VentureBeat

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Google’s Gemini series of AI large language models (LLMs) started off rough nearly a year ago with some embarrassing incidents of image generation gone awry, but it has steadily improved since then, and the company appears to be intent on making its second-generation effort — Gemini 2.0 — the biggest and best yet for consumers and enterprises. Today, the company announced the general release of Gemini 2.0 Flash, introduced Gemini 2.0 Flash-Lite, and rolled out an experimental version of Gemini 2.0 Pro. These models, designed to support developers and businesses, are now accessible through Google AI Studio and Vertex AI, with Flash-Lite in public preview and Pro available for early testing. “All of these models will feature multimodal input with text output on release, with more modalities ready for general availability in the coming months,” Koray Kavukcuoglu, CTO of Google DeepMind, wrote in the company’s announcement blog post — showcasing an advantage Google is bringing to the table even as competitors such as DeepSeek and OpenAI continue to launch powerful rivals. Google plays to its multimodal strenghts Neither DeepSeek-R1 nor OpenAI’s new o3-mini model can accept multimodal inputs — that is, images and file uploads or attachments. While R1 can accept them on its website and mobile app chat, The model performs optical character recognition (OCR) a more than 60-year-old technology, to extract the text only from these uploads — not actually understanding or analyzing any of the other features contained therein. However, both are a new class of “reasoning” models that deliberately take more time to think through answers and reflect on “chains-of-thought” and the correctness of their responses. That’s opposed to typical LLMs like the Gemini 2.0 pro series, so the comparison between Gemini 2.0, DeepSeek-R1 and OpenAI o3 is a bit of an apples-to-oranges. But there was some news on the reasoning front today from Google, too: Google CEO Sundar Pichai took to the social network X to declare that the Google Gemini mobile app for iOS and Android has been updated with Google’s own rival reasoning model Gemini 2.0 Flash Thinking. The model can be connected to Google Maps, YouTube and Google Search, allowing for a whole new range of AI-powered research and interactions that simply can’t be matched by upstarts without such services like DeepSeek and OpenAI. I tried it briefly on the Google Gemini iOS app on my iPhone while writing this piece, and it was impressive based on my initial queries, thinking through the commonalities of the top 10 most popular YouTube videos of the last month and also providing me a table of nearby doctors’ offices and opening/closing hours, all within seconds. Gemini 2.0 Flash enters general release The Gemini 2.0 Flash model, originally launched as an experimental version in December, is now production-ready. Designed for high-efficiency AI applications, it provides low-latency responses and supports large-scale multimodal reasoning. One major benefit over the competition is in its context window, or the number of tokens that the user can add in the form of a prompt and receive back in one back-and-forth interaction with an LLM-powered chatbot or application programming interface (API). While many leading models, such as OpenAI’s new o3-mini that debuted last week, only support 200,000 or fewer tokens — about the equivalent of a 400 to 500 page novel — Gemini 2.0 Flash supports 1 million, meaning it is is capable of handling vast amounts of information, making it particularly useful for high-frequency and large-scale tasks. Gemini 2.0 Flash-Lite arrives to bend the cost curve to the lowest yet Gemini 2.0 Flash-Lite, meanwhile, is an all-new LLM aimed at providing a cost-effective AI solution without compromising on quality. Google DeepMind states that Flash-Lite outperforms its full-size (larger parameter-count) predecessor, Gemini 1.5 Flash, on third-party benchmarks such as MMLU Pro (77.6% vs. 67.3%) and Bird SQL programming (57.4% vs. 45.6%), while maintaining the same pricing and speed. It also supports multimodal input and features a context window of 1 million tokens, similar to the full Flash model. Currently, Flash-Lite is available in public preview through Google AI Studio and Vertex AI, with general availability expected in the coming weeks. As shown in the table below, Gemini 2.0 Flash-Lite is priced at $0.075 per million tokens (input) and $0.30 per million tokens (output). Flash-Lite is positioned as a highly affordable option for developers, outperforming Gemini 1.5 Flash across most benchmarks while maintaining the same cost structure. Logan Kilpatrick highlighted the affordability and value of the models, stating on X: “Gemini 2.0 Flash is the best value prop of any LLM, it’s time to build!” Indeed, compared to other leading traditional LLMs available via provider API, such as OpenAI 4o-mini ($0.15/$0.6 per 1 million tokens in/out), Anthropic Claude ($0.8/$4! per 1M in/out) and even DeepSeek’s traditional LLM V3 ($0.14/$0.28), in Gemini 2.0 Flash appears to be the best bang for the buck. Gemini 2.0 Pro arrives in experimental availability with 2-million token context window For users requiring more advanced AI capabilities, the Gemini 2.0 Pro (experimental) model is now available for testing. Google DeepMind describes this as its strongest model for coding performance and the ability to handle complex prompts. It features a 2 million-token context window and improved reasoning capabilities, with the ability to integrate external tools like Google Search and code execution. Sam Witteveen, co-founder and CEO of Red Dragon AI and an external Google developer expert for machine learning who often partners with VentureBeat, discussed the Pro model in a YouTube review. “The new Gemini 2.0 Pro model has a two-million-token context window, supports tools, code execution, function calling and grounding with Google Search — everything we had in Pro 1.5, but improved.” He also noted of Google’s iterative approach to AI development: “One of the key differences in Google’s strategy is that they release experimental versions of models before they go GA (generally accessible), allowing for rapid iteration based

