Latest news with #ArunChandrasekaran
Time of India
5 days ago
- Business
- Time of India
Zoho courts the enterprise CIO with its own LLM and agentic AI stack
Zoho , a Chennai-headquartered SaaS company, has entered the foundational model arena with the launch of its proprietary large language model , Zia LLM . The announcement, made on the sidelines of its annual Zoholics India conference in Bengaluru, signals a growing ambition to deepen its AI footprint—while maintaining its hallmark focus on privacy and affordability. 'The announcement emphasises Zoho's long-standing aim to build foundational technology focused on the protection of customer data, breadth and depth of capabilities because of the business context, and value,' said Mani Vembu, CEO of Zoho. 'Our LLM model is trained specifically for business use cases, keeping privacy and governance at its core, which has resulted in lowering the inference cost, passing on that value to the customers, while also ensuring that they can utilise AI productively and efficiently.' How can Zoho's move into building its own LLM be viewed then: A bid for strategic differentiation, or a necessary pivot as enterprises demand more sovereignty and cost control over AI? Arun Chandrasekaran, Distinguished VP Analyst at Gartner, framed it as the former. 'The move is aimed at creating a differentiation for business use cases, targeting efficiency and privacy for customers who can't afford or don't want to rely on large external LLM providers,' he said. A right-sized, In-house stack Built fully in-house and trained using NVIDIA's AI-accelerated computing platform, Zia LLM comprises three model sizes—1.3 billion, 2.6 billion, and 7 billion parameters. Each is optimized separately for distinct business contexts such as structured data extraction, summarisation, retrieval-augmented generation (RAG), and code generation. This tiered model strategy enables Zoho to balance performance and compute efficiency across user scenarios, a principle the company refers to as 'right-sizing.' It also gives Zoho the flexibility to scale Zia LLM gradually; the first round of parameter increases is expected by the end of 2025. 'By building its own AI stack , Zoho is hoping to appeal to businesses that care about domain-specific AI, AI integrated into business workflows while simultaneously being sovereign and cost efficient,' Chandrasekaran said. To that end, Zoho has launched a Model Context Protocol (MCP) server that opens its library of workflow actions to third-party AI agents. While customers can still integrate with external models like ChatGPT, Llama, and DeepSeek, Zia LLM gives them the option to keep their data within Zoho's environment—benefiting from the latest AI capabilities without sending sensitive information to third-party clouds. The model is now deployed across Zoho data centers in the U.S., India, and Europe. Building agents without code The company also debuted Zia Agent Studio, a no-code platform that lets businesses create AI agents embedded directly within Zoho applications. More than 25 pre-built agents are already available, including several tailored for Indian businesses. 'Our differentiation comes from offering agents over our low-code platform so that there is a human in the loop for verification and modification. It is much simpler to verify and make changes in the UI screen than reading the code,' Vembu said. These agents work across business functions. For instance, a Customer Service Agent in Zoho Desk can process customer queries, respond to common issues, or route complex requests to a human agent. Meanwhile, Zoho's AI assistant, Ask Zia, enables interactive conversations to build reports, analyse data, and assist data teams in creating machine learning models. Zoho also launched India-specific agents that can verify documents like PAN, Voter ID, Udyog Aadhar, GSTIN, Driving Licences, and utility bills—targeting use cases in HR and financial services, such as employee background verification or onboarding checks. Eyeing the enterprise, quietly While Zoho has long positioned itself as a champion of small and mid-sized businesses, its expanding AI stack, with built-in governance and observability tools, may signal broader ambitions. 'While Zoho will remain committed to SMBs, its AI platform capabilities, governance tooling, and agentic capabilities suggest it is testing the waters for enterprise traction. Perhaps they are eyeing upmarket expansion without abandoning affordability,' Chandrasekaran said. That shift could mean rethinking pricing, too. As CIOs evaluate Zoho's AI agents and LLM stack for enterprise deployment, traditional SaaS models may not apply, Chandrasekaran noted: 'CIOs will scrutinize data privacy , extensibility, grounding mechanisms, observability, and interoperability, especially with existing enterprise data and APIs. Also, Agentic AI is giving rise to new pricing models that can potentially challenge the seat-based pricing model of SaaS. How well Zoho can embrace new pricing models (such as usage-based or outcome-based pricing) will be critical for its success with CIOs.'
