×
![]()
How do platform firms set prices and make money?
The post Prediction vs. Search Models: What Data Scientists Are Missing appeared first on Towards Data Science.
Jules Tools is our new, lightweight command-line interface. And Jules API will let you integrate Jules directly into your own systems.
Don't miss our most-read and -shared articles of the past month
The post TDS Newsletter: September Must-Reads on ML Career Roadmaps, Python Essentials, AI Agents, and More appeared first on Towards Data Science.
Get the scoop on how to use Nano Banana, the Gemini app's viral new image generation and editing model from Google DeepMind.
Bakshi will help shape and scale entrepreneurship education and platform at MIT.
How to maintain reliability in inherently stochastic systems
The post AI Engineering and Evals as New Layers of Software Work appeared first on Towards Data Science.
Optimized for generative AI, TX-GAIN is driving innovation in biodefense, materials discovery, cybersecurity, and other areas of research and development.
Get a gentle introduction to the standard that defines how artificial intelligence systems connect with the outside world.
How simple statistics reveal the visual fingerprints of 20 languages
The post What Makes a Language Look Like Itself? appeared first on Towards Data Science.
Salesforce Inc. is expanding its artificial intelligence platform with new data management and governance capabilities, aiming to address what the company says is a crisis in enterprise AI adoption where more than 80% of projects fail to deliver meaningful business value.The San Francisco-based software giant announced Thursday a suite of new tools designed to create what it calls a "trusted AI foundation" for enterprises struggling with fragmented data, weak governance, and security concerns that have hampered AI deployments across corporate America."We're seeing a lot of these AI projects really failing, and a lot of it's because customers still have fragmented data, they still have weak governance, they still have poor security," said Desiree Motamedi, Salesforce's senior vice president and chief marketing officer, in an exclusive interview with VentureBeat. "They really want a way that they can bring AI at scale that has the accuracy, the context and the control."The timing of Salesforce's announcement comes as the company prepares for its annual Dreamforce conference next week, where CEO Marc Benioff is expected to showcase the company's vision for what he calls the "agentic enterprise" — workplaces where AI agents work alongside humans across every business function.Why most corporate AI initiatives crash and burn before reaching productionThe scale of AI project failures has become a significant concern for enterprise technology leaders. According to a RAND Corporation study, poor data quality, inadequate governance frameworks, and fragmented system integration are the primary culprits behind the high failure rate of corporate AI initiatives.This challenge has created both pressure and opportunity for enterprise software providers. While companies face mounting pressure to deploy AI capabilities, many are discovering that their existing data infrastructure isn't equipped to support reliable AI applications at scale.Salesforce's response centers on what Motamedi describes as three core capabilities: ensuring AI outputs are grounded in unified business data, embedding security and compliance controls into every workflow, and connecting AI agents across different platforms and data sources."The Salesforce platform is a $7 billion business," Motamedi noted, highlighting the significant revenue opportunity the company sees in addressing enterprise AI infrastructure needs. "This is a significant opportunity where we're seeing meaningful differentiation from other vendors in the market."Inside Salesforce's new AI tools designed to fix enterprise data chaosThe company's latest announcements include several technically sophisticated solutions aimed at different aspects of the enterprise AI challenge:Data Cloud Context Indexing represents Salesforce's approach to handling unstructured content like contracts, technical diagrams, and decision trees. The system uses what the company calls a "business-aware lens" to help AI agents interpret complex documents within their proper business context."A good example is a field engineer who uploads a schematic for guided troubleshooting," Motamedi explained. "Now they have that capability at their disposal, because it's right there in that view."Data Cloud Clean Rooms, now generally available, allows organizations to securely share and analyze data with partners without exposing sensitive information. Using Salesforce's "zero copy" technology, companies can collaborate on data analysis without actually moving or duplicating datasets.The clean room technology extends beyond traditional advertising applications to sectors like banking, where institutions could "detect fraud, and they want to be able to do it with some of their partners. They could now do it in hours versus weeks," according to Motamedi.Tableau Semantics addresses one of the most persistent challenges in enterprise data management: ensuring consistent definitions of