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Riffusion, a San Francisco-based artificial intelligence startup, unveiled a free web platform today that enables anyone to create original music using AI, marking a significant advance in generative AI’s expansion into creative domains traditionally dominated by human artists. The company’s new AI model, called Fuzz, can generate complete songs from text descriptions, audio clips or visual prompts. What sets it apart is its ability to learn individual users’ musical preferences over time, creating an increasingly personalized creative experience. “We all love music, but so few of us make it. As we grow up, we’re told we don’t have enough creativity, talent or time. But we’re all inherently creative. We all have a sound within us,” said Seth Forsgren, Riffusion’s cofounder and CEO, in an exclusive interview with VentureBeat. A song created on Riffusion’s new AI platform demonstrates the technology with “Refresh Rate,” a track about keeping up with tech news. The song was generated from a simple text prompt by VentureBeat. (Credit: Riffusion) Fuzz: The AI that learns your musical taste and creates personalized tracks The platform’s release comes at a pivotal moment in AI music generation, as tech giants like Google, Meta and TikTok build upon Riffusion’s original insight of using diffusion models for audio synthesis. Industry veterans are taking notice. The Chainsmokers’ Alex Pall, who recently joined Riffusion’s advisory board, notes: “We always have a blast playing around with Riffusion and honestly have gotten some real value out of it professionally. Music is supposed to be fun, and unlocking creativity and ideas can be a challenge.” Professional musicians embrace AI as a creative partner, not a replacement The platform’s launch represents a significant shift in music creation accessibility. Unlike competitors that charge subscription fees, Riffusion is making its platform freely available worldwide. The company’s small team of 10 artists, engineers and researchers has focused on creating an intuitive interface that appeals to both professional musicians and casual enthusiasts. In blind human evaluations, Fuzz outperformed competitive models when given identical lyrics and sound prompts. The technology builds upon the company’s groundbreaking work in applying image diffusion techniques to audio spectrograms, a method that has since influenced research at major tech companies. From basement project to industry disruptor: Riffusion’s rapid rise The music industry faces a watershed moment as AI tools become more sophisticated. While some artists express concern about AI’s impact on creative professionals, others see it as an opportunity. Riffusion’s approach of positioning AI as a collaborative tool rather than a replacement for human creativity appears to be resonating with both amateur enthusiasts and professional musicians. The company, which raised $4 million in seed funding in October 2023, plans additional developments later this year. As the AI music generation space becomes increasingly competitive, Riffusion’s free-to-use model could prove disruptive to established players in the digital audio workspace market. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Cognida.ai, a Chicago-based AI company, has raised $15 million in Series A funding to help enterprises move beyond AI pilots to production-grade solutions that deliver measurable business impact. The funding round was led by Nexus Venture Partners. The investment comes at a critical time when enterprises are struggling to transform AI experiments into operational solutions. While 87% of enterprises are investing in AI, only 20% successfully deploy solutions into production, according to Cognida. “Enterprise AI adoption has reached its tipping point,” Feroze Mohammed, founder and CEO of Cognida, said in an exclusive interview with VentureBeat. “The biggest challenges enterprises face isn’t just building AI models — it’s getting them to work in production.” How Zunō platform cuts AI implementation time from 8 months to 12 weeks Mohammed, who previously led Hitachi Vantara as COO, identified three major barriers to enterprise AI adoption: data readiness, integration challenges with existing business processes and lack of AI expertise within organizations. To address these challenges, Cognida has developed Zunō, a platform that includes accelerators for predictive modeling, intelligent document processing and advanced graph-based solutions. The company claims its approach reduces typical AI implementation times from 6 to 8 months to 10 to 12 weeks. “Most conventional approaches require long lead times of doing consulting projects, doing a lot of change management with long timelines and long upfront investments,” Anup Gupta, managing director at Nexus Venture Partners, said in an interview with VentureBeat. “Cognida is one of the first times we have come across a business that can talk about demonstrable use cases across various industries.” The company has already deployed solutions at more than 30 enterprises. In one case, Cognida helped a major garage door manufacturer transform its catalog generation process from a six-month cycle to just weeks using generative AI. The solution allows the manufacturer to create virtual door designs