Time of India
28-05-2025
- Business
- Time of India
Top AI firms pivot to profitability track leaving price wars behind
With artificial intelligence companies such as OpenAI and Anthropic pausing steep price cuts of their generative AI models , several Indian startups may have to tap external funding to scale up their GenAI-based applications. OpenAI, Anthropic and Google — which reduced GenAI model pricing by 65%-90% in 2024 — are releasing new models at roughly flat or even higher rates. Experts said the cost of intelligence may still go down going forward, albeit at a slower rate because model companies are no longer hustling to train new models month after month. Instead, they are competing in the application layer with agents and enterprise AI. ETtech This is limiting the ability of Indian startups to scale AI applications, invest in R&D, and pass on cost savings to customers. Companies are shifting to more efficient and smarter AI usage to overcome the challenge while experts say most may have to tap external funding in the long term. Live Events "Cost is definitely a challenge right now, especially when scaling AI agents that require high inference loads, long context windows, memory, and tool use," said Somit Srivastava, CTO at a wealthtech platform. "Running production-grade agents at scale isn't cheap, and it impacts pricing strategies, performance tuning, and R&D investment decisions." Discover the stories of your interest Blockchain 5 Stories Cyber-safety 7 Stories Fintech 9 Stories E-comm 9 Stories ML 8 Stories Edtech 6 Stories He said has largely relied on open-source models which are free to use instead of paid alternatives. But there's a constant trade-off between performance versus viable cost-to-serve through open models. Many startups can't afford continuous improvement cycles without solid funding, said Arun Chandrasekaran, distinguished VP analyst at Gartner. Cutting-edge GenAI research and development (R&D) such as agent architecture, long-context models, and tool use, are compute-intensive and not easily subsidised by commercial revenues yet, he said. "Many teams are forced to 'build wrappers' instead of innovating, or to anchor themselves to public APIs to reduce infra lift," Chandrasekaran said. Experts said startups need to act early and design intelligently, and not just wait for prices to fall. Abhimanyu Singh, vice president-product at GenAI-based customer support platform said cost is becoming a primary scaling bottleneck, especially for agentic applications requiring multiple model calls.
CNET
05-05-2025
- CNET
OpenAI Pulled a Big ChatGPT Update. Why It's Changing How It Tests Models
Recent updates to ChatGPT made the chatbot far too agreeable, and OpenAI said it is taking steps to prevent the issue from happening again. In a blog post, the company detailed its testing and evaluation process for new models and outlined how the problem with the April 25 update to its GPT-4o model came to be. Essentially, a bunch of changes that individually seemed helpful combined to create a tool that was far too sycophantic and potentially harmful. How much of a suck-up was it? In some testing, we asked about a tendency to be overly sentimental, and ChatGPT laid on the flattery: "Hey, listen up — being sentimental isn't a weakness; it's one of your superpowers." And it was just getting started being fulsome. "This launch taught us a number of lessons. Even with what we thought were all the right ingredients in place (A/B tests, offline evals, expert reviews), we still missed this important issue," the company said. OpenAI rolled back the update at the end of April. To avoid causing new issues, it took about 24 hours to revert the model for everybody. The concern around sycophancy is not simply about the enjoyment level of the user experience. It posed a health and safety threat to users that OpenAI's existing safety checks missed. Any AI model can give questionable advice about topics like mental health, but one that is overly flattering can be dangerously deferential or convincing, like whether an investment is a sure thing or how thin you should seek to be. "One of the biggest lessons is fully recognizing how people have started to use ChatGPT for deeply personal advice — something we didn't see as much even a year ago," OpenAI said. "At the time, this wasn't a primary focus but as AI and society have co-evolved, it's become clear that we need to treat this use case with great care." Sycophantic large language models can reinforce biases and harden beliefs, whether they are about yourself or others, said Maarten Sap, assistant professor of computer science at Carnegie Mellon University. The large language model, or LLM, "can end up emboldening their opinions if these opinions are harmful or if they want to take actions that are harmful to themselves or others," he said. The issue is "more than just a quirk" and shows the need for better testing before models are released to the public, said Arun Chandrasekaran, a distinguished vice president analyst at Gartner. "It's a serious concern tied to truthfulness, reliability and user trust, and (the) updates from OpenAI hint at deeper efforts to address this, although the continued trend of prioritizing agility over safety is a concerning long-term issue," he