business metrics across different systems and teams. The AI-powered semantic layer translates raw data into standardized business language."We use terms like ACV or churn that have specific definitions within our organization," Motamedi said. "Making sure AI understands those definitions, and then having a standardized layer across organizations, really makes this seamless for enterprises."MuleSoft Agent Fabric tackles what Salesforce calls "agent sprawl" — the proliferation of AI agents across different platforms and vendors within large organizations. The system provides centralized registration, orchestration, and governance for AI agents regardless of where they were built.How Salesforce plans to battle Microsoft, Google and Amazon for AI dominanceSalesforce's comprehensive approach to AI infrastructure positions the company in direct competition with Microsoft, Google, Amazon, and ServiceNow, all of which are vying to become the dominant platform for enterprise AI deployment.The company's strategy relies heavily on integration advantages that come from building AI capabilities into an existing platform used by thousands of enterprises. "The power of the platform" lies in the fact that "all of this is natively into the platform. So these capabilities are just there, and they work and they work seamlessly together," Motamedi emphasized.This integrated approach contrasts with point solutions that require custom integration work. "Some of these point solutions, if you want these things to work together, you got to build those integrations. You got to have developer teams to make that happen," she noted.The company's pending $8 billion acquisition of data management company Informatica, expected to close soon, will significantly expand Salesforce's capabilities in enterprise metadata management — a critical component for AI accuracy."For the last 26 years, Salesforce has been rooted in our platform approach — we've built the metadata layer from day one," Motamedi said. "But with Informatica, we're going to see metadata across the entire enterprise, and that gives us another layer of accuracy for AI responses."Early enterprise customers reveal the reality of scaling AI in large organizationsDespite the technical capabilities, Salesforce acknowledges that enterprise AI adoption remains in early stages. The company reports having "over 12,000 live deployments of Agentforce" — its AI agent platform — but Motamedi describes a wide range of organizational readiness."Every company has a mandate right now to figure out how they can incorporate AI," she said. "We see very interesting ranges from people who are just getting started to people who are like, we're going to build like 80 different agents within their organization."Early customer implementations include AAA Washington, which is using Salesforce's unified data foundation to improve member experiences across roadside assistance, insurance, and travel services. UChicago Medicine is leveraging the platform to ensure reliable patient interactions while enabling healthcare staff to focus on complex, human-centered care.The maturity curve for enterprise AI adoption means "it's going to take a couple years to see it fully, fully embraced, but we already see the path," according to Motamedi.What Salesforce's AI governance push means for the future of enterprise softwareThe broader implications of Salesforce's strategy extend beyond technical capabilities to fundamental questions about how enterprises will manage AI risk and governance. The company's emphasis on built-in security and compliance reflects growing corporate awareness that AI deployment without proper controls can create significant business liability.Recent incidents involving AI agents accessing sensitive information or providing unreliable outputs have made corporate leaders more cautious about scaling AI initiatives. Salesforce's approach of embedding security directly into AI workflows — including automated threat detection partnerships with CrowdStrike and Okta, and built-in HIPAA compliance for healthcare applications — represents an attempt to address these concerns while accelerating adoption.However, market skepticism remains. CNBC's Jim Cramer recently noted concerns about Salesforce's performance despite strong quarterly reports, suggesting that investor expectations for AI-driven growth may be outpacing actual business results.The company's success will ultimately depend on whether it can help enterprises bridge the gap between AI experimentation and production-scale deployment. As Motamedi framed it: "We really believe that we have a trust layer for enterprise AI with all of these new announcements, and we're really helping companies move from cautious pilots to transformative action."Whether that vision becomes reality will depend on Salesforce's ability to prove that integrated platforms can solve enterprise AI's trust problem better than the patchwork of point solutions most companies rely on today. In an industry where 80% of projects fail, the company that finally cracks the code on reliable, scalable enterprise AI could reshape how business gets done — or discover that the technical challenges run deeper than any single platform can solve.
