aidigester

Claude Code Major Outage: Developers Forced into ‘Coffee Break

by

Claude Code Outage: 3 Key Points

  • API error rates surged across all of Anthropic’s Claude models
  • Claude Code users halted work due to 500 errors
  • Microsoft AI team also uses this service — impacting the entire industry

What Happened?

Claude Code experienced a major outage. Developers encountered 500 errors when accessing the service, and Anthropic officially announced an increase in API error rates “across all Claude models.”[The Verge]

Anthropic stated that they identified the issue and are working on a fix. The current status page indicates that the outage has been resolved.[Anthropic Status]

Why is it Important?

Claude Code is not just an AI tool. It has become a core infrastructure that many developers, including the Microsoft AI team, rely on in their daily work.

Frankly, such outages are rare. According to the Anthropic status page, Claude Code’s 90-day uptime is 99.69%. But the problem is that even less than 1% downtime has a significant impact on developer productivity.

Personally, I see this incident as a warning about the dependence on AI coding tools. If you put all your workflows on a single service, you have no alternative when an outage occurs.

Recent Anthropic Service Issues

It is also worth noting that this outage is not an isolated incident:

  • Yesterday (February 2nd): Errors occurred in Claude Opus 4.5
  • Earlier this week: Fixed issues with AI credit system purchases
  • January 31st: Claude Code 2.1.27 memory leak — patched to 2.1.29

The continuous occurrence of multiple issues in a short period is a disappointing aspect in terms of service stability.

What Happens Next?

It’s a good sign that Anthropic is responding quickly. However, it’s time for developers to consider a backup plan.

Alternatives to Claude Code include tools like Goose (free) and pi-mono (open source). They are not complete replacements, but they can help maintain minimal work continuity in the event of an outage.

Frequently Asked Questions (FAQ)

Q: How often does Claude Code outage occur?

A: According to Anthropic’s official data, the 90-day uptime is 99.69%. Major outages like this are rare, but there have been several minor issues in recent weeks. It’s not something to completely ignore.

Q: What are the alternatives in case of an outage?

A: Goose is a free AI coding agent, and pi-mono is an open-source alternative with 5.9k stars on GitHub. Neither covers all the features of Claude Code, but they are options to continue working in an emergency.

Q: Does Anthropic provide compensation?

A: To date, Anthropic has not announced a separate compensation policy for outages. For paid users, the fact that there are no charges during the outage time due to usage-based billing is a de facto compensation.

If you found this article helpful, please subscribe to AI Digester.

References

AWS SageMaker Data Agent: Weeks of Medical Data Analysis → Days

by

Healthcare Data Analysis, Weeks Reduced to Days

  • AWS SageMaker Data Agent: AI agent that analyzes healthcare data in natural language
  • Cohort comparison and survival analysis can be performed without code
  • Released in November 2025, free to use in SageMaker Unified Studio

What Happened?

AWS has unveiled SageMaker Data Agent, an AI agent for healthcare data analysis. When epidemiologists or clinical researchers ask questions in natural language, the AI automatically generates and executes SQL and Python code.[AWS]

Previously, healthcare data analysis required navigating multiple systems, waiting for data access permissions, understanding schemas, and writing code directly. This process took weeks. SageMaker Data Agent reduces this to days, or even hours.[AWS]

Why is it Important?

Frankly, healthcare data analysis has always been a bottleneck. Epidemiologists spent 80% of their time on data preparation and only 20% on actual analysis. The reality was that they could only conduct 2-3 studies per quarter.

SageMaker Data Agent reverses this ratio. It significantly reduces data preparation time, allowing for more focus on actual clinical analysis. Personally, I believe this will directly impact the speed of discovering patient treatment patterns.

It’s particularly impressive that complex tasks like cohort comparison and Kaplan-Meier survival analysis can be requested in natural language. Saying something like, “Perform survival analysis for male vs. female patients with viral sinusitis,” and the AI automatically plans, writes, and executes the code.[AWS]

How Does it Work?

SageMaker Data Agent operates in two modes. First, code can be generated directly from inline prompts in notebook cells. Second, the Data Agent panel can break down complex analysis tasks into structured steps for processing.[AWS]

The agent understands the current notebook state and generates contextually relevant code by understanding the data catalog and business metadata. It doesn’t just spit out code snippets, but creates an entire analysis plan.[AWS]

What Happens Next?

