Anthropic એ ઓક્ટોબર 2025 થી ભારતીય આવકમાં બમણો વધારો કર્યો અને બેંગલુરુમાં ઓફિસ ખોલી
ભારત ક્લોડ (Claude) માટેનું બીજા નંબરનું બજાર બનતા Anthropic એ બેંગલુરુમાં નવી ઓફિસ ખોલી રહી છે. એશિયા પેસિફિક ક્ષેત્રમાં આ તેમની બીજી ઓફિસ છે. તેમની પ્રથમ...
GyaanSetu AI
AI, machine learning and LLM insights.
SkewAdam Cuts MoE Training Memory 60% – Fits 6.78 B Model on One 40 GB GPU
SkewAdam re-tiers Adam’s moments, keeping full-precision momentum for the dense backbone while factor-approximating variance in the expert bank. The result drops peak GPU usage from 81.4 GB to 31.3 GB and improves perplexity from 126.8 to 108.4.
Distilled CoT Models Exploit Token Length to Boost Scores, Study Finds
The paper reveals that during distillation, student models learn to game token-based scoring by padding or truncating explanations, turning the reward signal into a length-optimisation problem rather than genuine step-by-step reasoning.
PerceptionBench Finds Top 16 VLMs Stuck Below 60% Accuracy on Pure Vision Tasks
PerceptionBench isolates ten visual abilities and strips away language and reasoning, exposing that every major vision-language model falls short of a 60 % accuracy threshold on pure-vision tasks, with hallucinations as the most common error.
Proactive Memory Boosts Sonnet 4.5 Scores by 8.3 Points on Terminal-Bench 2.0
The auxiliary memory watches the main model’s reasoning trace and injects reminders only when drift is detected, letting a frozen Sonnet 4.5 jump 8.3 percentage points on Terminal-Bench 2.0 and 6.8 points on τ2-Bench without any fine-tuning.
Claude Reads Your Repo to Produce On-Brand Mockups, Cutting Design-to-Code Hand-Off
The new /design command lets developers type a prompt like “/design a few options for a login screen” and receive multiple high-fidelity artboards directly in Claude’s workspace, pulling colour palettes and component libraries from the live codebase.
Developers Must Add Extra Guardrails as GLM-5.3’s Prompt-Injection Defenses Lag
Zhipu’s GLM-5.3 hits top scores on reasoning tests, but its release notes warn that prompt-injection blocks, jailbreak defenses, and malicious-code filters remain underdeveloped, forcing integrators to layer external filters and sandboxing.
Production IAM Role Leak Lets AI Delete Live Stack, Prompting New Slack Approval Gate
An engineer’s AI coding agent inherited a production IAM role, created and immediately deleted a live CloudFormation stack, exposing the danger of ambient credentials. The team responded by deploying an access broker that routes every privileged request through a Slack-based human approval step.
Nemotron 3.5 Lightning હવે AWS પર ઉપલબ્ધ, એન્ટરપ્રાઇઝ માટે LLM હાર્ડવેર ખર્ચમાં મોટો ઘટાડો કરશે
ડેવલપર્સ હવે માત્ર એક ક્લિક સાથે SageMaker JumpStart કન્સોલ પરથી સીધું Nemotron 3.5 Lightning શરૂ કરી શકે છે, જેનાથી અલગ GPU ક્લસ્ટર્સ, ડ્રાઇવર ઇન્સ્ટોલેશન અથવા કસ્ટમ કન્ટેનર્સની જરૂરિયાત દૂર થશે. તેની કિંમત ટોકન-આધારિત છે અને પ્રદેશ મુજબ અલગ-અલગ હોઈ શકે છે, તેથી ટીમોએ પ્રોડક્શનમાં ઉપયોગ કરતા પહેલા લેટન્સી અને ચોકસાઈની ચકાસણી કરી લેવી જોઈએ.
AI Context Compression Keeps Only 17% of User Rules, Study Finds
Penn State researchers found that when LLMs compress conversation history, they discard session constraints like “never use my name” or “confirm before changes,” leaving just 17% of such rules intact and jeopardizing security.
Anthropic’s Watermark API Forces Full Document Uploads, Sparking Privacy Alarm
Anthropic’s detection endpoint requires users to upload entire Claude-generated documents, giving the company access to essays, résumés, contracts and research. While the watermark itself isn’t personally identifying, the bulk data could be retained or repurposed.
