million in 2023 to settle discrimination allegations related to job advertising and hiring practices.
  • Instacart delivered a cease-and-desist letter to Jobs.now, claiming that sharing its job openings constitutes misappropriation and seeking to suspend the group’s website.

xAI Cuts 500 Data Annotators to Expand AI Tutor Team Tenfold

xAI laid off around 500 data annotation workers, shifting focus to expanding specialist AI tutors by 10x, with tests assessing skills in STEM, finance, medicine, safety, and behavior modeling.

  • xAI laid off approximately 500 data annotation workers, about one-third of its team, on September 12, 2025
  • The layoffs follow a strategic shift to prioritize specialist AI tutors over generalist roles, with plans to “surge” the team by 10x
  • The reorganization involved testing workers’ strengths in domains like STEM, finance, medicine, safety, and personality modeling to determine future roles

Emily Fontaine Uses Five Pillars to Evaluate Startups for IBM’s AI Investments

Emily Fontaine assesses startups based on five pillars—strategic fit, product quality, market potential, competition, and financial discipline—to align with IBM’s AI and hybrid cloud strategy.

  • Emily Fontaine, IBM’s global head of venture capital, uses five “pillars” to evaluate startup investments.
  • Criteria include strategic fit, technology quality, market opportunity, competitive positioning, and financial discipline.
  • Fontaine aims to connect with 800 startups worldwide in 2025 and oversees IBM’s $500 million Enterprise AI fund.

Community Tools Extend Life of Old Macs with Newer macOS and Browsers

Community tools enable installing newer macOS versions on unsupported Macs and running recent browsers on old hardware, extending usability of 10-15-year-old Apple devices.

  • Apple’s macOS 26 “Tahoe” launches soon, supporting only 2019-2020 Intel Macs; older models require tools like OpenCore Legacy Patcher.
  • macOS 10.15 “Catalina” can be unofficially installed on unsupported Macs, such as a 2010 MacBook Pro, using dosdude1’s Catalina Patcher.
  • Upgrading a 2010 MacBook Pro to macOS 10.15 enabled modern browsers like Firefox 142 and Chromium Legacy, improving web compatibility on 15-year-old hardware.
  • The latest OCLP version 2.4.1 supports macOS 11 “Big Sur” and requires at least 4 GB RAM; unsupported models can run newer macOS versions via patchers.
  • Running unsupported macOS versions may cause sluggishness and compatibility issues with legacy apps like 32-bit Office 2011; newer open-source options like LibreOffice are available.

▶️ Technology

Constraint Solvers Simplify Hard LeetCode Problems

Many hard LeetCode problems are effectively constraint problems solvable with tools like MiniZinc, simplifying implementation and handling complex constraints beyond traditional algorithmic approaches.

  • Many complex LeetCode problems can be reformulated as constraint satisfaction problems solvable with tools like MiniZinc, Z3, or OR-Tools
  • Examples include coin change, stock profit maximization, three-number sum to zero, and largest rectangle in histogram
  • Constraint solvers handle these problems efficiently, offering better flexibility for complex or evolving constraints compared to bespoke algorithms

Understanding UTF-8: Efficient Unicode Encoding with ASCII Compatibility

UTF-8 encodes Unicode characters with 1-4 bytes, maintaining backward compatibility with ASCII by using specific leading bit patterns, enabling representation of millions of characters efficiently.

  • UTF-8 encodes Unicode characters using 1 to 4 bytes, with the first 128 characters (U+0000 to U+007F) encoded as single-byte ASCII, ensuring backward compatibility.
  • The encoding pattern uses specific leading bits in the first byte (0xxxxxxx, 110xxxxx, 1110xxxx, 11110xxx) to determine total bytes; continuation bytes start with 10.
  • UTF-8 files containing only ASCII characters are valid ASCII files, exemplified by the text “Hey Buddy” with 9 bytes, while files with non-ASCII characters, like “Hey👋 Buddy,” use up to 13 bytes.

Thinking Machines Lab Aims to Boost AI Determinism by Controlling GPU Kernel Stitching

Thinking Machines Lab, backed by $2 billion seed funding, seeks to make AI models more deterministic by controlling GPU kernel stitching, improving response consistency and RL training.

  • Thinking Machines Lab, founded by Mira Murati with $2 billion seed funding, aims to improve AI model reproducibility
  • Published “Defeating Nondeterminism in LLM Inference” blog post identifying GPU kernel stitching as root cause of response randomness
  • Controlling GPU kernel orchestration could enhance response determinism, aiding enterprise reliability and reinforcement learning training

Rolling Stone sues Google over AI-generated article summaries

Rolling Stone publisher filed a lawsuit against Google on August 15, 2023, alleging copyright infringement and misappropriation of proprietary content through AI-generated article summaries.

  • Rolling Stone publisher sued Google over AI-generated article summaries on August 15, 2023
  • Allegation claims Google’s AI summaries infringe on copyright and misappropriate proprietary content
  • Lawsuit seeks damages and injunctive relief to prevent further use of Rolling Stone’s copyrighted material