According to a Deloitte survey, 92% of healthcare executives are investing in or experimenting with generative AI.[AWS] The demand for healthcare AI analysis tools will continue to increase.

If agentic AI like SageMaker Data Agent accelerates healthcare research, it could positively impact new drug development and the discovery of treatment patterns. However, one concern is data quality. No matter how fast the AI is, if the input data is messy, the results will be messy too.

Frequently Asked Questions (FAQ)

Q: What is the cost of SageMaker Data Agent?

A: SageMaker Unified Studio itself is free. However, you are charged for the actual computing resources used (EMR, Athena, Redshift, etc.). The notebook has a free tier of 250 hours for the first two months, so you can test it out lightly.

Q: What data sources are supported?

A: It connects to AWS Glue Data Catalog, Amazon S3, Amazon Redshift, and various other data sources. If you have an existing AWS data infrastructure, you can integrate it immediately. It is also compatible with healthcare data standards such as FHIR and OMOP CDM.

Q: Which regions are available?

A: It is available in all AWS regions where SageMaker Unified Studio is supported. It is best to check the AWS official documentation for Seoul region support.

If you found this article useful, please subscribe to AI Digester.

References

Lotus Health AI Raises $35 Million in Funding as a Free AI Primary Care Physician

by

Free AI Primary Care Physician Receives $35 Million Investment

  • Lotus Health AI secures $35 million in Series A funding from CRV and Kleiner Perkins
  • Provides 24/7 free primary care services in 50 languages, operating in all 50 US states
  • In an era where 230 million people ask ChatGPT health questions weekly, the AI healthcare market enters full-scale competition

What Happened?

Lotus Health AI received $35 million in a Series A round co-led by CRV and Kleiner Perkins.[TechCrunch] This startup utilizes large language models (LLMs) to provide 24/7 free primary care services in 50 languages.

Founder KJ Dhaliwal previously sold the South Asian dating app Dil Mil for $50 million in 2019.[Crunchbase] Inspired by his childhood experience of interpreting for his parents in medical settings, he launched Lotus Health AI in May 2024 with the goal of addressing inefficiencies in the US healthcare system.

Why is it Important?

Frankly, the size of this investment is notable. The average investment for AI healthcare startups is $34.4 million, and Lotus Health AI matched this level in its Series A.[Crunchbase]

Understanding the background helps. According to OpenAI, 230 million people ask ChatGPT health-related questions every week.[TechCrunch] This means people are already seeking health advice from AI. However, ChatGPT cannot provide medical treatment. Lotus Health AI is targeting this niche.

Personally, the \”free\” model is the most interesting. Considering how expensive healthcare is in the US, free primary care is a pretty disruptive value proposition. Of course, the revenue model is still unclear.

What Happens Next?

The AI healthcare market is expected to enter full-scale competition. OpenAI also entered this market last January with the launch of ChatGPT Health. It provides personalized health advice by integrating with Apple Health, MyFitnessPal, and more.[OpenAI]

Regulatory risks remain. Even OpenAI states in its terms of service, \”Do not use for diagnostic or treatment purposes.\” Several lawsuits have already been filed due to damages caused by AI medical advice. It is necessary to observe how Lotus Health AI will manage these risks.

Frequently Asked Questions (FAQ)

Q: Is Lotus Health AI really free?

A: It is free for patients. However, the specific revenue model has not yet been disclosed. There are various possibilities, such as a B2B model targeting insurance companies or employers, or adding premium services. It seems they are aiming for economies of scale by providing services in all 50 states.

Q: What is the difference between it and a general AI chatbot?

A: Lotus Health AI is a medical service specialized in primary care. Unlike general chatbots, it holds medical service licenses in all 50 US states. The key difference is that it can perform actual medical treatment, not just provide health information.

Q: Does it support Korean?

A: It stated that it supports 50 languages, but the specific language list has not been disclosed. It is necessary to confirm whether Korean is supported. Currently, the service is only available in the US, and there are no plans for overseas expansion announced yet.

If this article was helpful, please subscribe to AI Digester.

References

Apple Xcode 26.3: Simultaneously Equipped with Anthropic Claude Agent + OpenAI Codex

by

AI Coding Agents in a Two-Engine System, Landing Simultaneously on Xcode

  • Anthropic Claude Agent and OpenAI Codex are now directly available within Xcode.
  • Third-party agents can also be connected with Model Context Protocol support.
  • Release Candidate (RC) available to developer program members starting today.

