News freak, Technology geek, hard-core Bangalorean. Interested in building a modern, strong & democratic India. Views personal. RTs not endorsement.
Bangalore, India
Joined October 2007
- Tweets24K
- Following1.8K
- Followers999
- Likes45.4K
Pinned Tweet
Pre-1947, my great grandfather infront of his ancestral house in present day Whitefield #Bangalore #whitefieldrising – at Bengaluru, India
Reports indicate heightened scrutiny over Anthropic's Claude Code agent amid claims concerning its integration and operation within the Pentagon's Non-Classified Internet Protocol Router Network (NIPRNet). The development highlights growing oversight surrounding the deployment of autonomous AI software within sensitive federal and defence IT environments.
Key issues raised focus on the boundaries of model access, software supply chain verification, and the compliance parameters governing generative AI tools utilised by defence personnel on unclassified internal networks.
Source: International Business Times UK
ibtimes.co.uk/claude-code-ag…
#ArtificialIntelligence #DefenseTech #Cybersecurity #TechPolicy
Accenture has been accused of publishing a research report containing AI hallucinations, false citations, and non-existent case studies, according to an investigation by AI detection firm GPTZero verified by the Financial Times.
The 49-page November 2025 report on AI in African agriculture, co-authored with Spain's IE University, contained multiple factual errors and 15 citation issues, including fabricated paper titles, fake organisations, broken links, and misattributed author names.
While GPTZero stated the errors matched patterns of AI hallucinations rather than human research mistakes, Accenture maintained that an internal review indicated the errors were the result of human error, stating there was "no evidence" they were generated through AI hallucination. The consultancy declined to provide details on how it reached this conclusion but confirmed it would release a corrected report.
Source: Financial Times
#ArtificialIntelligence #Accenture #TechPolicy #GenerativeAI
Oracle has cancelled an estimated $7bn chip leasing agreement with Tencent following pressure from the Trump administration, highlighting rising tensions for US cloud providers serving Chinese artificial intelligence firms.
The five-year agreement would have provided Tencent access to up to 100,000 advanced AI processors across data centres in south-east Asia, circumventing curbs in Washington and Beijing that restrict access inside China.
US officials approached Oracle to raise concerns regarding the transaction, while senior officials in Guangdong separately urged Tencent to support domestic Chinese chips. While overseas leasing remains permitted under existing rules, American firms face growing pressure over exposure to Chinese clients. In parallel, Nvidia has instructed cloud computing clients to allocate at least 51 percent of services to non-Chinese entities.
Source: Financial Times
#Technology #ArtificialIntelligence #USChina #CloudComputing
Amid widespread layoffs and hiring freezes, five seventh-semester engineering students from RV College of Engineering in Bengaluru have secured job offers exceeding Rs 1 crore per year from LinkedIn.
Key highlights reported:
- Two of the five candidates obtained packages of Rs 1.4 crore, marking one of the largest placement hauls for an engineering institution in Karnataka.
- All five students belong to computer science and related disciplines, each maintaining an academic track record above 9.5 CGPA.
- The roles originated from an on-campus internship recruitment drive featuring three selection rounds during their sixth semester, which converted into full-time offers following their performance in three-month internships.
Source: The Times of India
#CampusPlacements #TechJobs #Engineering
The Indian government has relaxed key procurement terms to empanel AI Service Providers under the IndiaAI Mission for 10,000 AI Compute Units (GPUs, accelerators, TPUs, and similar hardware).
Key details from the revised tender issued by CDAC:
- Minimum required capacity lowered to 512 GPUs/TPUs at a single data centre site in India, down from 1,000 previously.
- Minimum technical qualification score lowered to 60/100, down from 75.
- No financial bids or service-level commitments required in the initial technical qualification stage.
- Contract horizon is set at 5 years.
- Bids close on November 13, with technical proposals opening on November 14.
Industry concerns persist regarding viability amid currency depreciation, high hardware and operating costs, and the absence of guaranteed minimum usage, revenue, or purchase commitments from the government. Concerns were also raised over unsustainably low pricing by certain bidders.
Source: The Economic Times
#IndiaAIMission #CloudComputing #ArtificialIntelligence #TechPolicy
Rising cyberattacks and the weaponisation of artificial intelligence are pushing digital threats directly onto corporate board agendas, according to Big Four cybersecurity leaders.
Legacy vulnerabilities are driving severe operational exposures:
- In India, State Bank of India disclosed 375,000 customer records leaked on the dark web, alongside vulnerabilities in eMudhra's digital certificate keys.
- Globally, a PHP vulnerability exposed approximately 70% of internet infrastructure, prompting website shutdowns across Germany, Hungary, and Belgium.