“Deep Think with Confidence,” a smarter way to scale reasoning tasks without wasting a massive amount of computation
The post Smarter, Not Harder: How AI’s Self-Doubt Unlocks Peak Performance appeared first on Towards Data Science.
Let's look at three feature selection techniques and see which one worked best.
Comparing model-free and model-based RL methods on a dynamic grid world
The post Temporal-Difference Learning and the Importance of Exploration: An Illustrated Guide appeared first on Towards Data Science.
Presented by HubSpotINBOUND, HubSpot's annual conference for marketing and sales professionals, took place in San Francisco this year, with three days of insights and events across marketing, sales, CX, and AI innovation. It was a mix of the new, like the Creators Corner and the Tech Stack Showcase Stage, and the familiar, like HubSpot HQ, the Spotlight Product Demo Stage, HubSpot Academy Labs for hands-on product learning, and attendee favorite, Braindates. The opening HubSpot Spotlight on Wednesday, by HubSpot co-founder and CTO, Dharmesh Shah, dug into the nitty gritty of AI, what transformation really looks like, why generative AI is more than just glorified autocomplete, and why it's good a thing that it's not going anywhere, anytime soon."Is it an exponential opportunity? Or is it an existential threat? I asked AI this question. It thought deeply about it for a while and came up with its answer. Yes.""Not everyone is necessarily sold on the merits of AI. It’s the most consequential, and also the most controversial, technology of our time," Shah said. "Is it an exponential opportunity? Or is it an existential threat? I asked AI this question. It thought deeply about it for a while and came up with its answer. Yes."How do you compete with AI?An informal poll of 6K respondents asked the question, "How do you compete with AI?” Shah said a third of the respondents read that as, how do I compete against it? But, he added, that frames it as a zero-sum game which isn't useful to anyone. "We should think about it as a positive-sum collaboration," he explained. "The goal isn’t to battle the machine. The goal is to build with the machine."This is critical because while AI capability is growing on an exponential curve, the learning curve for AI is much more linear, and so is the value seen from embracing the technology that's learned everything it knows from Shakespeare's full body of work, Reddit threads debating whether a hot dog is a sandwich, and Stephen Hawking's academic talks. "At the end of it all, this machine is capable of predicting, with such accuracy, a really good next word, that what it’s doing is indistinguishable from thinking. That’s why it feels so magical," Shah said. "It’s autocomplete with a PhD in everything. It’s like having 1,000 PhDs in your pocket. Suddenly you can write poetry. You can write prose. You can write computer programs. It’s like Neo in the Matrix, except it’s not 'I know kung fu.' It’s 'I know everything.' You have the world’s knowledge sitting there in your pocket.”There are some drawbacks. LLMs are limited to the training data they were given. Sometimes they hallucinate. They're frozen in time, and stateless — but they do learn. User interactions go into their long-term memories, including chats, resources they're asked to interact with, such as PDFs or images, and tools like the internet, databases, APIs with third-party systems and more. How to better use AI in our daily lives "My pro tip: every time you sit down at a computer to do something, try it first with AI. See if it can help, don’t overthink it."So if AI is the real deal, and the transformation is important, how should you use AI? "My pro tip: every time you sit down at a computer to do something, try it first with AI. See if it can help, don’t overthink it," Shah said. "You’ll be surprised. When it doesn’t work, don’t think, oh, well, this doesn’t work. Think, this doesn’t work yet. If it’s an important use case for you, leave yourself a calendar reminder to try it again in three months. Remember, AI is on an exponential curve. There’s a chance, three months or six months from now, that same thing that doesn’t work now will work."The quality of the results you get is based on the quality of the model you use, the quality of the prompt you write, and the quality of the context you put in the context window, otherwise known as prompt engineering. The quality of the model. There's been a Cambrian explosion in the world of large language models, but Shah recommends any of the three top-tier frontier models: OpenAI's GPT-5, Claude from Anthropic, or Google Gemini. But don't overthink it, he adds. Pick one you like or that your company is using.The quality of the prompt. Humans use considerably less than 10% of AI's