What Happened?

Apple announced official support for agentic coding in Xcode 26.3. [Apple Newsroom] Anthropic’s Claude Agent and OpenAI’s Codex can be used directly within the IDE.

Agentic coding goes beyond AI simply suggesting code snippets; it involves analyzing project structure, breaking down tasks independently, and autonomously running build-test-fix cycles. In simple terms, AI works like a junior developer.

Susan Prescott, Apple’s Vice President of Worldwide Developer Relations, stated that “Agentic coding maximizes productivity and creativity, allowing developers to focus on innovation.”[Apple Newsroom]

Why is it Important?

Personally, I think this is a pretty big change. For two reasons.

First, Apple has officially entered the AI coding tool competition. Independent tools like Cursor, GitHub Copilot, and Claude Code have been growing the market, but now the platform owner is directly involved.

Second, it embraces both Anthropic and OpenAI. Typically, Big Tech companies form exclusive partnerships with one AI vendor. But Apple is playing both sides. While claiming to give developers a choice, it seems like they’re hedging their bets because they don’t know which model will win.

The support for Model Context Protocol (MCP) is also noteworthy. This is an open standard for connecting AI agents with external tools, led by Anthropic.[TechCrunch] Apple’s adoption of this signals a departure from its closed ecosystem strategy.

What Will Happen in the Future?

Over a million iOS/macOS developers use Xcode. If they become accustomed to agentic coding, the development paradigm itself could change.

However, there are also concerns. If AI autonomously modifies code, security vulnerabilities or unexpected bugs could arise. We need to see how Apple manages this aspect.

The competitive landscape is also interesting. OpenAI independently released the Codex app for macOS a day earlier.[TechCrunch] The timing is curious, with the integration announcement with Apple coming the very next day.

Frequently Asked Questions (FAQ)

Q: When will Xcode 26.3 be officially released?

A: The Release Candidate (RC) version is currently available to Apple Developer Program members. The official version will be distributed through the App Store soon. The exact date has not yet been announced.

Q: Which should I use, Claude Agent or Codex?

A: It depends on the nature of the project. Claude excels at understanding long code and ensuring safety, while Codex specializes in rapid code generation. Try both and choose the one that suits you best. That’s why Apple gives you the choice.

Q: Can existing Xcode 26 users upgrade?

A: Yes. This is an extension of the Swift coding assistant feature introduced in Xcode 26, so existing users can immediately use the agentic coding feature by updating to 26.3.

If you found this article helpful, please subscribe to AI Digester.

References

Apple Xcode 26.3: Simultaneously Loading Anthropic Claude Agent + OpenAI Codex

by

AI 2-Layer Coding Agent System Lands on Xcode Simultaneously

  • Anthropic Claude Agent and OpenAI Codex are now available directly within Xcode
  • Third-party agents will also connect with Model Context Protocol support
  • Release Candidate (RC) available to Developer Program members starting today

Apple has officially announced.

Agentic coding support in Xcode 26.3. [Apple Newsroom] Anthropic’s Claude Agent and OpenAI’s Codex are available directly within the IDE.

Agentic coding means AI doesn’t just write code. Beyond the suggested level, it analyzes the project structure, independently divides tasks, and autonomously executes build, test, and modification cycles. Simply put, AI acts like a junior developer.

“Agentic coding maximizes productivity and creativity, allowing developers to focus on innovation,” said Susan Prescott, Apple’s Vice President of Worldwide Developer Relations. [Apple Newsroom]

Why is it important?

Personally, I think this is a pretty big change. There are two reasons:

First, Apple has officially entered the AI coding tool competition. While independent tools like Cursor, GitHub Copilot, and Claude Code have been growing the market, now the platform owner is entering directly.

Second, they are adopting Anthropic and OpenAI simultaneously. Usually, large tech companies form exclusive partnerships with one AI company. But Apple crossed the line. The cause is to give developers a choice, but honestly, it feels like insurance because I don’t know which model will be the winner.

Also noteworthy is the support for the Model Context Protocol (MCP). This is an open standard for connecting AI agents and external tools, with Anthropic leading the way. [TechCrunch] Apple’s adoption of this is a step away from its closed ecosystem strategy. This is a sign that they have resigned.

What will happen in the future?

More than 1 million iOS/macOS developers use Xcode. As they become familiar with agentic coding, the development paradigm itself may change.