and render them in different settings, enabling rapid testing with dealers. Other successful implementations include a 70% improvement in invoice processing speed and a one percent reduction in customer churn for SaaS clients — metrics that translate to significant revenue impact for large enterprises. The future of enterprise software: Every stack is being rewritten with AI The funding will support three primary initiatives: market expansion, intellectual property development and capability building. Mohammed envisions Cognida becoming “the practical AI company for the enterprise” within five years. “Every software stack is being rewritten leveraging AI,” said Gupta. “In the next few years, every workflow in all enterprises will have a lot more AI than is being used today.” This investment reflects a broader trend in enterprise AI, where focus is shifting from experimental projects to practical implementations that deliver clear return on investment. As businesses seek to operationalize AI while maintaining existing systems, Cognida’s approach of building solutions that integrate with current workflows appears particularly timely. The company plans to expand its AI solution library, advance its Zunō platform and grow its implementation teams to meet increasing enterprise demand. With offices in Chicago, Silicon Valley and Hyderabad, India, Cognida serves clients across manufacturing, healthcare, finance and technology sectors. Industry analysts suggest this funding round could signal a new phase in enterprise AI adoption, where practical implementation and measurable outcomes take precedence over experimental pilots. As organizations continue to grapple with AI integration challenges, solutions that can demonstrate concrete business impact while working within existing systems may find increasing traction in the market. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Reasoning models like OpenAI o1 and DeepSeek-R1 have a problem: They overthink. Ask them a simple question such as “What is 1+1?” and they will think for several seconds before answering. Ideally, like humans, AI models should be able to tell when to give a direct answer and when to spend extra time and resources to reason before responding. A new technique presented by researchers at Meta AI and the University of Illinois Chicago trains models to allocate inference budgets based on the difficulty of the query. This results in faster responses, reduced costs, and better allocation of compute resources. DeepSeek solving 1+1 Costly reasoning Large language models (LLMs) can improve their performance on reasoning problems when they produce longer reasoning chains, often referred to as “chain-of-thought” (CoT).  The success of CoT has led to an entire range of inference-time scaling techniques that prompt the model to “think” longer about the problem, produce and review multiple answers and choose the best one. One of the main ways used in reasoning models is to generate multiple answers and choose the one that recurs most often, also known as “majority voting” (MV). The problem with this approach is that the model adopts a uniform behavior, treating every prompt as a hard reasoning problem and spending unnecessary resources to generate multiple answers. Smart reasoning The new paper proposes a series of training techniques that make reasoning models more efficient at responding. The first step is “sequential voting” (SV), where the model aborts the reasoning process as soon as an answer appears a certain number of times. For example, the model is prompted to generate a maximum of eight answers and choose the answer that comes up at least three times. If the model is given the simple query mentioned above, the first three answers will probably be similar, which will trigger the early-stopping, saving time and compute resources. Their experiments show that SV outperforms classic MV in math competition problems when it generates the same number of answers. However, SV requires extra instructions and token generation, which puts it on par with MV in terms of token-to-accuracy ratio. SV outperforms MV on number of responses but matches it on number of tokens (source: arXiv) The second technique, “adaptive sequential voting” (ASV), improves SV by prompting the model to examine the problem and only generate multiple answers when the problem is difficult. For simple problems (such as the 1+1 prompt), the model simply generates a single answer without going through the voting process. This makes the model much more efficient at handling both simple and complex problems.  Reinforcement learning While both SV and ASV improve the model’s efficiency, they require a lot of hand-labeled data. To alleviate this problem, the researchers propose “Inference Budget-Constrained Policy Optimization” (IBPO), a reinforcement learning algorithm that teaches the model to adjust the length of reasoning traces based on the difficulty of the query. IBPO is designed to allow LLMs to optimize their responses while remaining within an inference budget constraint. The RL algorithm enables the model to surpass the gains obtained through training on manually labeled data by constantly generating ASV traces, evaluating the responses, and choosing outcomes that provide the correct answer and the optimal inference budget. Their experiments show that IBPO improves the Pareto front, which means for a fixed inference budget, a model trained on IBPO outperforms other baselines. IBPO (green circles) outperforms other baselines on the Pareto front (source: arXiv) The findings come against the backdrop of researchers warning that current AI models are hitting a wall. Companies are struggling to find quality training data and are exploring alternative methods to improve their models. One promising solution is reinforcement learning, where the model is given an objective and allowed to find its own solutions as opposed to supervised fine-tuning (SFT), where the model is trained on manually labeled examples. Surprisingly, the model often finds solutions that humans haven’t thought of. This is a formula that seems to have worked well for DeepSeek-R1, which has challenged the dominance of U.S.