potential, Shah said. About 95% of the time, users repeat one of the handful of prompts they've found that have worked for the majority of their use cases. Instead, he recommends that 60% of the time, you use the prompts that work, carving out 30% of the time for iteration, or taking the prompts you're using right now and seeing if you can improve them for better results, spending 10% of your time on experimentation, or using AI for things that you’ve never tried before, and you're not sure will work. Metaprompting can be useful here — actually asking AI to make the prompt betterBeyond prompt engineeringContext engineering is just what it sounds like — adding the right context to a request so AI offers better results. Improving the quality of the context includes things like adding custom instructions, which every major LLM includes. That's basically telling the AI what perspective it's working from — the personality, how it should behave, how it should answer and so on. Once you provide custom instructions, it will use them for every answer, and your results improve.The other improvement is a recent change — MCP, or model context protocol, a relatively new way to supply tools to the LLM. Any AI application that supports MCP can immediately and directly connect with the thousands of applications that also support MCP, like HubSpot, to bring that context directly into the LLM. Taking action with AI agents "Last year, at Inbound 24, I predicted that this year would be the year of AI agents. I was wrong," Shah said. "This is not the year of AI agents. This is the decade of AI agents. We are just getting started on this massive transformational wave in AI." Last year Shah launched Agent.ai, a place to discover, use, and build your own AI agents. Over 2 million people are now using the platform. Around 26,000 users have already built agents on Agenti.ai, including Shah, who took all of the information from his keynote prep and put it into an AI agent anyone can chat with, at You.ai. The TEAM strategy: triage, experiment, automate, and measure. The next stepsHow do you take your newfound AI enthusiasm and bring it to your team? Shah offered a gameplan, the TEAM strategy: triage, experiment, automate, and measure. "The goal behind that strategy is to go from AI being led by individual heroics, where you have the super ambitious, super creative person, but translate those individual heroics into team habits," he explained. "Take those things and apply TEAM to go through that process again and again. We are big believers that tomorrow’s teams will be hybrid. This is a time to get started. Even in a small way, start building out your hybrid team."But the most important thing to remember is that as smart as AI is, humans win on EQ, or emotional quotient. Marrying lived experience with an AI tool will improve your life, and improve your experiences, he added. "But my challenge to you is, don’t stop there," he said. "The future does not belong to artificial intelligence. It belongs to you, with augmented intelligence. AI isn’t here to replace us. It’s here to replace the parts of our work that don’t bring us joy. To handle the repetitive so we can focus on the remarkable. The better AI gets, the more it allows us to be human."Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.
IBM today announced the release of Granite 4.0, the newest generation of its homemade family of open source large language models (LLMs) designed to balance high performance with lower memory and cost requirements. Despite being one of the oldest active tech companies in the U.S. (founded in 1911, 114 years ago!), "Big Blue" as its often nicknamed has already wowed many AI industry workers and followers with this new Granite 4.0 family of LLMs, as they offer high performance on third-party benchmarks; a permissive, business friendly license (Apache 2.0) that allows developers and enterprises to freely take, modify and deploy the models for their own commercial purposes; and, perhaps most importantly, have symbolically put the U.S. back into a competitive place with the growing raft of high-performing new generation open source Chinese LLMs, especially from Alibaba's prolific Qwen team — alongside OpenAI with its gpt-oss model family released earlier this summer.Meta, the parent company of Facebook and Instagram, was once seen as the world and U.S. leader of open source LLMs with its Llama models, but after the disappointing release of the Llama 4 family in April and lack of its planned, most powerful Llama 4 Behemoth, it has since pursued a different strategy and is now partnering with outside labs like Midjourney on AI products, while it continues to build out an expensive, in-house AI "Superintelligence" team, as well. Little wonder AI engineer Alexander Doria (aka Pierre-Carl Langlais) observed, with a hilarious Lethal Weapon meme, that "ibm suiting up again after llama 4 fumbled," and "we finally have western qwen."Hybrid (Transformer/Mamba) theoryAt the heart of IBM's Granite 4.0 release is a new hybrid design that combines two very different architectures, or underlying organizational structures, for the LLMs in question: transformers and Mamba.Transformers, introduced in 2017 by Vaswani and colleagues in the famous Google paper “Attention Is All You Need,” power most large language models in use today.In this design, every token — essentially a small chunk of text, like a word or part of a word — can compare itself to every other token in the input. This “all-to-all” comparison is what gives transformers their strong ability to capture context and meaning across a passage. The trade-off is efficiency: because the model must calculate relationships between every possible pair of tokens in the context window, computation and memory demands grow rapidly as the input gets longer. This quadratic scaling makes transformers costly to run on very long documents or at high volume.Mamba, by contrast, is a newer architecture developed in late 2023 by researchers Albert Gu and Tri Dao at Carnegie Mellon University and Princeton University. Instead of comparing every token against all the others at once, it processes tokens one at a time, updating its internal state as it moves through the sequence. This design scales only linearly with input length, making it far more efficient at handling long documents or multiple requests at once. The trade-off is that transformers still tend to perform better in certain kinds of reasoning and “few-shot” learning, where it helps to hold many detailed token-to-token comparisons in memory.This enables much greater efficiency, especially for long documents or multi-session inference, though transformers retain advantages for some reasoning and few-shot learning tasks. But whether the model is built on transformers, Mamba, or a hybrid of the two, the way it generates new words works the same way. At each step, the model doesn’t just pick from what’s already in the context window. Instead, it uses its internal weights — built from training on trillions of text samples — to predict the most likely next token from its entire vocabulary. That’s why, when prompted with “The capital of France is…,” the model can output “Paris” even if “Paris” isn’t in the input text. It has learned from countless training examples that “Paris” is a highly probable continuation in that context. In other words, the context window guides the prediction, but the embedding space — the model’s learned representation of all tokens it knows — supplies the actual words it can generate.By combining Mamba-2 layers with transformer blocks, Granite 4.0 seeks to offer the best of both worlds: the efficiency of Mamba and the contextual precision of transformers.This is the first official Granite release to adopt the hybrid approach. IBM previewed it earlier in 2025 with the Granite-4.0-Tiny-Preview, but Granite 4.0 marks the company’s first full family of models built on the Mamba-transformer combination.Granite 4.0 is being positioned as an enterprise-ready alternative to conventional transformer-based models, with particular emphasis on agentic AI tasks such as instruction following, function calling, and retrieval-augmented generation (RAG). The models are open sourced under the Apache 2.0 license, cryptographically signed for authenticity, and stand out as the first open language model family certified under ISO 42001, an international standard for AI governance and transparency.Reducing memory needs, expanding accessibilityOne of Granite 4.0’s defining features is its ability to significantly reduce GPU memory consumption compared to traditional large language models. IBM reports that the hybrid Mamba-transformer design can cut RAM requirements by more than 70% (!!!) in production environments, especially for workloads involving long contexts and multiple concurrent sessions.Benchmarks released alongside the launch illustrate these improvements. Granite-4.0-H-Small, a 32B-parameter mixture-of-experts model with 9B active parameters, maintains strong throughput on a single NVIDIA H100 GPU, continuing to accelerate even under workloads that typically strain transformer-only systems. This efficiency translates directly into lower hardware costs for enterprises running intensive inference tasks.For smaller-scale or edge deployments, Granite 4.0 offers two lighter options: Granite-4.0-H-Tiny, a 7B-parameter hybrid with 1B active parameters, and Granite-4.0-H-Micro, a 3B dense hybrid. IBM is also releasing Granite-4.0-Micro, a 3B transformer-only model intended for platforms not yet