InfoQ's New Certification Focuses on Practical Skills for Senior Developers and Architects - Related to skills, model, architects, llm, focuses

DeepSeek Open-Sources DeepSeek-R1 LLM with Performance Comparable to OpenAI's o1 Model

DeepSeek open-sourced DeepSeek-R1, an LLM fine-tuned with reinforcement learning (RL) to improve reasoning capability. DeepSeek-R1 achieves results on par with OpenAI's o1 model on several benchmarks, including MATH-500 and SWE-bench.

DeepSeek-R1 is based on DeepSeek-V3, a mixture of experts (MoE) model lately open-sourced by DeepSeek. This base model is fine-tuned using Group Relative Policy Optimization (GRPO), a reasoning-oriented variant of RL. The research team also performed knowledge distillation from DeepSeek-R1 to open-source Qwen and Llama models and released several versions of each; these models outperform larger models, including GPT-4, on math and coding benchmarks.

[DeepSeek-R1 is] the first step toward improving language model reasoning capabilities using pure reinforcement learning (RL). Our goal is to explore the potential of LLMs to develop reasoning capabilities without any supervised data, focusing on their self-evolution through a pure RL [website] in a wide range of tasks, including creative writing, general question answering, editing, summarization, and more. Additionally, DeepSeek-R1 demonstrates outstanding performance on tasks requiring long-context understanding, substantially outperforming DeepSeek-V3 on long-context benchmarks.

To develop the model, DeepSeek started with DeepSeek-V3 as a base. They first tried fine-tuning it only with RL, and without any supervised fine-tuning (SFT), producing a model called DeepSeek-R1-Zero, which they have also released. This model exhibits strong reasoning performance, but " powerful reasoning behaviors, it faces several issues. For instance, DeepSeek-R1-Zero struggles with challenges like poor readability and language mixing."

To address this, the team used a short stage of SFT to prevent the "cold start" problem of RL. They collected several thousand examples of chain-of-thought reasoning to use in SFT of DeepSeek-V3 before running RL. After the RL process converged, they then collected more SFT data using rejection sampling, resulting in a dataset of 800k samples. This dataset was used for further fine-tuning and to produce the distilled models from Llama and Qwen.

DeepSeek evaluated their model on a variety of reasoning, math, and coding benchmarks and compared it to other models, including [website], GPT-4o, and o1. DeepSeek-R1 outperformed all of them on several of the benchmarks, including AIME 2024 and MATH-500.

Within a few days of its release, the LMArena showcased that DeepSeek-R1 was ranked #3 overall in the arena and #1 in coding and math. It was also tied for #1 with o1 in "Hard Prompt with Style Control" category.

Django framework co-creator Simon Willison wrote about his experiments with one of the DeepSeek distilled Llama models on his blog:

Each response starts with a ... pseudo-XML tag containing the chain of thought used to help generate the response. [Given the prompt] "a joke about a pelican and a walrus who run a tea room together"[website] then thought for 20 paragraphs before outputting the joke!...[T]he joke is awful. But the process of getting there was such an interesting insight into how these new models work.

DeepSeek is rapidly emerging as a strong builder of open models. Not only are these models great performers, but their license permits use of their outputs for distillation, potentially pushing forward the state of the art for language models (and multimodal models) of all sizes.

The DeepSeek-R1 models are available on HuggingFace.

Hugging Face has launched the integration of four serverless inference providers Fal, Replicate, SambaNova, and Together AI, directly into its model p......

How do you get testers and developers to cooperate on tests?

If developers help out with tests at all — that's already a good start. But even then, t......

This is part two of a Makers series on the state of observability. Part one featured Christine Yen , CEO and co-founder of [website].

InfoQ's New Certification Focuses on Practical Skills for Senior Developers and Architects

InfoQ is introducing its first hands-on software architecture certification at QCon London 2025 (April 7-11), the international software development conference. The certification will combine practitioner-led conference sessions with a hands-on workshop focused on real-world architectural challenges.

Wesley Reisz, sixteen-time QCon chair, co-host of the InfoQ Podcast, and technical principal consultant at Equal Experts, will lead the InfoQ Certified Software Architect in Emerging Technologies (ICSAET) certification workshop.

"While many certification programs exist, few address the practical challenges of implementing emerging technologies at enterprise scale," expressed Reisz. "This certification bridges that gap by focusing on real-world scenarios software architects face daily."

The certification requires participants to have a minimum of five years of senior technical experience. Participants will work through architectural challenges drawn from actual enterprise implementations inspired by the conference sessions, including "Architectures You've Always Wondered About" which capabilities real-world examples from companies scaling systems for massive traffic and complexity, "Modern Data Architectures" which addresses critical challenges in building scalable systems that integrate AI and machine learning, and "The Changing Face of Architectural Practice" that will examine how traditional approaches are evolving to meet new challenges.

The InfoQ Certified Software Architect in Emerging Technologies (ICSAET) certification differs from traditional architecture certifications through its:

Integration with QCon’s cutting-edge software architecture tracks.

Focus on practical implementation challenges.

Real enterprise implementation scenarios.

"Participants will work through actual challenges we've encountered in enterprise implementations," Reisz explained. "The focus is on the trade-offs and constraints that shape real-world solutions, not theoretical frameworks."

Dio Synodinos, president of C4Media, Inc., the makers of InfoQ and QCon, expressed:

"This certification emerges from nearly two decades of InfoQ and QCon's work with international enterprise software teams. Through our conferences and community, we've helped senior software developers navigate emerging technologies and architectural challenges since 2006. We've seen firsthand what senior architects need: practical insights into implementing emerging technologies at scale. That's exactly what this certification delivers - real-world knowledge from practitioners who are actively solving today's complex enterprise-scale architectural challenges."