-based AI labs. The researchers note that “prompting-based and SFT-based methods struggle with both absolute improvement and efficiency, supporting the conjecture that SFT alone does not enable self-correction capabilities. This observation is also partially supported by concurrent work, which suggests that such self-correction behavior emerges automatically during RL rather than manually created by prompting or SFT.” source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Agentic AI is all the rage today across multiple sectors, including application development and coding. Today at long last, GitHub has joined the agentic AI party with the launch of GitHub Copilot agent mode. The promise of agentic AI in development is about enabling developers to build more code with just a simple prompt. The new agent mode will enable Copilot to iterate on its own code and fix errors automatically. Looking forward, GitHub is also previewing a fully autonomous software engineering agent, Project Padawan, that can independently handle entire development tasks. The new agentic AI features mark the latest step in the multi-year evolution of the AI-powered coding development space that GitHub helped to pioneer. The Microsoft-owned GitHub first previewed GitHub Copilot in 2021, with general availability coming in 2022. In the AI world, that’s a long time ago, before ChatGPT became a household name and most people had ever heard the term “generative AI.” GitHub has been steadily iterating on Copilot. Initially, the service relied on the OpenAI Codex large language model (LLM). In October 2024, users gained the ability to choose from a variety of LLMs, including Anthropic’s Claude, Google’s Gemini 1.5 and OpenAI’s GPT4o. Alongside the agent mode launch, GitHub is now also adding support for Gemini 2.0 Flash and OpenAI’s o3-mini. Microsoft overall has been emphasizing agentic AI, assembling one of the largest AI agent ecosystems in the market. AI that supports ‘peer programming’ The new GitHub Copilot agent mode service comes as a series of rivals, mostly led by startups, have shaken up the development landscape. Cursor, Replit, Bolt and Lovable are all chasing the growing market for AI-powered development that GitHub helped to create. When GitHub Copilot first emerged, it was positioned as a pair programming tool, which pairs with a developer. Now, GitHub is leaning into the term peer programming as it embraces agentic AI. “Developer teams will soon be joined by teams of intelligent, increasingly advanced AI agents that act as peer-programmers for everyday tasks,” said GitHub CEO Thomas Dohmke. “With today’s launch of GitHub Copilot agent mode, developers can generate, refactor and deploy code across the files of any organization’s codebase with a single prompt command.” Technical breakdown: How GitHub’s new agent architecture works Since its initial debut, GitHub Copilot has provided a series of core features. Among them is intelligent code completion, which is the ability to suggest code snippets to execute a given function. Copilot also functions as an assistant, allowing developers to input natural language queries to generate code, or get answers about a specific code base. The system, while intelligent, still requires a non-trivial amount of human interaction. Agent mode goes beyond that. According to GitHub, the platform enables Copilot to iterate on its own output, as well as the results of that output. This can significantly improve results and code output. Here’s a detailed breakdown of agent mode operation. Task understanding and planning: When given a prompt, agent mode doesn’t just generate code — it analyzes complete task requirements; According to GitHub, the system can “infer additional tasks that were not specified, but are also necessary for the primary request to work”.  Iterative execution: The agent iterates on both its own output and the result of that output; It continues iteration until all subtasks are completed. Self-healing capabilities: Automatically recognizes errors in its output; Can fix identified issues without developer intervention; Analyzes runtime errors and implements corrections; Suggests and executes necessary terminal commands. Project Padawan brings the ‘force’ to development While agent mode certainly is more powerful than the basic GitHub Copilot operation, it’s still not quite a fully automated experience. To get to that full experience, GitHub is previewing Project Padawan. In popular culture, a ‘Padawan’ is a reference to a Jedi apprentice from the Star Wars science fiction franchise.  Project Padawan builds on the agent mode and extends it with more automation. In a blog post, Dohmke noted that Padawan will allow users to assign an issue to GitHub Copilot, and the agentic AI system will handle the entire task. That task can include code development, setting up a repository and assigning humans to review the final code. “In a sense, it will be like onboarding Copilot as a contributor to every repository on GitHub,” Dohmke said. Comparing GitHub’s agent to other agentic AI coding options GitHub in some respects is a late entrant to the agentic AI coding race. Cursor AI and Bolt AI debuted their first AI agents in 2023, while Replit released its agent in 2024. Those tools have had over a year to iterate, gain a following and develop brand loyalty. I personally have been experimenting with Replit agents for the last several months. Just this week, the company brought the technology to its mobile app — which you wouldn’t think is a big deal, but it is. The ability to use a simple prompt, without the need for a full desktop setup to build software, is powerful. Replit’s agent also provides AI prompt tuning to help generate the best possible code. The Replit system runs entirely in the cloud and users like me don’t need to download anything.  