optimized for Mamba-based architectures.Performance benchmarksPerformance metrics suggest that the new models not only reduce costs but also compete with larger systems on enterprise-critical tasks. According to Stanford HELM’s IFEval benchmark, which measures how well LLMs follow instructions from users, Granite-4.0-H-Small surpasses nearly all open weight models in instruction-following accuracy, ranking just behind Meta’s much larger Llama 4 Maverick.The models also show strong results on the Berkeley Function Calling Leaderboard v3, where Granite-4.0-H-Small achieves a favorable trade-off between accuracy and hosted API pricing. On retrieval-augmented generation tasks, Granite 4.0 models post some of the highest mean accuracy scores among open competitors.Notably, IBM highlights that even Granite 4.0’s smallest models outperform Granite 3.3 8B, despite being less than half its size, underscoring the gains achieved through both architectural changes and refined training methods.Trust, safety, and securityAlongside technical efficiency, IBM is emphasizing governance and trust. Granite is the first open model family to achieve ISO/IEC 42001:2023 certification, demonstrating compliance with international standards for AI accountability, data privacy, and explainability.The company has also partnered with HackerOne to run a bug bounty program for Granite, offering up to $100,000 for vulnerabilities that could expose security flaws or adversarial risks. Additionally, every Granite 4.0 model checkpoint is cryptographically signed, enabling developers to verify provenance and integrity before deployment.IBM provides indemnification for customers using Granite on its watsonx.ai platform, covering third-party intellectual property claims against AI-generated content.Training and roadmapGranite 4.0 models were trained on a 22-trillion-token corpus sourced from enterprise-relevant datasets including DataComp-LM, Wikipedia, and curated subsets designed to support language, code, math, multilingual tasks, and cybersecurity. Post-training is split between instruction-tuned models, released today, and reasoning-focused “Thinking” variants, which are expected later this fall.IBM plans to expand the family by the end of 2025 with additional models, including Granite 4.0 Medium for heavier enterprise workloads and Granite 4.0 Nano for edge deployments.Broad availability across platformsGranite 4.0 models are available immediately on Hugging Face, IBM watsonx.ai, with distribution also through partners such as Dell Technologies, Docker Hub, Kaggle, LM Studio, NVIDIA NIM, Ollama, OPAQUE, and Replicate. Support through Amazon SageMaker JumpStart and Microsoft Azure AI Foundry is expected soon.The hybrid architecture is supported in major inference frameworks, including vLLM 0.10.2 and Hugging Face Transformers. Compatibility has also been extended to llama.cpp and MLX, although optimization work is ongoing. The models are also usable in Unsloth for fine-tuning and in Continue for custom AI coding assistants.Enterprise focusEarly access testing by enterprise partners, including EY and Lockheed Martin, has guided the launch. IBM highlights that the models are tailored for real-world enterprise needs, such as supporting multi-agent workflows, customer support automation, and large-scale retrieval systems.Granite 4.0 models are available in both Base and Instruct forms, with Instruct variants optimized for enterprise instruction-following tasks. The upcoming “Thinking” series will target advanced reasoning.Alternate hybrid Mamba / Transformer modelsBesides IBM, several major efforts are already charting different designs for mixing Transformers with Mamba architecture:ModelHybrid strategy / architectureHighlightsAI21 JambaInterleaves Transformer blocks and Mamba layers, with Mixture-of-Experts (MoE) in some layersSupports context lengths up to 256K tokens and offers higher throughput and lower memory usage than pure Transformers while maintaining competitive benchmarksNvidia Nemotron-HReplaces most attention layers with Mamba-2 blocks, retaining a few attention layers where neededDemonstrates up to 3× faster inference throughput compared to pure-Transformer peers while keeping benchmark accuracy comparableNemotron-Nano-2A reasoning-optimized hybrid built on Nemotron’s designReports up to 6× throughput improvement on reasoning tasks while matching or surpassing accuracyDomain-specific variantsHybridized architectures in multimodal models, such as swapping in Mamba layers for decoder componentsShows that the hybrid approach extends beyond text into vision-language applicationsThe Qwen family from Alibaba remains a dense, decoder-only Transformer architecture, with no Mamba or SSM layers in its mainline models. However, experimental offshoots like