The ICSAET certification signals participants’ authority as senior technical leaders and their dedication to continuous growth. Successfully completing both the QCon London conference and the workshop will earn participants the InfoQ Certified Software Architect in Emerging Technologies (ICSAET) credential, which demonstrates leadership and expertise in modern software architecture.

The certification will initially be available at QCon London 2025, with plans to expand to QCon San Francisco 2025 in November. The workshop component will be limited in size to ensure quality interactions and meaningful peer discussions.

Registration for the ICSAET certification is now open. More information and registration details can be found at [website].

As the Trump administration revokes Executive Order 14110, the [website] shifts toward a market-driven AI strategy, departing from the Biden administration......

State management and reactivity are at the heart of modern frontend development, determining how applications enhancement and respond to user interactions.......

Overview of Databases and Their Significance in Data Management.

Databases are structured repositories of information that can be readily accessed, co......

Vercel Rolls Out More Cost-Effective Infrastructure Model

Gone are the days of the edge worker/runtime for frontend cloud hosting provider Vercel, CEO Guillermo Rauch tweeted early yesterday.

“In fact, as of last week, all Edge middleware runs on Fluid in the Vercel cloud,” he added.

Edge Network is a Vercel offering that is both a Content Delivery Network (CDN) and a globally distributed platform for running compute in regions around the globe.

Fluid is a new web application infrastructure model that purports to blend the best of servers and serverless, while offering efficient resource utilization and — more importantly for Vercel consumers — reduced costs.

Fluid is multiregion, but that doesn’t mean “dozens or hundreds of CDN edges,” Rauch wrote.

Edge computing is best for routing, static assets and pre-renders, but it’s not for apps, APIs and databases, Rauch stated in his tweet and reiterated in a webcast later that day with CTO Malte Ubl.

“The big insight is that your application, [the] real logic, has to run close to the data, because you’re going back and forth as data waterfalls; that’s going to be slow,” Ubl stated. “The dream of edge compute, that you suddenly have your data in all these 200 locations, that dream is just not reality.”.

Plus, the vast majority of Vercel end-people host their data in one location, he added.

“The example I like to give is that Google has eight data centers for Google search,” Rauch added. “Google did pretty well with eight data centers. You’re not going to copy petabytes of data to 200 locations.”.

With Fluid, the compute runs closer to where your data already lives instead of “attempting unrealistic replication across every edge location,” wrote Mariano Cocirio, product manager for CI/CD and Compute, in a post introducing Fluid.

“The dream of edge compute, that you suddenly have your data in all these 200 locations, that dream is just not reality.”.

“Rather than forcing widespread data distribution, this approach ensures your compute is placed in regions that align with your data, optimizing for both performance and consistency,” he expressed. “Dynamic requests are routed to the nearest healthy compute region — among your designated locations — ensuring efficient and reliable execution.”.

For enterprise consumers, multi-region failover is the default when activating Fluid, he added.

Serverless computing can suffer from cold starts, which are delays that occur when a function is invoked for the first time or after inactivity.

Serverless functions run in containers. When the function is deployed, the cloud provider packages the code and dependencies into a container. Containers need initialization when a function is invoked, a process that takes time as the container is allocated, initialized, and the code loaded. Also, containers shut down due to inactivity to save resources.

“If I turn on Fluid, which I just did, I’m going to send another four requests, 1-2-3-4, they’re all hitting the same instance,” Ubl noted. “What you can see in the demo is it counts the total duration of the time the function was alive, in this case, [website] seconds. So that’s all I’m billed for. Yes, so you get built for [website] seconds instead of 12 seconds […] that’s what Fluid is.”.

Fluid reduces the frequency of cold starts by maintaining a ‘warm instance.’ It trades single-invocation functions for high-performance mini-servers, Cocirio stated.

“When cold starts do happen, a Rust-based runtime with full [website] and Python support accelerates initialization,” Cocirio wrote. “Bytecode caching further speeds up invocation by pre-compiling function code, reducing startup overhead.”.

As a result, the model maximizes resource efficiency and, in early adopters, has reduced costs by up to 85%, he added.

Fluid bills on actual compute usage, minimizing waste, he emphasized. It also prioritizes using existing resources before creating new instances, “eliminating hard scaling limits and leveraging warm compute for faster, more efficient scaling,” Cociro noted. “By scaling functions before instances, Fluid shifts to a many-to-one model that can handle tens of thousands of concurrent invocations.”.

Fluid also mitigates the risk of uncontrolled execution, which can drive up costs, Cicirio explained. Functions that are waiting on backend responses can process other requests instead of wasting compute.

Fluid also supports advanced tasks such as streaming and post-response processing. It’s fully managed, which retains one of the appealing aspects of the serverless model.

Despite that, Vercel is not automatically switching clients over to Fluid — it does require clients to flip one switch in the functions tab under Project Settings to enable it. Rauch explained that Vercel decided not to enable it for everyone because the execution model slightly changes.

“It requires no code changes. We have … mitigations built in. It’s powered by Node and Python and more open runtimes to come, and you’re ready to enable it today,” Rauch presented.

We call this an event when a button is pressed; a sensor detects a temperature change, or a transaction flows through. An event is an action or state ......

AI-driven data trends in Indian governance in 2025 are revolutionizing decision-making, enhancing efficiency, and improving public servi......

AI is being used to open fake feature requests in open source repos, . So far, AI-driven issues have been reported in Cur......