However, there are also concerns. If AI autonomously modifies code, security vulnerabilities or unexpected bugs may occur. It is not yet known how Apple will manage this part.

The competitive landscape is also interesting. OpenAI independently released the Codex app for macOS the day before. [TechCrunch] The timing is strange because the integration with Apple was announced the next day.

Frequently Asked Questions (FAQ)

Q: When will Xcode 26.3 be officially released?

A: Release Candidate (RC) version is now available on Apple Developer Program members. The full version will be distributed through the App Store soon. The exact date has not yet been announced.

Q: Which should I use, Claude Agent or Codex?

A: It depends on the nature of the project. Claude excels at understanding long code and safety, while Codex specializes in rapid code generation. Try both and choose the one that suits you. That’s why Apple gave us a choice.

Q: Can existing Xcode 26 users also upgrade?

A: Yes. This is an extension of the Swift coding assistant features introduced in Xcode.

  • Xcode 26.3 unlocks the power of agentic coding – Apple Newsroom (2026-02-03)
  • Agentic coding comes to Apple’s Xcode 26.3 with agents from Anthropic and OpenAI – TechCrunch (2026-02-03)
  • OpenAI launches new macOS app for agentic coding – TechCrunch (2026-02-02)

    • Microsoft Building AI Content Licensing ‘App Store’: Announces Publisher Content Marketplace

    by

    MS, Building an AI Content Licensing Marketplace: 3 Key Points

    • Microsoft is building the Publisher Content Marketplace (PCM) – a platform where AI companies can search for and contract content licensing terms.
    • Co-designed with major media outlets such as Vox Media, AP, Conde Nast, and Hearst.
    • Usage-based compensation model benefits both publishers and AI companies.

    What Happened?

    Microsoft is creating an app store-like platform for AI content licensing. This platform, called the Publisher Content Marketplace (PCM), allows AI companies to directly search for licensing terms for premium content, and publishers can receive reports on how their content is being used.[The Verge]

    Microsoft co-designed PCM with major publishers such as Vox Media (The Verge’s parent company), AP, Conde Nast, People, Business Insider, Hearst, and USA TODAY. Yahoo is onboarding as the first demand partner.[Search Engine Land]

    Why is it Important?

    Frankly, the issue of unauthorized use of content in the AI industry has already reached a breaking point. NYT, The Intercept, and others are pursuing copyright lawsuits against Microsoft and OpenAI. The problem has become too large to be solved by individual contracts.

    The interesting thing about PCM is that it’s a two-sided marketplace. Publishers set licensing terms, and AI companies can compare and contract terms as if they were shopping. Personally, I think this is one of the realistic solutions to the AI learning data problem.

    It’s also significant that Microsoft is moving first in this market. From the publisher’s perspective, MS has consistently delivered the message that “fair compensation must be paid for content quality.”[Digiday]

    What Happens Next?

    Microsoft is currently expanding its partners in the pilot phase. Simply put, it’s a platform that could become the standard for content licensing in the AI era.

    But one question remains. It is still unclear how PCM will work with the Really Simple Licensing (RSL) open standard that publishers are pushing for on their own. Microsoft has not commented on this.

    In conclusion, this is the first sign that AI content licensing is shifting from individual negotiations to platform-based transactions. It is necessary to watch how Google and OpenAI will respond.

    Frequently Asked Questions (FAQ)

    Q: Can anyone participate in PCM?

    A: According to Microsoft, it supports publishers of all sizes, from large media companies to small specialized media outlets. However, it is currently in the pilot phase and is being tested with invited publishers. The timing of general participation has not yet been announced.

    Q: How do publishers earn revenue?

    A: It is a usage-based compensation model. Every time an AI product uses publisher content for grounding (reference), it is measured, and compensation is paid accordingly. Publishers can check reports on where and how much value their content has created.

    Q: What is different from existing AI licensing agreements?

    A: In the past, publishers and AI companies had to negotiate individually on a 1:1 basis. PCM is in the form of a marketplace, so multiple AI companies can compare and select the terms of multiple publishers on one platform. This greatly reduces negotiation costs and time.

    If you found this article useful, please subscribe to AI Digester.