Vamba-Qwen2-VL-7B show that hybrids derived from Qwen are possible, especially in vision-language settings. For now, though, Qwen itself is not part of the hybrid wave.What Granite 4.0 means for enterprises and what's nextGranite 4.0 reflects IBM’s strategy of combining open access with enterprise-grade safety, scalability, and efficiency. By focusing on lowering inference costs and reinforcing trust with governance standards, IBM positions the Granite family as a practical foundation for enterprises building AI applications at scale.For the U.S., the release carries symbolic weight: with Meta stepping back from leading the open-weight frontier after the uneven reception of Llama 4, and with Alibaba’s Qwen family rapidly advancing in China, IBM’s move positions American enterprise once again as a competitive force in globally available models. By making Granite 4.0 Apache-licensed, cryptographically signed, and ISO 42001-certified, IBM is signaling both openness and responsibility at a moment when trust, efficiency, and affordability are top of mind. This is especially enticing to U.S. and Western-based organizations who may be interested in open source models, but wary of those originating from China — rightly or not — over possible political ramifications and U.S. government contracts.For practitioners inside organizations, this positioning is not abstract. Lead AI engineers tasked with managing the full lifecycle of LLMs will see Granite 4.0’s smaller memory footprint as a way to deploy faster and scale with leaner teams. Senior AI engineers in orchestration roles, who must balance budget limits with the need for efficiency, can take advantage of Granite’s compatibility with mainstream platforms like SageMaker and Hugging Face to streamline pipelines without locking into proprietary ecosystems. Senior data engineers, responsible for integrating AI with complex data systems, will note the hybrid models’ efficiency on long-context inputs, enabling retrieval-augmented generation on large datasets at lower cost. And for IT security directors charged with managing day-to-day defense, IBM’s bug bounty program, cryptographic signing, and ISO accreditation provide clear governance signals that align with enterprise compliance needs.By targeting these distinct roles with a model family that is efficient, open, and hardened for enterprise use, IBM is not only courting adoption but also shaping a uniquely American answer to the open-source challenge posed by Qwen and other Chinese entrants. In doing so, Granite 4.0 places IBM at the center of a new phase in the global LLM race — one defined not just by size and speed, but by trust, cost efficiency, and readiness for real-world deployment.With additional models scheduled for release before the end of the year and broader availability across major AI development platforms, Granite 4.0 is set to play a central role in IBM’s vision of enterprise-ready, open-source AI.
Vibe search, Sora access, AI love, self-aware AI, OK Computer, and more...
A complete review of architectures to make zero-shot predictions in the most common types of datasets.
The post Are Foundation Models Ready for Your Production Tabular Data? appeared first on Towards Data Science.
Learn how to diagnose and resolve bottlenecks in PyTorch using the num_workers, pin_memory, and profiler parameters to maximize training performance.
The post How to Improve the Efficiency of Your PyTorch Training Loop appeared first on Towards Data Science.
OpenAI’s latest app encourages users to generate a personal digital avatar and scroll AI-generated videos of themselves and their friends.
In this post, we share how Hapag-Lloyd developed and implemented a machine learning (ML)-powered assistant predicting vessel arrival and departure times that revolutionizes their schedule planning. By using Amazon SageMaker AI and implementing robust MLOps practices, Hapag-Lloyd has enhanced its schedule reliability—a key performance indicator in the industry and quality promise to their customers.
We’re excited to announce that Rox is generally available, with Rox infrastructure built on AWS and delivered across web, Slack, macOS, and iOS. In this post, we share how Rox accelerates sales productivity with AI agents powered by Amazon Bedrock.
A Harvard Business School study shows that several AI companions use various tricks to keep a conversation from ending.
Thinking Machines Lab, led by a group of prominent former OpenAI researchers, is betting that fine-tuning cutting-edge models will be the next frontier in AI.
A non-technical and accessible guide to the underlying concept behind visual design: visual encoding channels
The post Data Visualization Explained (Part 2): An Introduction to Visual Variables appeared first on Towards Data Science.