Bolt doesn’t have a mobile app, but it does have a really nice web interface that makes it easy for beginners to get started. Cursor is a bit more bulky in that it involves a download, but it is a powerful tool for professional developers. So how does GitHub Copilot agent mode compare? GitHub is the de facto standard for code repositories on the internet today. More than 150 million developers, including more than 90% of the Fortune 100 companies, use GitHub. According to the company, more than 77,000 organizations have adopted GitHub Copilot. That makes the technology very sticky. Those organizations already relying heavily on GitHub and Copilot are not going to move away from the technology easily. In

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Thomson Reuters is bringing AI to tax professionals in a big way. The company has partnered with Anthropic to use its Claude AI technology in its tax tools, marking one of the largest AI rollouts in the tax and accounting industry. At the heart of this initiative is CoCounsel, Thomson Reuters’ AI platform for legal and tax professionals. The system runs on Amazon’s secure cloud infrastructure, ensuring that sensitive client information remains protected while delivering AI-powered insights. “We combine real expert human knowledge with advanced technology,” Joel Hron, CTO at Thomson Reuters, said in an exclusive interview with VentureBeat. “We have experts across many different domains generating content and workflows. For us, AI is a tool to facilitate the distribution of that expertise through our software.” How Thomson Reuters built a tax AI platform using 150 years of professional content The company has built a comprehensive retrieval-augmented generation (RAG) architecture that connects Claude to Thomson Reuters’ vast knowledge base, including content from more than 3,000 subject matter experts and 150 years of professional publications. Rob Greenlee, head of industries at Anthropic, explained the technical approach in an exclusive interview: “Claude’s foundation in understanding complex professional domains like law and tax comes from comprehensive training on a diverse range of high-quality texts, including professional and academic content. For work with Thomson Reuters, we’ve taken several additional steps… We then work closely with Thomson Reuters to optimize Claude’s performance through advanced prompting strategies and carefully designed workflows that leverage their authoritative content and domain expertise.” Inside the strategic deployment of multiple AI models for professional services Thomson Reuters is strategically deploying different versions of Claude based on task complexity. The company uses Claude 3 Haiku for rapid processing tasks and Claude 3.5 Sonnet for deeper analyses requiring detailed insights. Early results show significant efficiency gains. “Customers are reporting transformative efficiency gains with CoCounsel,” said Hron. “Professionals are not only saving time but, also elevating the level of work they focus on, maintaining quality while delivering more strategic value to their clients.” Security remains paramount in the implementation. Amazon Bedrock provides what Hron called “a robust and battle-tested cloud infrastructure that adheres to our enterprise-grade security standards throughout the entire life cycle.” Enterprise AI deployment sets new standard for security and professional trust The collaboration between Thomson Reuters and Anthropic represents a new model for enterprise AI deployment, combining advanced AI capabilities with domain expertise and secure infrastructure. “What makes this partnership particularly valuable is the combination of Anthropic’s advanced AI capabilities with Thomson Reuters’ deep domain expertise and authoritative content,” said Greenlee. Looking ahead, Thomson Reuters plans to expand its use of Claude, exploring agent frameworks for complex tax workflows and computer vision capabilities to help editorial teams curate content more efficiently. “We’ve been vocal about our AI investment as a strategic part of our products going forward,” said Hron. “Our editorial workforce spends significant time building and curating content — we see tremendous potential to accelerate these processes with Anthropic’s computer vision and tool use capabilities.” The implementation comes as tax and accounting professionals increasingly adopt AI tools to streamline their work. Thomson Reuters’ approach could serve as a blueprint for other enterprises looking to deploy AI while maintaining professional standards and data security. Correction: Feb. 11, 2025: An earlier version of this article misstated Thomson Reuters’ use of Claude AI. The technology is implemented specifically for tax services within CoCounsel, not for legal services. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More CodeSignal, which is known for its technical skills assessment platform, is making a significant pivot into leadership and communication training with a new AI-powered conversation simulation tool. The move represents a strategic expansion beyond the company’s core technical assessment business, which has attracted clients like Netflix, Capital One, Meta and Dropbox. The San Francisco-based company’s new offering uses voice-enabled AI to simulate workplace scenarios, allowing users to practice difficult conversations like delivering feedback or resolving conflicts. The system provides real-time coaching through an AI mentor named Cosmo. “Leadership training often costs $20,000 to $40,000 per person for a one-month program,” Tigran Sloyan, CEO of CodeSignal, said in an exclusive interview with VentureBeat. “For a 10-person executive team, you’re looking at a multi-million dollar price tag, which means it’s never been accessible to anyone below the executive level.” The platform’s rapid growth suggests strong market demand for accessible skills development tools. CodeSignal’s learning platform has attracted one million users in less than a year since launch, with user base and usage doubling every two months, according to Sloyan. The technology leverages recent advances in generative AI and voice models to create realistic conversational partners. Early feedback suggests the AI-powered approach may have advantages over traditional role-playing with human actors. CodeSignal’s new AI conversation simulator presents users with workplace scenarios and a 5-minute window to practice difficult management conversations, guided by Cosmo, an AI mentor. (Credit: CodeSignal) How AI simulation outperforms human role-playing for leadership training “The feedback has been higher on average than with human role-playing,” said Sloyan. “When you’re doing real actors, usually it’s a workshop with other people watching. The fact that you have no judgment with AI and can practice as many times as needed takes the pressure off.” The company is targeting the underserved middle management segment, where traditional leadership development programs have been cost-prohibitive. The solution is priced at $25 monthly for individuals and $39 per user monthly for enterprise licenses. Beta customers have deployed the platform for management training, sales training and interviewer preparation. “We don’t want sales people practicing on actual customers,” Sloyan explained. “Now they can practice customer scenarios with AI first.” Market expansion and mobile learning: What’s next for CodeSignal The expansion comes as businesses increasingly focus on developing leadership capabilities across organizational levels. CodeSignal appears well-positioned to capitalize on this trend, with plans to launch a mobile version later this year to enable on-the-go learning. “Skills are changing faster than ever, driven by transformative technologies like AI,” said Sloyan. “This idea of lifelong, constant learning is more important than it has ever been before.” The move into leadership development could help CodeSignal capture a larger share of corporate training budgets while diversifying beyond technical assessments. The company faces competition from established leadership development firms and new AI-powered learning platforms, but its combination of technical expertise and AI capabilities may provide competitive advantages in the rapidly evolving market for workplace skills development. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Adobe is expanding its push into AI with new features that aim to demystify complex contracts and legal documents for both businesses and consumers, as the company seeks to maintain its dominant position in the document management market. The software giant announced today that its Acrobat AI Assistant can now automatically detect contracts, summarize key terms and compare differences across multiple versions — capabilities that Adobe says could help address a widespread problem: Most people don’t fully read agreements before signing them. According to Adobe’s research, nearly 70% of consumers have signed contracts without understanding all the terms. The problem extends into the business world, too, where 64% of small business owners have avoided signing contracts due to uncertainty about the content. “Control F is dead,” said Lori DeFurio, a product executive at Adobe, referring to the traditional way people search documents. “Why would I ever search again when I can just ask?” The shift from keyword searching to conversational AI reflects Adobe’s broader vision for making complex documents more accessible to everyone. Credit: Adobe How Adobe’s AI actually reads your contracts The new features represent a significant enhancement to Adobe’s AI capabilities — but notably stop short of providing legal advice. Instead, the system acts more like an intelligent research assistant, helping users locate and understand important contract terms while providing clear citations back to source material. “This is not a replacement for legal advice,” Michi Alexander, VP of product marketing at Adobe, emphasized in an exclusive interview with VentureBeat. “This is just to help you understand as a starting point your contracts, and where you potentially might want to ask questions.” The technology works by analyzing contract text and presenting information in more digestible formats. For example, users can compare up to 10 different contract versions in a table that highlights specific changes. The system can also process scanned documents, even if they’re wrinkled or imperfectly captured. A key differentiator, according to Adobe executives, is the system’s ability to provide specific citations for its analyses. “The answer AI assistant gives you is your guide on where in the document you should find the answer,” said Alexander. Your documents are safe, Adobe says: Inside the security architecture As AI features become more prevalent in enterprise software, questions about data security take center stage. Adobe emphasizes that all contract analysis happens in a transient fashion — documents are processed in the cloud but aren’t stored or used to train AI models. “Your data is always your data,” Lori DeFurio, a product executive at Adobe, explained during a product