    Reference Materials

    Intel’s Full-Scale Entry into the GPU Market: Shaking Nvidia’s Monopoly?

    by

    Intel CEO, Officially Announces Entry into the GPU Market — 3 Key Points

    • Lip-Bu Tan CEO announces the full-scale launch of the GPU business at the Cisco AI Summit
    • Recruitment of a new GPU Chief Architect — Crescent Island for data centers to be sampled in the second half of 2026
    • Intel challenges Nvidia’s exclusive market as the third player

    What happened?

    Intel CEO Lip-Bu Tan officially announced the company’s entry into the GPU market at the Cisco AI Summit held in San Francisco on February 3rd.[TechCrunch] This is a market currently dominated by Nvidia.

    Tan revealed that they have recruited a new GPU Chief Architect. He didn’t disclose the name but mentioned that it took quite an effort to persuade him.[CNBC]

    Intel is already preparing a GPU codenamed Crescent Island for data centers. Based on the Xe3P microarchitecture and equipped with 160GB of LPDDR5X memory, customer sampling is scheduled for the second half of 2026.[Intel Newsroom]

    Why is this important?

    Honestly, I was a bit surprised. I didn’t expect Intel to fully enter the GPU market.

    Currently, the GPU market is dominated by Nvidia. Their market share in the AI learning GPU market exceeds 80%. AMD is challenging with the MI350, but it is still difficult to overcome Nvidia’s CUDA ecosystem.

    Intel’s entry provides a third option in the market. In particular, Crescent Island targets the AI inference market. Not learning, but inference. This is important.

    This is because the AI inference market is growing faster than the learning market. The demand for agent AI and real-time inference is exploding. Intel CTO Sachin Katti also emphasized this point.[Intel Newsroom]

    Personally, I think Intel’s timing is not bad. Nvidia GPU prices are so high that many companies are looking for alternatives. Intel’s pursuit of a cost-effectiveness strategy with Gaudi is also in this context.

    What will happen in the future?

    Once Crescent Island sampling begins in the second half of 2026, we will be able to see its actual performance. Intel is also planning 14A node risk production by 2028.

    But there is a problem. As Tan himself admitted, memory is hindering AI growth. Memory bottlenecks are as serious as GPU performance. Cooling is also an issue. Tan said that air cooling has reached its limit and water cooling solutions are needed.[Capacity]

    It is uncertain whether Intel will be able to break down Nvidia’s stronghold. But at least the emergence of competition is good news for consumers.

    Frequently Asked Questions (FAQ)

    Q: When will Intel’s new GPU be released?

    A: Customer sampling of the Crescent Island GPU for data centers is scheduled for the second half of 2026. The official release date has not yet been announced. For consumer GPUs, there is a separate Arc series lineup, and products based on the current Xe2 architecture are being sold.

    Q: What are the strengths of Intel GPUs compared to Nvidia?

    A: Intel emphasizes price competitiveness. While the Nvidia H100 consumes 700 watts per unit and is expensive, Intel Gaudi and Crescent Island emphasize power efficiency relative to performance. In addition, Intel’s ability to provide CPU-GPU integrated solutions is also a differentiating factor.

    Q: Will consumer gaming GPUs be affected?

    A: There is little direct correlation. This announcement targets the data center AI inference market. However, the Intel Arc series is growing in the gaming market, exceeding 1% market share, and the 12GB VRAM configuration of the B580 is attracting attention in the cost-effective market.

    If you found this article helpful, please subscribe to AI Digester.

    Reference Materials

    MIT Kitchen Cosmo: AI Creates Recipes from Your Fridge Ingredients

    by

    3 Key Points

    • MIT develops AI recipe-generating kitchen appliance ‘Kitchen Cosmo’
    • Recognizes ingredients with a camera and prints user-customized recipes with a printer
    • Presents the concept of ‘Large Language Objects’ that extends LLMs to the physical world

    What Happened?

    MIT architecture students have developed an AI-based kitchen appliance called ‘Kitchen Cosmo’.[MIT News] The device, about 45cm (18 inches) tall, recognizes ingredients with a webcam, receives user input via a dial, and prints recipes with a built-in thermal transfer printer.

    The project was conducted at the Design Intelligence Lab led by MIT professor Marcelo Coelho. Architecture graduate student Jacob Payne and fourth-year design major Ayah Mahmoud participated.[MIT News]

    Why is it Important?

    Frankly, the interesting thing about this project is its philosophy rather than the technology itself. Professor Coelho calls it ‘Large Language Objects (LLOs)’. It’s the concept of taking LLMs out of the screen and turning them into physical objects.

