<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Microsoft AI — FoundationIndex</title>
    <link>https://foundationindex.ustunozgur.com/s/microsoft-ai</link>
    <description>Microsoft&apos;s AI research and product organization—spanning foundational Microsoft Research breakthroughs (ResNet, DeepSpeed, LoRA, and the open-weight Phi small language models), Azure AI supercomputing and custom Maia silicon, and the first-party MAI frontier model family powering Copilot and GitHub Copilot.</description>
    <language>en</language>
    <atom:link href="https://foundationindex.ustunozgur.com/s/microsoft-ai/feed.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Microsoft AI · Release · 2026-11-17</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-ignite-2026-mai-code-2-licensed-weights</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-ignite-2026-mai-code-2-licensed-weights</guid>
      <pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;At Microsoft Ignite 2026, Microsoft AI introduced MAI Code 2 and MAI Code 2 Flash for GitHub Copilot and Microsoft Foundry, MAI Cyber 2 Flash (1M-token cybersecurity defense), MAI Realtime 1 (native speech-to-speech at $0.01/minute), MAI Image 2.7, MAI Transcribe 2, and MAI Licensed Weights enabling enterprises to customize and self-host frontier MAI checkpoints inside their own Azure environments.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/&quot;&gt;Introducing MAI Code 2 and MAI Code 2 Flash&lt;/a&gt;, &lt;a href=&quot;https://microsoft.ai/news/introducing-mai-licensed-weights/&quot;&gt;Introducing MAI Licensed Weights&lt;/a&gt;, &lt;a href=&quot;https://microsoft.ai/news/microsoft-ignite-2026/&quot;&gt;Microsoft Ignite 2026&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Release · 2026-06-02</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-build-2026-mai-thinking-1</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-build-2026-mai-thinking-1</guid>
      <pubDate>Tue, 02 Jun 2026 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;At Microsoft Build 2026, Microsoft AI launched seven first-party MAI models led by MAI-Thinking-1 (a ~1T-total / 35B-active sparse MoE reasoning model trained from scratch without third-party distillation), alongside MAI-Code-1 Flash, MAI-Cyber-1 Flash, MAI-Image-2, MAI-Voice-2, and MAI-Transcribe-1 in Microsoft Foundry.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://microsoft.ai/news/building-a-hillclimbing-machine-launching-seven-new-mai-models/&quot;&gt;Building a Hill-Climbing Machine: Launching Seven New MAI Models&lt;/a&gt;, &lt;a href=&quot;https://microsoft.ai/news/introducing-mai-thinking-1/&quot;&gt;Introducing MAI-Thinking-1&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Milestone · 2025-11-06</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-humanist-superintelligence-2025</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-humanist-superintelligence-2025</guid>
      <pubDate>Thu, 06 Nov 2025 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft AI launched the MAI Superintelligence Team under Mustafa Suleyman and Karén Simonyan and published its Towards Humanist Superintelligence charter prioritizing controllable, domain-specific AI across software engineering, medical diagnostics, and clean energy—followed on November 18 by a $30 billion Azure compute partnership with Anthropic and NVIDIA.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://microsoft.ai/news/towards-humanist-superintelligence/&quot;&gt;Towards Humanist Superintelligence&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Release · 2025-08-28</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-mai-1-preview-voice-1-2025</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-mai-1-preview-voice-1-2025</guid>
      <pubDate>Thu, 28 Aug 2025 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft AI unveiled its first two in-house foundation models trained independently of OpenAI: MAI-Voice-1 for low-latency expressive speech generation and MAI-1-preview, a sparse Mixture-of-Experts model trained across ~15,000 NVIDIA H100 GPUs.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://microsoft.ai/news/two-new-in-house-models/&quot;&gt;Two new in-house models in service of people&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Release · 2024-12-12</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-phi-3-phi-4-2024</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-phi-3-phi-4-2024</guid>
      <pubDate>Thu, 12 Dec 2024 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Following the April 2024 rollout of the mobile-ready Phi-3 series (3.8B–14B), Microsoft Research released Phi-4 (14B) under an MIT license, combining multi-agent Synthetic Data &amp;amp; Self-Play curricula with pivotal-token DPO to surpass much larger models on mathematical and STEM reasoning.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/2412.08905&quot;&gt;Phi-4 Technical Report&lt;/a&gt;, &lt;a href=&quot;https://arxiv.org/abs/2404.14219&quot;&gt;Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Milestone · 2024-03-19</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-mustafa-suleyman-mai-2024</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-mustafa-suleyman-mai-2024</guid>