said. (Disclosure: Ziff Davis, the parent company of CNET, in April filed a lawsuit against OpenAI, alleging that it infringed on Ziff Davis copyrights in training and operating its AI systems.) How OpenAI tests models and what is changing The company offered some insight into how it tests its models and updates. This was the fifth major update to GPT-4o focused on personality and helpfulness. The changes involved new post-training work or fine-tuning on the existing models, including the rating and evaluation of various responses to prompts to make it more likely to produce those responses that rated more highly. Prospective model updates are evaluated on their usefulness across a variety of situations, such as in coding and math, along with specific tests by experts to experience how it behaves in practice. The company also runs safety evaluations to see how it responds to safety, health and other potentially dangerous queries. Finally, OpenAI runs A/B tests with a small number of users to see how it performs in the real world. The April 25 update performed well in these tests, but some expert testers noted the personality seemed a bit off. The tests didn't specifically look at sycophancy, and OpenAI decided to move forward despite the issues raised by testers. Take note, readers: AI companies are in a tail-on-fire hurry, which doesn't always square well with well thought-out product development. "Looking back, the qualitative assessments were hinting at something important and we should've paid closer attention," OpenAI said. Among its takeaways, the company said it needs to treat model behavior issues the same as it would other safety issues and halt a launch if there are concerns. For some model releases, the company said it would have an opt-in "alpha" phase to get more feedback from users before a broader launch. Is ChatGPT too sycophantic? You decide. (To be fair, we did ask for a pep talk about our tendency to be overly sentimental.) Katie Collins/CNET Sap said evaluating an LLM based on whether a user likes the response isn't necessarily going to get you the most honest chatbot. In a recent study, Sap and others found a conflict between the usefulness and truthfulness of a chatbot. He compared it to situations where the truth is not necessarily what people are told: Think of a car salesperson trying to sell a flawed vehicle. "The issue here is that they were trusting the users' thumbs-up/thumbs-down response to the model's outputs and that has some limitations because people are likely to upvote something that is more sycophantic than others," Sap said, adding that OpenAI is right to be more critical of quantitative feedback, such as user up/down responses, as they can reinforce biases. The issue also highlighted the speed at which companies push updates and changes out to existing users, Sap said, an issue not limited to one tech company. "The tech industry has really taken a 'release it and every user is a beta tester' approach to things," he said. A process with more testing before updates are pushed to users can bring such issues to light before they become widespread. Chandrasekaran said more testing will help because better calibration can teach models when to agree and when to push back. Testing can also let researchers identify and measure problems and reduce the susceptibility of models to manipulation. "LLMs are complex and non-deterministic systems, which is why extensive testing is critical to mitigating unintended consequences, although eliminating such behaviors is super hard," he said in an email.
CNN
19-03-2025
- Business
- CNN
AI is getting better at thinking like a person. Nvidia says its upgraded platform makes it even better
Nvidia on Tuesday revealed more details about its next artificial intelligence chip platform, Blackwell Ultra, which it says will help apps reason and act on a user's behalf – two capabilities that would take AI beyond chatbots further into real life. Blackwell Ultra, which the company detailed at the its annual GTC conference, builds on Nvidia's existing sought-after Blackwell chip. The additional computing power in the new Ultra version should make it easier for AI models to break complicated queries down into multiple steps and consider different options – in other words, reason, the company said. Demand for AI chips has surged in the wake of OpenAI's ChatGPT in 2022, fueling a massive surge in Nvidia share prices. Its chips fuel the data centers that power popular, power-hungry AI and cloud-powered services from companies like Microsoft, Amazon and Google. But the arrival of Chinese tech startup DeepSeek – whose R1 model sent shockwaves through Wall Street for its reasoning capabilities and supposedly low cost – sparked speculation that expensive hardware may not be necessary to run high-performing AI models. Nvidia, however, appears to be skirting such concerns, as evidenced by its January quarter earnings in which it breezed past Wall Street's expectations. Nvidia wants its chips to be central to the types of reasoning models that the Chinese tech startup helped popularize. It claims a DeepSeek R1 query that would have taken a minute and a half to answer on Nvidia's previous-generation Hopper chip would only take 10 seconds with Blackwell Ultra. Cisco, Dell, HP, Lenovo and Supermicro are among the companies working on new servers based on Blackwell Ultra. The first products with Blackwell Ultra are expected to arrive in the second half of 2025. Being able to reason, or think through an answer before responding, will allow AI apps and agents to handle more complex and specific types of questions, experts say. Instead of just spitting out an answer, a chatbot with reasoning abilities could dissect a question and provide multiple, specific responses accounting for different scenarios. Nvidia cited an example of using a reasoning model to help create a seating arrangement for a wedding that takes into account preferences such as where to sit parents and in-laws and ensuring the bride is seated on the left. 