Filling the data gap: A machine learning approach to pollen identification in ecology and biotechnology
The post Visual Pollen Classification Using CNNs and Vision Transformers appeared first on Towards Data Science.
In this article, we're going to break down cross-validation in plain English, provide reasons why it is more reliable than the hold-out method, and demonstrate how to use it with basic code and images.
Google DeepMind collaborated with designer Ross Lovegrove and design office Modem.
Microsoft’s multi-agent framework, AutoGen, acts as the backbone for many enterprise projects, particularly with the release of AutoGen v0.4 in January. However, the company aims to harmonize all of its agent framework offerings and bring more observability capabilities to the forefront as well. Microsoft released the Agent Framework on public preview, which will now essentially be the company's sole orchestration and agent framework.Microsoft said AutoGen and Semantic Kernel will “remain in maintenance mode, which means they will not receive new feature investments but will continue to receive bug fixes, security patches and stability updates.”“For future-facing work, however, the roadmap is centered on Microsoft Agent Framework, and customers should plan migration to capture the benefits of open standards, durability and Azure AI Foundry Integration,” the company said in an email to VentureBeat. The company assured existing workloads on AutoGen or Semantic Kernel will be safe because “no breaking changes are planned.”Microsoft’s move to consolidate agent frameworks into one shows the company’s strategy in agentic AI. By closely tying observability and data protection to the framework for building agents, the company aims to enable the creation of agents through post-deployment, all in one place.Agent Framework and FoundryThe Agent Framework consolidates AI workloads into a single SDK, combining the capabilities of both Semantic Kernel and AutoGen, allowing users to build AI agents, manage multi-agent deployments, and set up observability systems. Sarah Bird, chief product officer for Responsible AI at Microsoft, told VentureBeat in an interview that so many developers and businesses have been rapidly experimenting and adopting AI agents, but needed a way to bring a lot of capabilities together. “What's really exciting about what we're releasing this week is a lot of capabilities to help people more successfully build and manage agents in a way that, of course, allows them to be powerful,” Bird said. “But, one that also ensures that they are trustworthy by giving you the tools to, you know, observe their behavior and new guardrails to help them stay on task.”The new framework offers five capabilities for enterprises building AI agents:Local experimentation before deployment in Azure AI FoundryAPI integration through OpenAPI and collaboration across runtimes with A2A and MCP connectionsUse Magentic One and other orchestration agents Reduce context switching across platformsBuild multi-agent systems across different agent platforms like AI Foundry, M365 Copilot or othersMicrosoft is also adding Agent Framework services, such as multi-agent workflows, to its cloud-based Foundry Agent Service. Safety, security and monitoringBird said one of the differentiators for Agent Framework lies in its responsible AI features. Microsoft added:
Task Adherence, which keeps agents aligned to tasksPII Detection, which alerts administrators if an agent accesses sensitive dataPrompt Shields that help protect against prompt injection and highlight risky agent behavior. “For what enterprises need to think about with agents, I believe, are three important categories,” Bird said. “Number one is the quality of the agent, does it actually work and is it staying and completing the ask. The second is security, both traditional security and new types of risks like prompt injection attacks or leaking sensitive data. And the third thing is management because future organizations will have thousands of agents who could have access to different things and tasks.”Microsoft will be contributing to the OpenTelemetry standard for observability. Through AI Foundry, developers with agents built on Agent Framework can track quality, performance and cost. AI Foundry does offer OpenTelemetry observability to agents built with other frameworks and not just Agent Framework.All-in-one agent frameworksAutoGen competed with other agent builders and multi-agent frameworks from LangChain, CrewAI and LlamaIndex and there’s no doubt Agent Framework would either. Microsoft is not the only one hoping to bring all the needed tools to build, deploy and monitor AI agents. LangChain has been building towards offering these tools even as it aims towards a 1.0 release. As agents become more ubiquitous at enterprises, more platforms could look into providing access to building, deployment and observability tools in one.
Google and the University of Waterloo are bringing together their respective expertise in AI and innovative co-op education. This collaboration will fund a new research …