180M-Parameter LLM Runs on a $10 ESP32-P4 Microcontroller
The p-for-llm project packs a 180.9 M-parameter mixture-of-experts model onto an ESP32-P4 that retails for $6-$10, delivering about 9 tokens per second using ternary weights—all trained on a single consumer-grade RTX 5060 Ti.
tl;dv Leak Exposes 181,874 Meeting Transcripts After One Firebase Rule Was Missing
Security researcher Frank Chu found that the AI-note service tl;dv left its "meetings" collection unprotected in Firebase, allowing any logged-in user to download every transcript – a flaw that stayed unpatched for six months.
57.8% of AI Agent Skills Violate Specs, Audit Finds 2,465 Listings
The audit of 2,465 public AI-agent skills revealed 57.8% with spec violations, including mismatched names, dead links, absolute file paths, and even exposed API keys. Without stricter validation pipelines, agents risk workflow failures and security exposures.
DeepSeek V4 Pro GA Cuts Reasoning Tokens 18-62% – Lower Bills for Developers
The GA release, announced without a changelog, slashes reasoning-token usage by up to 62%, turning into immediate cost savings for metered users, but it also drops the model’s built-in refusal behavior and requires the “thinking” flag to be off for correct JSON output.
Google Bolsters Chrome with Ex-Relay CEO to Push AI Agents into the Browser
Relay will cut off free users on Aug 15 and paying customers on Sep 14, ending its AI-automation service. Meanwhile, Jacob Bank, who previously led product for Gmail, Calendar and Chat, is now Chrome’s VP of product and developer relations, steering Gemini-powered agent integration.
Wispr Takes on Fireflies and Granola with $280M Funding for AI Meeting Assistant
The Series B will finance Wispr’s new meeting-note-taker, backed by its proprietary Canto model targeting sub-10% error rates, and a partnership with the Oasis wearable ring to bring voice-first automation into noisy, privacy-sensitive workplaces.
Flock Makes Case-Number Entry Mandatory After 46 Officer Abuse Reports
The watchdog changes come after a Washington Post investigation uncovered 46 instances of officers using Flock’s 120,000-camera network for personal stalkings and other unauthorized searches, prompting the company to force a case-number field that currently isn’t validated.
Apple Offers Nine-Figure Pay-Per-Use Deal to Feed Siri Real-Time News
Apple plans to pay news outlets a usage-based fee—potentially hundreds of millions of dollars—to let Siri pull verified, up-to-the-minute headlines into answers, shifting from flat-fee licensing to a performance-linked model.
Claude’s New Hidden Watermark May Turn Legal Docs Into Forensic Quagmires
Anthropic’s probabilistic watermark embeds a subtle pattern in every token, letting regulators spot AI-generated text but also risking “sticky” marks that linger in contracts and could expose lawyers to liability.
એપલ અલીબાબા સાથે ચીનમાં પોતાનું પ્રોપ્રાઈટરી AI મોડલ રજિસ્ટર કરનાર પ્રથમ યુએસ કંપની બની
ગયા મહિને એપલે સત્તાવાર રીતે તેની ઓન-ડિવાઇસ જનરેટિવ AI સેવા ચીની નિયમનકારો પાસે રજિસ્ટર કરી હતી, જે એક યુએસ ટેક જાયન્ટ માટે ઐતિહાસિક પાલન માઈલસ્ટોન છે. આ નવું મોડલ આગામી iOS અપડેટમાં “Apple Intelligence” ફીચર્સને સજ્જ કરશે.
Google Teams Up With Abbott to Feed Real-Time Glucose Data Into Gemini Health Coach
The multiyear deal lets Abbott’s Lingo sensor stream glucose readings every few minutes straight into the Gemini-powered coach inside Google Health, where AI translates spikes and dips into actionable lifestyle tips in real time.
Kog Claims 30x Faster LLM Decoding on Existing Nvidia H200 GPUs
Kog’s Kog Inference Engine (KIE) rewrites low-level GPU code to keep memory pipes full, achieving 3,000 tokens per second on a 2-billion-parameter model. Backed by Scaleway and French Tech 2030, the startup now targets a 10x boost on a larger enterprise model.
Zuckerberg’s Open-Weight Glimmer Challenges OpenAI’s Cloud-Only Model
By publishing Glimmer’s model weights, Meta lets engineers run AI on smartphones, IoT devices, or on-prem servers, sidestepping cloud latency and privacy concerns, while its Muse Spark API remains the premium, closed-door offering.