      <pubDate>Tue, 19 Mar 2024 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft established the Microsoft AI (MAI) organization and appointed DeepMind co-founder and Inflection AI CEO Mustafa Suleyman as EVP and CEO (with Karén Simonyan as Chief Scientist), unifying Copilot product engineering with a mandate to train proprietary first-party frontier models.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://blogs.microsoft.com/blog/2024/03/19/mustafa-suleyman-deepmind-and-inflection-co-founder-joins-microsoft-to-lead-copilot/&quot;&gt;Mustafa Suleyman, DeepMind and Inflection Co-founder, joins Microsoft to lead Copilot&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Release · 2023-11-15</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-maia-100-cobalt-2023</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-maia-100-cobalt-2023</guid>
      <pubDate>Wed, 15 Nov 2023 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;At Microsoft Ignite 2023, Microsoft unveiled its first custom in-house AI accelerator silicon, Azure Maia 100 (a 5nm tensor chip paired with rack-level liquid cooling and 1.8 TB/s HBM memory for cloud training and inference), alongside the 128-core ARM-based Azure Cobalt 100 CPU.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://news.microsoft.com/source/features/ai/in-house-chips-silicon-maia-cobalt/&quot;&gt;With a systems approach to chips, Microsoft aims to tailor everything &amp;apos;from silicon to service&amp;apos; to meet AI demand&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Paper · 2023-06-20</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-phi-1-textbooks-2023</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-phi-1-textbooks-2023</guid>
      <pubDate>Tue, 20 Jun 2023 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft Research published Textbooks Are All You Need, introducing Phi-1 (1.3B) and demonstrating that curating &amp;quot;textbook-quality&amp;quot; Synthetic Data &amp;amp; Self-Play allows small language models to match or beat models 10x larger on Python coding benchmarks.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/2306.11644&quot;&gt;Textbooks Are All You Need&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Paper · 2021-06-17</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-lora-paper-2021</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-lora-paper-2021</guid>
      <pubDate>Thu, 17 Jun 2021 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft researchers (Edward Hu et al.) introduced LoRA (Low-Rank Adaptation), freezing pretrained model weights and injecting trainable low-rank matrices into Transformer layers—cutting fine-tuning VRAM up to 3x with zero inference latency penalty.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/2106.09685&quot;&gt;LoRA: Low-Rank Adaptation of Large Language Models&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Release · 2020-02-10</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-deepspeed-zero-2020</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-deepspeed-zero-2020</guid>
      <pubDate>Mon, 10 Feb 2020 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft open-sourced DeepSpeed built around the ZeRO (Zero Redundancy Optimizer) paper, partitioning optimizer states, gradients, and parameters across data-parallel GPUs to eliminate memory redundancy in multi-hundred-billion-parameter Distributed Training.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/1910.02054&quot;&gt;ZeRO: Memory Optimizations Toward Training Trillion Parameter Models&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Ecosystem · 2019-07-22</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-openai-1b-azure-2019</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-openai-1b-azure-2019</guid>
      <pubDate>Mon, 22 Jul 2019 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft invested $1 billion in OpenAI (later expanded by ~$10 billion in January 2023) to co-design large-scale Azure AI supercomputing clusters and integrate frontier GPT models across GitHub Copilot, Bing, Microsoft 365, and Azure OpenAI Service.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://openai.com/index/microsoft-invests-in-and-partners-with-openai/&quot;&gt;Microsoft invests in and partners with OpenAI to support us building beneficial AGI&lt;/a&gt;&lt;/p&gt;</description>
    </item>
    <item>
      <title>Microsoft AI · Paper · 2015-12-10</title>
      <link>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-resnet-2015</link>
      <guid>https://foundationindex.ustunozgur.com/s/microsoft-ai#microsoft-ai-resnet-2015</guid>
      <pubDate>Thu, 10 Dec 2015 00:00:00 GMT</pubDate>
      <description>&lt;p&gt;Microsoft Research (Kaiming He et al.) published Deep Residual Learning for Image Recognition (ResNet), introducing residual skip connections ($y = F(x) + x$) that allowed gradient flow across 152-layer CNNs and later became an essential structural primitive inside every Transformer layer.&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/1512.03385&quot;&gt;Deep Residual Learning for Image Recognition&lt;/a&gt;&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
