Latest news with #DeedyDas
Time of India
27-05-2025
- Business
- Time of India
Sarvam AI unveils multilingual LLM; tepid response poses questions over India's AI chops
Indian artificial intelligence (AI) startup Sarvam AI , the first company picked by the government to build a sovereign foundational model under the India AI Mission , dropped its multilingual open-source large language model (LLM) Sarvam M last week. It claims the LLM has achieved benchmarks across mathematics, programming, and a range of Indian languages. But the launch has thrown up questions about the country's AI scene, with the model getting only a few hundred downloads in the first two days. Let's take a deep dive to understand why. About the model Sarvam M is a 24-billion-parameter model, built on top of French AI company Mistral's model Small. The startup said the model is "built for versatility" and designed to support a wide range of applications, including conversational agents, translation and educational tools. It supports 10 Indian languages, including Hindi, Bengali, Gujarati, Kannada and Malayalam. Live Events — SarvamAI (@SarvamAI) "This is the first in a series of contributions as we help build out the Sovereign AI Ecosystem in India," the startup said in a post on X. 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 Key features Hybrid thinking mode: This mode supports complex logical reasoning, solves mathematical problems and handles coding tasks. It also has a 'non-think' mode for general purpose conversation. Indic skills: It is post-trained in Indian languages along with English to better suit local needs. Reasoning capabilities: The company claims Sarvam M outperforms similarly sized models on coding and math benchmarks. No traction In the first two days after the launch, the model had just over 300 downloads. As of May 27, it had 1,200 downloads on Hugging Face, can be tested on Sarvam AI's playground, and accessed through its APIs. This sparked a debate on social media platforms about the standing of India's AI scene when compared with global rivals OpenAI and China's DeepSeek. Social media backlash In a scathing post, Deedy Das, a venture capitalist at Menlo Ventures, hit out at "India's biggest AI startup $1B Sarvam" for a muted launch with only 23 downloads in 2 days. "In contrast, 2 Korean college trained an open-source model that did ~200K last month (sic)," his post on X read. — deedydas (@deedydas) Das added in another post that he had nothing against Sarvam or India trying to build AI, but was "disappointed at their direction". Many came to the defence of the model. "Before Sarvam M came out, people were complaining about the lack of IndicLLMs. After Sarvam M came out, people are still complaining," one user wrote. Another person said we must offer encouragement, not shame entrepreneurs. "Teams in India are trying and they are working to build the muscle and culture of what good or amazing looks like." — svembu (@svembu) Sridhar Vembu of Zoho also jumped to the defence of Sarvam, saying that there was no such thing as an "instant hit". "Even when we were the first mover in a new market and we had done a lot of technical work, we only got slow traction. Instant success is neither necessary nor sufficient to succeed long term," he said in a post on X, telling the Sarvam team to "fight the good fight". Sarvam's trajectory The Bengaluru-based startup was the first to be selected to build an indigenous foundational model under the Rs 10,000 crore IndiaAI Mission. It was provided access to 4,096 Nvidia H100 graphics processing units (GPU) for six months from the mission's common compute cluster to train its model. According to Tracxn, the company has so far raised $53.6 million in total and is valued at $111 million.
Miami Herald
13-05-2025
- Business
- Miami Herald
U.S.-based Microsoft to ax roughly 6,000 jobs in global workforce
May 13 (UPI) -- Microsoft said Tuesday that nearly 3% of its workforce is being cut as other tech giants have recently acted similarly. "We continue to implement organizational changes necessary to best position the company for success in a dynamic marketplace," a Microsoft spokesperson said in a statement to CNBC and GeekWire. Microsoft reported better-than-expected financial returns last month with more than $25 billion in quarterly net income. As of June 2024, Microsoft had 228,000 reported employees around the globe and 126,000 U.S.-based workers. It's turning out to be likely Microsoft's biggest round of job cuts since 2023, when 10,000 people were laid off. But according to a venture capitalist, at around $200,000 a person, that sheds roughly $1.4 billion a year for Microsoft in this new round of cuts. "CS college graduates can't find jobs anyway, and it's getting worse," Deedy Das, a former Google engineer and principal partner at Menlo Park, Calif.-based Menlo Ventures, posted Tuesday afternoon on X. He said Microsoft grew ~10% a year from 2012-2022 but "has effectively plateaued" in the three years since. A Microsoft spokesman added that one objective has been to reduce layers of management and that this new round of cuts were not related to a smaller, performance-based set of layoffs in January. CFO Amy Hood said on the company's April 30 earnings call that Microsoft was focused on building "high-performing teams and increasing our agility by reducing layers with fewer managers." Meanwhile, stock shares in Microsoft ended Monday in the day's trading at its highest price this year so far at $449.26 after last July's record at $467.56. Nearly 60,000 tech employees have been laid off at 127 companies so far this year, according to It arrived on top of last week's announcement that software provider CrowdStrike will lay off about 5% of its own workforce. Meanwhile, several economists told USA Today how the wave of tech industry layoffs could indicate that American consumers habits shifted since the COVID-19 pandemic. Copyright 2025 UPI News Corporation. All Rights Reserved.