demonstration for VentureBeat. “We do not look at any of the documents that you don’t tell it to…your content is never used to train AI models.” The feature integrates into Adobe’s existing Acrobat ecosystem, which the company says serves more than 650 million monthly active users. It’s available for an additional $4.99 monthly fee for individual users, with enterprise pricing available for larger organizations. The real-world impact: How businesses are already using contract AI The release comes at a time when companies are increasingly looking to AI to streamline operations. According to Adobe’s survey, 96% of technology leaders believe AI would make them more confident in employees’ ability to handle contracts responsibly. In interviews the company performed with three dozen early users, most reported cutting their contract review time by 70 to 80%. “I used to spend 45 minutes on initial contract reviews,” said Austin Bailey, a real estate development executive who has been testing the feature since January. “Now I typically finish in under 10 minutes.” While Adobe isn’t the first company to apply AI to contract analysis, its massive user base and deep integration with existing document workflows could give it an advantage in the growing market for AI-powered business tools. The move also reflects a broader trend of traditional software companies embedding AI capabilities into their core products, rather than treating them as standalone features. For Adobe, which has invested in AI development for more than five years, the strategy appears to be paying off — the company reports that customer conversations with its AI assistant doubled quarter-over-quarter in late 2024. The future of contract analysis may increasingly rely on AI assistance, but human judgment remains crucial. As Alexander puts it, the tool is meant to “guide you to the parts of the document” rather than replace careful review entirely. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More App Orchid, an AI company specializing in enterprise data analytics, has announced an expanded collaboration with Google Cloud that aims to help businesses more easily extract insights from their corporate data using natural language queries powered by Google’s Gemini AI models. The partnership centers on integrating App Orchid’s technology with Google Cloud’s Cortex Framework, which helps organizations manage data from systems like SAP and Salesforce. The combination allows business users to ask questions about their data in plain English and receive AI-generated analytics and insights. “For too long, organizations have struggled to get answers to business critical questions in a timely manner,” Wes Kapsa, chief revenue officer at App Orchid, said in an exclusive interview with VentureBeat. “With App Orchid and Google Cloud Cortex Framework, customers can get instant, timely and accurate insights.” Knowledge graphs and natural language: A new approach to enterprise data The company’s approach differs from competitors by using what it calls an “ontology-driven framework” that creates knowledge graphs of enterprise data. This allows AI models to better understand complex business information and relationships. “It’s really hard to teach a language model how to understand database data, because as a human, it’s hard for us to understand database data as well, Rehan Refai, VP of solutions at App Orchid, explained in an interview with VentureBeat. “What we’ve developed is a method where you can access cortex data using our product called Easy Answers.” According to Refai, the company’s text-to-SQL accuracy rate of 99.8% significantly outperforms competitors who achieve around 90% accuracy. The system also provides detailed explanations for how it arrives at answers — a critical feature for enterprises concerned about AI hallucinations. Breaking down data silos with AI-powered analytics The partnership represents a strategic expansion for App Orchid, which has historically focused on specific industry verticals like manufacturing and logistics. “This is our first endeavor in an offering that is more horizontal,” noted Kapsa. The collaboration comes as enterprises increasingly look to democratize access to their data while maintaining accuracy and trust. App Orchid claims its technology can reduce data preparation time by up to 85% compared to traditional methods. AI agents and predictive analytics: What’s next for enterprise data Looking ahead, App Orchid is developing new capabilities, including probabilistic ontologies that can predict likely outcomes, and integration with emerging AI agent architectures. “Agents are the next big thing with LLMs,” said Refai. “We are one of the top choices when it comes to if you want to have an agent that’s an LLM player that wants to understand structured data.” The Easy Answers application is available now on the Google Cloud Platform Marketplace. While financial terms of the partnership were not disclosed, it represents a significant validation of App Orchid’s approach to making enterprise data more accessible through AI. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More While the AI market has in recent days seemed to collapse around DeepSeek and OpenAI, there are of course many other teams of brilliant engineers fielding large language models (LLMs) that are worth a look as a user or enterprise seeking to leverage the latest and greatest. Take Mistral AI, the French startup that made headlines even before it launched with a record-setting seed funding round for Europe of $640 million, and which has quietly been training and releasing a mix of open source and proprietary models for consumers and enterprises. Even as the rise of new reasoning models and agents have dominated the AI landscape recently, Mistral is still positioning itself as a viable alternative to OpenAI’s signature chatbot ChatGPT and DeepSeek’s hit mobile app, especially for those concerned with data privacy and security. Today, Mistral finally launched its own free, mobile version of its chatbot Le Chat for iOS and Android, as well as a new Enterprise tier for private infrastructure, and a Pro plan at $14.99 per month, the move suggests that Mistral AI is making a concerted push to convince companies there are worthwhile alternatives to DeepSeek and OpenAI. Mistral’s le Chat offers business leaders an AI tool that integrates with enterprise environments, operates with high-speed performance, and—importantly for some customers—does not send user data to China, unlike DeepSeek. Mistral targets both consumer and enterprise with savvy new releases Mistral AI’s latest rollout comes at a time when enterprises are increasingly evaluating AI partners based on data privacy, security, and deployment flexibility. The launch of le Chat’s Enterprise tier, which allows businesses to deploy the assistant on private infrastructure, SaaS, or virtual private cloud (VPC), suggests that Mistral is targeting the same corporate users who may have previously considered OpenAI’s GPT-4 or Anthropic’s Claude but want more control over their data and models. Mistral’s strategy mirrors a recent move by DeepSeek, a Chinese AI company that released DeepSeek-R1, a powerful reasoning model that offers capabilities and performance similar to OpenAI’s “o” series of models (o1, o1-mini, o3-mini out now with o3 full to follow soon) but at a fraction of the lost (30 times less expensive for enterprise users than OpenAI o1). However, DeepSeek’s expansion has been met in the West with privacy and security concerns related to China’s data retention and censorship laws. Some analysts have raised questions about whether AI models developed by Chinese firms could be subject to Beijing’s data access regulations, prompting enterprises to proceed cautiously when integrating such systems. For companies concerned about where their AI models process and store data, Le Chat’s non-Chinese infrastructure could be a key selling point. Unlike DeepSeek, which operates under a Chinese legal framework, Mistral AI is a European company based in France, subject to EU data privacy laws (GDPR) rather than China’s Cybersecurity Law or the Personal Information Protection Law (PIPL). Mistral AI is betting that Le Chat’s performance advantages will also help it stand out. The mobile app is powered by the company’s latest low-latency AI models, which, according to Mistral, enable “Flash Answers”—a feature that generates responses at speeds of up to 1,000 words per second. Beyond speed, Le Chat differentiates itself by integrating real-time web search and sourcing from journalistic and social media platforms, allowing for fact-grounded responses rather than relying solely on pre-trained knowledge. This makes Le Chat a potential alternative for businesses that require more up-to-date, evidence-based AI insights rather than static model training data. For enterprises, Le Chat also includes: • Code Interpreter: Allows in-place execution of scripts, scientific computations, and data visualization. • OCR & Document Processing: Industry-grade optical character recognition (OCR) for PDFs, spreadsheets, and even complex or low-quality images. • Image Generation: Powered by Black Forest Labs’ Flux Ultra, enabling photorealistic content creation. Undercutting OpenAI and Anthropic on price Mistral AI is also taking a different approach to pricing compared to competitors. While OpenAI charges $20 per month for ChatGPT Plus and Anthropic’s Claude has varying pricing based on token limits, Le Chat’s Pro plan starts at $14.99 per month. Additionally, most features—including the latest models, document uploads, and even image generation—are free, with limits only kicking in for power users. For businesses looking at team-wide adoption, Le Chat Team provides priority support, unified billing, and integration credits, while Enterprise deployments allow companies to use their own custom AI models tailored to their organization’s needs. Quick hands-on comparison I downloaded and tested Mistral’s Le Chat iOS app on my iPhone briefly while writing and editing this piece, and compared some of my prompts to my default AI assistant, OpenAI’s ChatGPT powered by GPT-4o. Le Chat was typically noticeably faster in its outputs than ChatGPT, but its Black Forest Labs Ultra model image generation capabilities were surprisingly not as adherent to my prompt as ChatGPT’s built-in connection to OpenAI’s DALL-E 3 image model, which is now 5 months old and hardly state-of-the-art anymore. Also, OpenAI’s connectivity to web search provided a more rich diversity of sources than Le Chat, which defaulted to the AFP, a French yet English-language publishing news outlet and wire service that Mistral partnered with back in January 2025. See some of my comparisons of Le Chat and ChatGPT below. Le Chat: ChatGPT: Le Chat: ChatGPT: AI competition continues to intensify Le Chat’s launch underscores a broader industry shift: while OpenAI and Anthropic remain dominant players, enterprises are actively evaluating alternative AI providers that offer better pricing, more flexible deployment options, and clearer data privacy guarantees. With DeepSeek facing scrutiny over its Chinese data links and OpenAI dealing with ongoing enterprise adoption challenges, Mistral AI’s European positioning, fast performance, and competitive pricing could make it an increasingly attractive choice for businesses looking to integrate AI assistants into their workflows. For companies weighing their AI options, the latest iteration of Le Chat is a signal that viable non-U.S., non-Chinese AI alternatives are beginning to