    Professor Coelho said, “This new form of intelligence is powerful, but it is still ignorant of the world outside of language.”[MIT News] Simply put, ChatGPT knows text well but doesn’t know what’s in the refrigerator. Kitchen Cosmo bridges that gap.

    Personally, I think this shows the future of AI interfaces. Instead of touching the screen and typing, you show objects and turn a dial. This is especially useful in situations where your hands are busy, like cooking.

    What Happens Next?

    The research team plans to add real-time cooking tips and role-sharing features for multiple people cooking together in the next version.[MIT News] Student Jacob Payne said, “When you’re wondering what to make with leftover ingredients, AI can find creative ways to use them.”

    It is uncertain whether this research will lead to a commercial product. However, attempts to extend LLMs to physical interfaces will increase in the future.

    Frequently Asked Questions (FAQ)

    Q: What ingredients can Kitchen Cosmo recognize?

    A: It uses a Vision Language Model to recognize ingredients photographed with a camera. It can identify common food ingredients such as fruits, vegetables, and meats, and generates recipes considering basic seasonings and condiments at home. However, the specific recognition accuracy has not been disclosed.

    Q: What factors are reflected in recipe generation?

    A: You can enter meal type, cooking skill, available time, mood, dietary restrictions, and number of people. You can also select a taste profile and regional cuisine style (e.g., Korean, Italian). It combines all these conditions to create a customized recipe.

    Q: Can the general public purchase it?

    A: Currently, it is in the prototype stage at the MIT lab, and there are no plans for commercial release. Since it started as an academic research project, it will take time to commercialize. However, there is a possibility that similar concept products will be released by other companies.

    If you found this article useful, please subscribe to AI Digester.

    References

    Jensen Huang: “Everything Will Be Expressed as a Virtual Twin” — NVIDIA-Dassault’s Biggest Partnership in 25 Years of Collaboration

    by

    Jensen Huang: “Everything Will Be Represented in a Virtual Twin” — NVIDIA-Dassault’s Largest Partnership in 25 Years of Collaboration

    • NVIDIA and Dassault Systèmes announce the largest strategic partnership in their 25-year history of collaboration.
    • Aim to expand design and manufacturing processes by 100-1000 times with physics-based AI and Virtual Twin.
    • AI factories to be deployed across 3 continents, providing industrial AI to 45 million users.

    What Happened?

    NVIDIA CEO Jensen Huang and Dassault Systèmes CEO Pascal Daloz announced their largest partnership ever at 3DEXPERIENCE World in Houston on February 3, 2026.[NVIDIA Blog] The two companies have been collaborating for over 25 years, but this announcement marks the first time NVIDIA’s accelerated computing and AI libraries will be fully integrated with Dassault’s Virtual Twin platform.

    Huang said, “AI will become infrastructure, like water, electricity, and the internet,” and “Engineers will be able to work at scales 100x, 1000x, and ultimately a million times larger.”[NVIDIA Blog] He added that engineers will have teams of AI companions.

    The core of this partnership is Industry World Models. AI systems validated by the laws of physics simulate products, factories, and even biological systems before they are actually built. NVIDIA Omniverse libraries and Nemotron open models will be integrated into Dassault’s 3DEXPERIENCE platform, allowing AI agents called Virtual Companions to support design in real-time.[Dassault Systèmes]

    Why Is It Important?

    Frankly, this is not just a partnership announcement. It’s a move that could change the landscape of industrial AI.

    Virtual Twin is a more advanced concept than the traditional Digital Twin. While Digital Twin is a static 3D replica, Virtual Twin simulates real-time behavior and evolution. This means you can design not only the geometric shape of a product but also how it works at the same time.

    Personally, I think the real significance of this partnership lies in the concept of “AI companions.” Instead of an engineer running CAD alone, AI simulates and suggests thousands of design options in real-time. This allows for exploring a much wider design space in the early stages of design.

    There have been similar attempts already. Siemens and NVIDIA also announced an Industrial AI Operating System at CES 2026, and PepsiCo achieved a 20% improvement in throughput with an AI digital twin in its factory.[NVIDIA Newsroom] However, Dassault has a massive installed base of 45 million users and 400,000 customers. The impact will be different when NVIDIA AI is integrated into a platform of this scale.

    What Happens Next?

    Dassault’s OUTSCALE brand will deploy AI factories across 3 continents. This structure operates industrial AI models while ensuring data sovereignty and privacy.

