1.
Enterprise sales push OpenAI's quarterly revenue to $10 billion, though it still trails Anthropic
OpenAI's recent quarterly revenue hit $10 billion, marking a significant 50 percent increase. However, they’re still facing hefty adjusted operating losses of $10.9 billion, which shows they’re spending more than they’re earning. In contrast, Anthropic is doing better with $17 billion in revenue while managing costs more effectively. To catch up, OpenAI is diving into a price war aimed at attracting enterprise customers. This shift could reshape the competitive landscape, as they look to balance their books while still pushing for growth. It’s fascinating to see how these dynamics play out in the tech space.
2.
Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three days
Anthropic has made some intriguing strides in AI by leveraging its own technology to shift from Nvidia's CUDA to AMD GPUs. Their model, Mythos, was able to port Claude to AMD hardware in just three days, which is a significant reduction from the months it typically takes. This transition was largely automated, requiring minimal human input, which is pretty fascinating. By running Claude on four different chip families now, Anthropic is not just diversifying its tech stack but also aiming to cut costs and lessen reliance on a single vendor. It’s a clever move that could shake up the landscape a bit.
3.
Experience the art and legacy of Dr. Esther Mahlangu
Today marks the launch of an online retrospective celebrating Dr. Esther Mahlangu, coinciding with her 91st birthday. This project, hosted on Google Arts & Culture, features over 45 stories and a treasure trove of her artwork, life, and legacy, making it accessible to a global audience. The retrospective emerged from a deep-rooted relationship with Esther and her family, aiming to shine a light on South African women artists.
What’s fascinating is how the team used AI to fill in gaps from Esther’s life before 1980, a time that wasn’t well documented. They held workshops and interviews, allowing Esther to share memories that had never been captured before, like her early painting experiences at just 10 years old. This blend of personal storytelling and cutting-edge technology created vivid visualizations of her past, transforming her memories into engaging video content.
The exhibition is a collaboration with various institutions, showcasing how Esther’s journey unfolded through her art and personal history. It’s a thoughtful approach to preserving cultural heritage, ensuring that her voice remains influential in contemporary art. I really hope you take a moment to explore this incredible collection; it’s a beautiful tribute to an artist who has so much to offer.
4.
9 Key LLM Inference Metrics Explained in 5 Minutes
So, this article dives into how we can measure the performance of large language models, focusing on nine key metrics that really help us understand what's happening under the hood. It starts with the idea that just having high throughput doesn’t mean users are having a great experience. The first token might take too long, or the flow of subsequent tokens could slow down, especially when the server's busy.
The metrics cover everything from the time it takes to get that first token, known as Time to First Token (TTFT), to the overall End-to-End latency (E2E), which includes the entire wait time until the last token is received. It also breaks down how long it takes to generate each token after the first one, and provides insights into server load with Requests per Second (RPS) and Tokens per Second (TPS).
What’s interesting is how they measure these metrics over multiple requests to get a clearer picture of performance. They look at averages, medians, and tail percentiles to account for variations. And then there's Goodput, which is all about how many requests meet your service-level objectives, giving a real sense of efficiency.
Overall, these metrics help pinpoint bottlenecks and track performance over time, making it easier to optimize user experience in LLM applications. It’s a fascinating look at the intricacies of how these systems operate!
5.
What Senior Data Science Job Postings Actually Ask for in 2026
In 2026, the landscape for senior data science roles has shifted significantly, revealing a gap between what candidates think is needed and what employers actually want. The focus has moved from just building models to managing them throughout their lifecycle, emphasizing production-grade skills. Python remains essential, but now demands proficiency in specific libraries like scikit-learn and TensorFlow. SQL requirements have also deepened, with advanced capabilities becoming standard.
Cloud expertise is crucial, with many postings specifying services rather than just familiarity, and about 20% even requesting cloud certifications. MLOps skills are increasingly important, highlighting the need for hands-on experience with tools like MLflow and Docker. Generative AI skills are on the rise, with a notable 31% of postings mentioning them, reflecting a broader expectation for candidates to be fluent in AI technologies. Statistical knowledge continues to be a requirement, underscoring the analytical foundation still needed in the field. Overall, the role of a senior data scientist in 2026 demands a blend of technical depth, cloud fluency, and business acumen.