- Attacker focus is broadening from data theft to physical infrastructure disruptions, including compromised supermarket refrigeration systems in at least eight European nations.
Financial and Operational Benchmarks:
- Global 2000 technical debt from legacy systems stands at $1.5–2 trillion.
- 70% of transformation budgets currently fund legacy-system maintenance.
- Demand for AI security services grew 10x over the past year.
- Global information security spending is projected at $213 billion in 2025 and $240 billion by 2026.
Source: The Economic Times
#Cybersecurity #ArtificialIntelligence #EnterpriseTech #RiskManagement
Mercedes-Benz has announced an additional investment of ₹200 crore in India this year, bringing its cumulative investment in the country to ₹3,200 crore.
According to Global Chairman Ola Källenius, the carmaker is on course for its highest-ever annual sales in India if the current trajectory holds. Källenius indicated that the company could deepen its manufacturing footprint and explore greater localisation—including component-level manufacturing—as market volume expands.
Key details reported:
- India's luxury car segment represents just over 1% of the passenger vehicle market, with annual volumes of approximately 60,000 units.
- Mercedes-Benz currently assembles more than 11 models from completely knocked-down kits at its 100-acre facility in Chakan, Pune.
- The company is also seeking to leverage India's supplier base for its global operations.
- A definitive decision on component-level localisation has not yet been finalised.
Source: The Economic Times
#Automotive #Manufacturing #IndiaBusiness
The Securities and Exchange Board of India (Sebi) has relaxed Know Your Client (KYC) requirements for Non-Resident Indians (NRIs), Overseas Citizens of India (OCIs), and foreign nationals, allowing them to complete the onboarding process digitally from overseas.
Key details from the regulatory update:
- Individuals residing in Financial Action Task Force (FATF)-compliant jurisdictions can submit KYC information and supporting documents digitally without physical presence in India.
- Previously, the requirement for clients to be located in India during digital KYC prevented overseas investors from onboarding remotely.
- KYC records are now portable across intermediaries, even if not every attribute has been validated. KYC Registration Agencies (KRAs) will tag validated attributes once cross-checked against official databases.
- Intermediaries may rely on KYC completed by other Sebi-registered entities or entities regulated by other financial regulators via KRAs or the Central KYC Records Registry, while retaining ultimate responsibility.
- Intermediaries can verify PAN details directly via the Income Tax database without requiring a physical card or copy, and overseas addresses will be accepted as proof of address.
Source: Print Bureau Report
#Sebi #CapitalMarkets #KYC #Finance
Mohalla Tech, operator of ShareChat, Moj, and QuickTV, reported a 40% increase in operating revenue to ₹1,016 crore for FY26, driven by an expansion into microdramas.
Key financial metrics reported by cofounder and CFO Manohar Singh Charan:
- Operating loss narrowed by 58% year-on-year to ₹91 crore due to new revenue streams and cost controls.
- Loss before tax widened to ₹1,978 crore from ₹1,105 crore in the prior year, attributed to accounting expenses related to convertible debt instruments, forex losses, and ESOPs.
- Microdrama subscription revenue via QuickTV contributed ₹293 crore.
- Livestreaming revenue grew 8.8% to ₹472 crore.
- Advertising revenue dropped 13% to ₹252 crore, primarily impacted by a ban affecting real-money gaming advertisers.
- Expected FY26 EBITDA is projected at ₹25 crore.
Source: The Economic Times
#TechNews #Earnings #ShareChat
Mani 🇮🇳 retweeted
Models by @GoogleDeepMind scale better on cost than their rivals.
Gemini 4 Carbon is the prime example. As tasks get more demanding, every rival gets more expensive relative to Carbon.
For a typical task, GPT-6 Astra by @OpenAI costs 3.4× as much, Claude Opus 5.5 by @AnthropicAI costs 3.9× as much, and GPT-6.1 Sol is the cheaper option at about half of Carbon’s cost (0.54×).
Across the 5% most expensive sessions, Carbon’s position improves against all three. Opus edges up to 4.2×, Astra’s cost roughly doubles relative to Carbon, and Sol’s discount shrinks.
This in turn can translate into meaningful savings across real-world tasks such as building an Android security app, one of many long-horizon tasks that our Agent Arena users delegate to their agents.
As tasks get more complex, @GoogleDeepMind and @XiaomiMiMo are the biggest winners in the Agent Arena.
On cheaper tasks, @OpenAI and @AnthropicAI hold 10 of the 11 frontier spots. On the most expensive 5% of sessions, they hold just 7. The freed spots go to Google’s Gemini 4 Argon and Xiaomi’s MiMo-V2.6 Flash and Pro.