Yahoo
06-04-2025
- Business
- Yahoo
OpenAI's New Image Generator Is Incredible for Creating Fraudulent Documents
OpenAI's latest image-generating 4o model is surprisingly good at generating text inside images, a feat that had proved particularly difficult for its many predecessors. And that makes it a powerful tool for generating images of fraudulent documents, as users have found. Case in point, Menlo Ventures principal Deedy Das tweeted a photo of a fake receipt for a lavish meal at a real San Francisco steakhouse, as spotted by TechCrunch. "You can use 4o to generate fake receipts," Das wrote. "There are too many real-world verification flows that rely on 'real images' as proof. That era is over." The image itself, at first blush, is pretty convincing and includes a breakdown of a multicourse meal, a correct subtotal, and even a tip calculation. Another user even managed to edit the image further by adding a realistic filter and food stains — the perfect way to commit expense fraud, if you were so inclined. And that's the tip of the iceberg. Das also found that 4o was happy to generate fraudulent prescriptions for controlled substances like Zoloft. The development highlights how far AI-powered image generators have come. Previous models infamously struggled with recreating letters, often resulting in garbled shapes and unintentionally hilarious phrases. Beyond faking expenses for lavish meals, OpenAI's increasingly canny ability to generate fake documents could open up the door for everything from phony tax forms and bank cheques to fake IDs and birth certificates. Whether our ability to detect this storm of faked documents remains to be seen. But given AI companies' current efforts, it's not looking good. Guardrails like appended metadata or watermarks that divulge whether an image was generated by an AI are easily overcome. Even before the advent of powerful AI-powered image generators, a 2015 survey found that 85 percent of respondents admitted to lying to get reimbursed with more money. Many of these kinds of fraud cases fall through the cracks due to a lack of internal controls and flawed accounts payable processes. In other words, you can no longer believe anything you see online. More on image generators: Lawyer Says Studio Ghibli Could Take Legal Action Against OpenAI
Yahoo
31-03-2025
- Business
- Yahoo
ChatGPT's new image generator is really good at faking receipts
This month, ChatGPT unveiled a new image generator as part of its 4o model that is a lot better at generating text inside images. People are already using it to generate fake restaurant receipts, potentially adding another tool to the already-extensive toolkit of AI deepfakes used by fraudsters. Prolific social media poster and VC Deedy Das posted on X a photo of a fake receipt for a (real) San Francisco steakhouse that he says was created with 4o. Others were able to replicate similar results, including one with food or drink stains to make it look even more authentic: The most real-looking example TechCrunch found was actually from France, where a LinkedIn user posted a crinkled-up AI-generated receipt for a local restaurant chain: TechCrunch tested 4o and was also able to generate a fake receipt for an Applebee's in San Francisco: But our attempt had a couple of dead giveaways that it was faked. For one, the total uses a comma instead of a period. For another, the math doesn't add up. LLMs still struggle to do basic math, so this isn't particularly surprising. But it wouldn't be hard for a fraudster to quickly fix a few of the numbers with either photo editing software or, possibly, more precise prompts. It's clear that making it really easy to generate fake receipts presents huge opportunities for fraud. It wouldn't be hard to imagine this kind of tech being used by bad actors to get "reimbursed" for entirely fake expenses. OpenAI spokesperson Taya Christianson told TechCrunch that all of its images include metadata indicating they were made by ChatGPT. Christianson added that OpenAI "takes action" when users violate its usage policies and that it's "always learning" from real-world use and feedback. TechCrunch then asked why ChatGPT allows people to generate fake receipts in the first place, and whether this is in line with OpenAI's usage policies (which ban fraud.) Christianson replied that OpenAI's "goal is to give users as much creative freedom as possible" and that fake AI receipts could be used in non-fraud situations like "teaching people about financial literacy" along with creating original art and product ads. This article originally appeared on TechCrunch at Sign in to access your portfolio