    But we’ll have to see how much of this actually comes to fruition. “1 million times expansion” is a vision, not an immediate reality. The important thing is whether existing 3DEXPERIENCE users can use this feature without additional cost, or whether a new license is required. Pricing policies have not yet been announced.

    The theme of the 3DEXPERIENCE User Conference in Boston in March 2026 is “AI-Powered Virtual Twin Experiences.”[Dassault Systèmes] A more specific roadmap is expected to be revealed then.

    Frequently Asked Questions (FAQ)

    Q: What is the difference between Virtual Twin and Digital Twin?

    A: Digital Twin is a static 3D replica of a physical product. Virtual Twin includes real-time behavior simulation and evolution over time. It can simulate and predict not only the shape of the product but also how it works and its entire life cycle, allowing for more optimization in the design phase.

    Q: How will this partnership affect existing 3DEXPERIENCE users?

    A: When NVIDIA’s AI libraries and Nemotron models are integrated into the 3DEXPERIENCE platform, users will be able to receive real-time design support from AI companions. However, specific pricing policies and compatibility with existing licenses have not yet been announced, so more information is expected to be released at the March User Conference.

    Q: Didn’t NVIDIA announce a similar partnership with Siemens?

    A: That’s right. NVIDIA announced an Industrial AI Operating System partnership with Siemens at CES 2026. Siemens has strengths in manufacturing automation and factory systems, while Dassault has strengths in product design and PLM. From NVIDIA’s perspective, both partnerships are strategies to expand the Omniverse ecosystem and are more complementary than competitive.

    If you found this article useful, please subscribe to AI Digester.

    References

    Microsoft Paza: Public Benchmark for Speech Recognition in 39 African Languages

    by

    Microsoft Paza: Public Benchmark for Speech Recognition in 39 African Languages

    • Launched the first dedicated ASR leaderboard for low-resource languages
    • Performance comparison of 52 latest models available
    • Also released 3 fine-tuned models for 6 Kenyan languages

    What happened?

    Microsoft Research has released Paza, a speech recognition (ASR) benchmark platform for low-resource languages.[Microsoft Research] Paza comes from the Swahili word meaning ‘raise your voice’. This project consists of two parts: the PazaBench leaderboard and the Paza ASR model.

    PazaBench is the first ASR leaderboard dedicated to low-resource languages. It measures the performance of 52 state-of-the-art ASR and language models for 39 African languages.[Microsoft Research] It tracks three metrics: Character Error Rate (CER), Word Error Rate (WER), and Real-Time Factor (RTFx).

    Why is it important?

    Currently, most speech recognition systems are optimized for major languages such as English and Chinese. Although there are over 1 billion African language users, technical support for them has been lacking. Microsoft’s Project Gecko research also revealed that “speech systems fail in real low-resource environments.”[Microsoft Research]

    The Paza team emphasized that “creating useful speech models in low-resource environments is not just a data problem, but also a design and evaluation problem.” The key is to not simply add languages, but to create technology together with local communities.

    What happens next?

    Paza has released three fine-tuned models for six Kenyan languages (Swahili, Dholuo, Kalenjin, Kikuyu, Maasai, and Somali). These are Paza-Phi-4-Multimodal-Instruct, Paza-MMS-1B-All, and Paza-Whisper-Large-v3-Turbo. It is expected to expand to more African languages in the future. It is released in the form of an open benchmark, allowing researchers to freely test and improve models.

    Frequently Asked Questions (FAQ)

    Q: Which languages does the Paza benchmark support?

    A: It currently supports 39 African languages, including Swahili, Yoruba, and Hausa, and also provides fine-tuned models for 6 Kenyan languages. It is operated in the form of a leaderboard, allowing researchers to directly compare model performance.

    Q: What performance metrics does PazaBench measure?

    A: It measures three metrics. Character Error Rate (CER) measures errors in individual characters, and Word Error Rate (WER) measures errors in words. RTFx represents real-time processing speed and is used to predict response speed during actual deployment.

    Q: Why is speech recognition difficult for low-resource languages?

    A: There is an absolute lack of training data. While English has tens of thousands of hours of speech data, African languages often have only hundreds of hours. In addition, evaluation itself is difficult because there is a large diversity of dialects and some languages lack standard notation.

    If you found this article useful, please subscribe to AI Digester.

    References