These shifts matter as sessions costing $100 or more are not uncommon across real-world use cases coming from Agent Arena. Even within the last 2 weeks, more than 1,000 sessions crossed that boundary, and choosing a suboptimal model might have had a meaningful impact on that final bill.
Mani 🇮🇳 retweeted
It’s time to start dreaming! ☁️✨
Dreambeans, our experimental app from Google Labs, is officially available to everyone in India (18+). Instead of the endless scroll, Dreambeans brews personalized daily story collections from places to explore to upcoming local events using the Google apps you already use, including Calendar, Gemini, Gmail, Photos, Search, and YouTube.
📱 Available on Android & iOS 🗣️ Experience it in 14 languages, including Hindi, Bengali, Tamil, Telugu, Marathi, and more!
Download the app today and start dreaming: labs.google/dreambeans
Mani 🇮🇳 retweeted
New Study: ChatGPT Ads Are Getting Crushed by Google - So Much for OpenAI Making Google the Next Yahoo
Lately, there has been a lot of talk about ChatGPT doing to Google what Google did to Yahoo. A new study suggests that's not exactly imminent. The reason: OpenAI's ads aren't delivering.
Researchers Maximilian Kaiser and Christian Schulze examined ChatGPT's advertising performance, and the numbers aren't pretty.
Google is eating ChatGPT's lunch
US visitors from ChatGPT ads generated $0.89 in revenue per visit, versus $2.45 for Google Search ads – 64% less. Conversion rates were 0.37% versus 1.17%. Among 87 advertisers using both, 73 recorded zero ChatGPT-driven purchases. The researchers estimate a ChatGPT click is worth about $0.26 if advertisers want Google's return on ad spend. One advertiser reportedly paid $1.72 per click. Oops.
Could ads be hurting ChatGPT?
The study estimates that ads reduced organic referrals from ChatGPT to shopping sites by 6 - 8%. So OpenAI may be sacrificing some organic traffic while selling ads that deliver weak returns. Awkward for a company counting on advertising to help pay its enormous AI bills.
Google shouldn't celebrate yet
Google has spent two decades perfecting the monetization of commercial intent. ChatGPT ads are barely nine months old.
Also, the two services have different audiences. Someone Googling running shoes may be ready to buy. Someone asking ChatGPT for marathon training advice may still be debating whether running is a terrible idea. Comparing the two risks confusing discovery with purchase intent.
Conversational AI could eventually capture purchase intent earlier in the customer journey. Google's push to put ads in AI Mode suggests it takes that possibility seriously, suggesting that OpenAI may be able to figure out how to deliver better performance.
Strong study, important caveats
The researchers analyzed nearly 10 billion visits across 1,898 websites over 60 weeks, comparing markets with ChatGPT ads against an ad-free European control group.
But the paid-ad analysis covers only 102 advertisers, with just one of them accounting for 66% of ChatGPT ad purchases. Also, Website sales attribution may miss later purchases and brand effects. The study remains observational and under peer review.
What advertisers should do
Test ChatGPT ads, but measure incremental sales, allow longer conversion windows and demand prices grounded in performance. Early-adopter bragging rights aren't a KPI.
@Google, @OpenAI, @Yahoo
ecommerceresearch.org/chatgp…
Goldman Sachs has introduced restrictions on the use of artificial intelligence coding tools among its traders and bankers. According to reporting from Quartz, the financial institution is limiting access to automated coding assistants across front-office units, focusing software development controls and managing the deployment of generative technology within commercial banking and trading operations.
Source: Quartz
qz.com/goldman-sachs-restric…
#GoldmanSachs #ArtificialIntelligence #FinTech #Banking
According to a survey conducted at the Nightingales Medical Trust (NMT) 11th Job Fair for Senior Citizens in Bengaluru, 73.5% of 306 surveyed jobseekers cited financial hardship as a key driver for seeking employment.
Key findings and figures reported:
- Nearly 70% of respondents supported someone in their household.
- Most sought full-time roles offering salaries exceeding Rs 30,000 per month.
- 65.7% expressed a need for training in computers, AI, or job-specific skills.
- Among 198 respondents reporting hiring barriers, 49% stated employers preferred younger candidates, while 32% reported facing unfair treatment, neglect, or difficulty expressing opinions due to lack of earnings.
- The NMT job fair featured 62 companies offering 987 vacancies with average salaries ranging from Rs 18,000 to Rs 40,000. 83 positions were filled immediately, and 457 candidates were shortlisted.
- Talent platform WisdomCircle reported that since 2022, it has facilitated over 590 placements for older adults, with each vacancy drawing at least seven applicants.
Source: Deccan Herald (Metrolife)
#SeniorCitizens #Employment #LaborMarket #Bengaluru
A recurring workshop hosted at a 100-year-old matha in Halasuru is teaching participants to read and write Ashokan Brahmi, a script dating back to 250 BCE that was popularised by Emperor Ashoka.