6.
Try Tabletop: Customizable, collaborative spaces for groups to plan, play and do more together.
So, there's this new tool called Tabletop from Google Labs that’s all about making group planning way easier. Imagine you're trying to organize a family trip or a potluck; instead of juggling a million messages, you can just set up a shared space where everyone can contribute ideas and files in real time. You can start by adding your links and thoughts, and then just type @tabletop or choose a template to whip up things like trip itineraries, dinner polls, or even tournament brackets. Once you’ve got your space set up, you can invite everyone to join in, just like you would with a Google Doc. Right now, it’s only available in the U.S. for folks over 18, and you have to sign up on a waitlist. It seems like a neat way to get everyone on the same page without the usual chaos.
7.
NSF’s X-Labs Program Draws Support From Charities and Investors
In October, they announced that consortium partners would provide practical resources—like lab space and expert guidance—to help X-Labs winners, who will receive $1.5 million for a nine-month planning phase. This collaboration is designed to prevent winners from wasting funds on basic infrastructure, allowing them to focus on ambitious research goals.
The program emphasizes autonomy, requiring project leaders to work full-time and make independent decisions, which sets it apart from traditional academic settings. Consortium partners, like Ginkgo Bioworks, are stepping in to offer significant resources, including access to their cloud lab for complex experiments. This shift reflects a broader trend of integrating private sector capabilities into government-led science, as companies have grown much larger and more capable since the mid-20th century.
Experts are optimistic about this new model, likening it to "gig work for science," which could foster innovation in ways traditional funding hasn't. Some academics are already embracing the X-Labs approach, finding it a valuable avenue for exploring larger projects that exceed typical university grants. Supporters argue that as technologies like AI evolve, so too must the infrastructure supporting research, creating exciting opportunities for those with an entrepreneurial spirit.
8.
When Does It Actually Make Sense to Put AI Into Products?
The tech world has been racing to embed AI into everything, from toothbrushes to household robots, but it’s becoming clear that not every product benefits from this integration. Savannah Kunovsky from Ideo highlights the importance of discerning when AI enhances user experience and when it just complicates things. Many companies feel pressured to adopt AI, often driven by investor expectations or competitive trends, but the key is understanding the user’s needs first.
Kunovsky points out that AI can be useful, especially in areas like accessibility, but it can also backfire when it replaces intuitive, familiar controls with something less reliable. For example, a cooktop that relies solely on AI led to frustrating experiences for users who needed immediate control. The best AI integrations are those that enhance the experience without adding unnecessary complexity, allowing users to interact with products in multiple ways.
Ultimately, successful AI use should feel seamless and intuitive, enhancing our ability to achieve tasks rather than creating barriers. It’s about finding that balance and ensuring that technology serves the user, not the other way around.
9.
UK regulator threatens universities with sanctions unless they adopt AI-proof exam formats
The Office for Students in England is raising alarms about the impact of AI on academic integrity. They’ve discovered that students are using tools like ChatGPT to produce entire papers without a grasp of the material, which undermines the value of degrees. As a response, the regulator is urging universities to revert to traditional exam formats to ensure fair assessments. If institutions don’t comply, they could face sanctions. This shift back to conventional testing methods is aimed at preserving the credibility of higher education in a landscape increasingly influenced by AI technologies. It’s a significant move that could reshape how we think about assessments.
10.
Grok-built apps exposed patient data, plaintext passwords, and SpaceXSI's own sales pipeline
So, here’s the scoop. A security firm named Claroc discovered that over 35,000 apps built by Grok were left wide open for anyone to access without needing a login. This includes sensitive patient data and even confidential documents related to SpaceXSI. The issue stemmed from a default setting that allowed access to “anyone with the link,” which is a pretty big oversight. SpaceXSI reacted by taking down just 24 of its own employee apps, but the majority of the exposed apps are still out there, accessible to anyone who stumbles upon them. It’s a stark reminder of how crucial it is to double-check those privacy settings.