'The models are now starting to mimic a little bit of human-like behavior,' said Arun Chandrasekaran, an analyst covering artificial intelligence for market research firm Gartner. And it's not just DeepSeek and OpenAI creating models that can reason. Google also updated its Gemini models with more reasoning capabilities late last year, and Anthropic introduced a hybrid reasoning model called Claude 3.7 Sonnet in February. Some experts also believe reasoning models could pave the way for 'AI agents,' or AI assistants that can take actions rather than just answering questions. Companies like Google, Amazon and Qualcomm have been vocal about their visions for AI-powered helpers that can do things like book a vacation for you based on your preferences rather than just churning out answers to questions about flights and destinations. 'What agentic AI excels at is multitasks,' said Gene Munster, managing partner at Deepwater Asset Management. 'And being able to reason in each of those tasks is going to make the agents more capable.'
CNN
18-03-2025
- Business
- CNN
AI is getting better at thinking like a person. Nvidia says its upgraded platform makes it even better
Nvidia on Tuesday revealed more details about its next artificial intelligence chip platform, Blackwell Ultra, which it says will help apps reason and act on a user's behalf – two capabilities that would take AI beyond chatbots further into real life. Blackwell Ultra, which the company detailed at the its annual GTC conference, builds on Nvidia's existing sought-after Blackwell chip. The additional computing power in the new Ultra version should make it easier for AI models to break complicated queries down into multiple steps and consider different options – in other words, reason, the company said. Demand for AI chips has surged in the wake of OpenAI's ChatGPT in 2022, fueling a massive surge in Nvidia share prices. Its chips fuel the data centers that power popular, power-hungry AI and cloud-powered services from companies like Microsoft, Amazon and Google. But the arrival of Chinese tech startup DeepSeek – whose R1 model sent shockwaves through Wall Street for its reasoning capabilities and supposedly low cost – sparked speculation that expensive hardware may not be necessary to run high-performing AI models. Nvidia, however, appears to be skirting such concerns, as evidenced by its January quarter earnings in which it breezed past Wall Street's expectations. Nvidia wants its chips to be central to the types of reasoning models that the Chinese tech startup helped popularize. It claims a DeepSeek R1 query that would have taken a minute and a half to answer on Nvidia's previous-generation Hopper chip would only take 10 seconds with Blackwell Ultra. Cisco, Dell, HP, Lenovo and Supermicro are among the companies working on new servers based on Blackwell Ultra. The first products with Blackwell Ultra are expected to arrive in the second half of 2025. Being able to reason, or think through an answer before responding, will allow AI apps and agents to handle more complex and specific types of questions, experts say. Instead of just spitting out an answer, a chatbot with reasoning abilities could dissect a question and provide multiple, specific responses accounting for different scenarios. Nvidia cited an example of using a reasoning model to help create a seating arrangement for a wedding that takes into account preferences such as where to sit parents and in-laws and ensuring the bride is seated on the left. 'The models are now starting to mimic a little bit of human-like behavior,' said Arun Chandrasekaran, an analyst covering artificial intelligence for market research firm Gartner. And it's not just DeepSeek and OpenAI creating models that can reason. Google also updated its Gemini models with more reasoning capabilities late last year, and Anthropic introduced a hybrid reasoning model called Claude 3.7 Sonnet in February. Some experts also believe reasoning models could pave the way for 'AI agents,' or AI assistants that can take actions rather than just answering questions. Companies like Google, Amazon and Qualcomm have been vocal about their visions for AI-powered helpers that can do things like book a vacation for you based on your preferences rather than just churning out answers to questions about flights and destinations. 'What agentic AI excels at is multitasks,' said Gene Munster, managing partner at Deepwater Asset Management. 'And being able to reason in each of those tasks is going to make the agents more capable.'