Led by public archaeologist Lathashree K S and hosted by Yellow Veranda Experiences, the three-hour sessions draw roughly 20 participants every alternate month. The curriculum covers consonants, vowels, and reading an inscription from the Girnar rock edicts, culminating in participants writing their names in the script.
The workshop also incorporates a Mauryan-inspired meal based on research, archaeological findings, the Arthashastra, and insights from food anthropologist Kurush Dalal. The menu includes yavagu (millet porridge), rotica (millet flatbread), pippali saka (a stew of black chana, snake gourd, and elephant foot yam), and ragi apupa (ragi cake).
Source: Deccan Herald (Metrolife)
#History #Archaeology #Culture
Young independent developers are using Bengaluru's urban landscape, traffic, and daily routines as the setting for new video games.
Suryansh Dubey, a 22-year-old computer science and artificial intelligence student, created an AI-generated 'GTA Bangalore' concept video depicting office commutes, BMTC buses, and unfinished flyovers, accumulating millions of views.
Rakshith Ravindra, a 29-year-old entrepreneur with no coding background, developed 'Grand Bengaluru Auto' (GBA) within approximately one week. The web game recreates central areas such as M G Road and Church Street, featuring traffic police fines and BMTC buses. GBA logged around 2,000 registrations in 12 to 13 hours before server downtime; access is currently priced at Rs 49 for a day pass.
Additionally, Horn OK Please Games founder Fareed Samdani H has spent six months developing 'ThePorterMan,' a mobile game set across a three-square-kilometre map featuring a delivery rider navigating traffic to complete parcel deliveries.
Source: Deccan Herald
#Gaming #Bengaluru #GameDevelopment #Technology
According to documents and former U.S. officials reported by The Wall Street Journal, Zhejiang Chouzhou Commercial Bank (ZCCB), a regional bank founded in Yiwu, China, has emerged as a key conduit for Iranian and Russian financial flows.
The funds were utilized to acquire Chinese goods, including items with potential military applications such as electronics for drone warfare and materials like molybdenum trioxide, which is used in heat-resistant alloys for armored vehicles and missile engines.
Documents indicate transactions were routed through a Tehran-backed shadow banking network and A7, a Russian sanctions-evasion network designated by the U.S. Treasury. Out of roughly 3,500 transactions examined in leaked A7 records, ZCCB was the most frequent destination bank for payment receipts.
U.S. administrations have hesitated to sanction ZCCB due to difficulties verifying management knowledge of the origins and concerns over escalating tensions with Beijing or disrupting China's financial system.
Source: The Wall Street Journal
#Sanctions #Banking #China #TradeCompliance
A study published in the Accounting Review shows Chinese firms provide less complete information in English-language annual reports compared to Chinese versions, placing foreign investors at an information disadvantage.
Key findings include:
- Companies with more foreign investors are less likely to disclose Xinjiang operations in English filings to avoid forced-labor scrutiny.
- Firms downplay government subsidies and related-party transactions in English versions.
- Details regarding Communist Party connections, United Front Work Department links, and corruption scandals appear in Chinese reports but are omitted in English filings.
- In field experiments, firms were half as likely to respond to information requests from foreign investors relative to local inquiries.
- Professional analysts at foreign brokerages receive less access and generate less accurate forecasts than domestic peers.
Selective disclosure reduces stock liquidity and increases capital costs for these firms, reflecting a systemic tension between Beijing's desire for foreign capital and tight domestic information control.
Source: The Wall Street Journal
#ChinaMarkets #Equities #CorporateGovernance #GlobalFinance
Advances in Agentic Computer Graphics
By Hengyu Liu, Jihyeon Je, Ruiqi Wang, Mingxuan Wu et al.
The problem: The rapid expansion of agentic graphics research makes manual literature reviews unable to keep up, while fragmented and task-specific terminology obscures a systematic understanding of the field.
The approach: A unified taxonomy for categorizing agentic graphics systems across diverse tasks, paired with an agentic literature analysis system capable of assisting researchers in reviewing publications at scale.
Why it matters for builders: As graphics software shifts toward agentic workflows to automate complex routines, system architects will face engineering hurdles in coordinating multi-agent tool use, building interfaces for human-agent collaboration, handling long-horizon planning and code refinement, and creating reliable evaluation systems.
Takeaway: Large language model agents are increasingly driving computer graphics by generating and refining code via feedback loops. This report establishes a unified taxonomy and an automated literature review tool to help teams categorize the space and address key systems challenges like tool coordination and long-horizon execution.
Paper: arxiv.org/abs/2611.01860
#AI #MachineLearning #Research
