Jeff Dean Leaves Google to Launch AI Startup Discovery Loop
Jeff Dean Leaves Google to Launch AI Startup Discovery Loop The landscape of artificial intelligence is shifting as one of Google's most iconic figures, Jeff Dean, departs the sea…
GyaanSetu AI
AI, ujifunzaji wa mashine na maarifa ya LLM.
Meta Launches Muse Code, a Terminal AI Agent That Writes and Tests Whole Features
Muse Code runs on Meta’s Muse Spark model and uses a fan-out system of sandboxed sub-agents to plan, generate, and validate code directly in the terminal, promising enterprise teams faster feature delivery without merge conflicts.
$67M Series B Fuels Omilia’s Push into U.S. Customer Support Market
Omilia’s CEO Dimitris Vassos says the company’s “knife-not-bazooka” approach lets it automate routine queries without costly LLMs, a strategy that helped lift ARR from $6M to $60M and win clients like Capital One, Discover and over 1,000 Taco Bell locations. The new capital will fund a U.S. office,
OpenAI Inapinga Kufutwa kwa Kesi ya Siri za Biashara ya Apple, Ikidai Udhaifu wa Usalama wa iCloud wa Apple
Ombi la OpenAI linadai kuwa Apple haikuwahi kulinda data za vifaa zinazodaiwa kwa sababu wahandisi walihifadhi faili kwenye akaunti binafsi za iCloud na wafanyakazi wa zamani walibaki na ufikiaji wa mfumo, jambo linalodhoofisha madai ya siri za biashara.
MIT Study Finds AI Explanations Help Laypeople but Trip Up Doctors
The MIT experiment showed non-medical participants improved diagnostic scores by deferring to AI heatmaps, while primary-care physicians reported confusion and reduced confidence when the same explanations clashed with their clinical reasoning.
AI Agents Now Launch Companies Solo: Naïve’s API Automates LLC, Stripe, QuickBooks
The Series-A-backed startup lets developers send a single prompt to spin up an LLC, email, virtual card, Stripe and QuickBooks, with only a final human KYC step.
Mapato ya AI ya Microsoft Yanategemea OpenAI kwa 70%, Yenye Thamani ya $24.1B
Ripoti ya Bloomberg inaonyesha kuwa dola bilioni 24.1 – takriban 70% ya mapato ya AI ya Microsoft kwa mwaka wa kifedha unaoishia Juni – yalitokana na mkataba wake na OpenAI, hata wakati kampuni hiyo ikilenga mauzo ya kila mwaka ya AI ya dola bilioni 37 na kupanua mkusanyiko wake wa mifano ya AI.
Suno Inaongeza Alama za Maji Zisizoonekana na Vikomo vya Upakuaji ili Kuzuia Spam ya Muziki wa AI
Mfumo mpya wa kampuni hiyo changa unaweka nambari iliyofichwa katika kila wimbo uliotengenezwa na AI na unaweka kikomo cha idadi ya faili ambazo watumiaji wanaolipa wanaweza kuzitoa kila mwezi, hatua inayolenga kuzuia mbinu za kupakua kwa wingi ambazo zimezidi majukwaa ya utiririshaji.
Virusi Zilizoundwa na AI Zinajizidisha kwa Haraka Zaidi Kuliko Phage za Asili, Stanford Yaonyesha
Stanford na Taasisi ya Arc walitumia modeli ya Evo kutengeneza vinasaba 285 vya virusi, ambavyo 16 vilitengeneza bacteriophages zinazojizidisha kikamilifu—baadhi zikizidi kasi ya kielelezo cha phiX174—zikionyesha uwezo wa AI katika kuunda uhai nje ya miundo ya asili.
OpenAI Inalenga Soko la Spika Janja za Kifahari kwa Kifaa chenye Umbo la Dona cha $300-$400
Kifaa hicho, ambacho ni spika ya chuma yenye umbo la dona na sehemu zinazojisogeza, kinatengenezwa kwa kushirikiana na studio ya LoveFrom ya Jony Ive na kinatarajiwa kuzinduliwa mwaka 2027, kikionyesha hatua ya kwanza ya OpenAI katika uingiliaji wa soko la vifaa vya teknolojia vya hali ya juu.
Alibaba Inajaribu Mikataba ya Kugawana Mapato kwa Watumiaji Wakubwa wa Qwen
Alibaba itawaruhusu watumiaji wakubwa wa kibiashara kutumia uzito wa bure wa Qwen, lakini itawataka kusaini mikataba ya kugawana sehemu ya mapato yoyote yanayotokana na huduma za API, mifumo iliyoboreshwa, au bidhaa zinazotokana na hiyo, huku wapenzi wa teknolojia wakibaki bila vikwazo.
Green Dashboard Masked AI Hallucination: Support Bot Gave Fake 2FA Steps
An AI-driven support agent answered a 2FA reset request with steps that never existed. The HTTP call returned 200 OK and every metric stayed green, but a missing guard-rail let the model hallucinate a plausible-sounding solution.
Klarna Inakuwa Huduma ya Kwanza ya BNPL kwenye J.P. Morgan Payments, Ikitumia Jukwaa la Dola Trilioni 2
Masharti ya kifedha hayakuwekwa wazi, lakini makubaliano hayo yanabadilisha msimamo wa zamani wa JPMorgan wa kusita kushirikiana na makampuni ya fintech, na kumpa Klarna njia ya moja kwa moja ya kufikia mtandao mkubwa wa wafanyabiashara wa benki hiyo.
Google Cloud Launches Managed GKE Agent Sandbox to Isolate AI-Generated Code
The new GKE Agent Sandbox service provides a disposable Linux container for each AI agent, with optional warm-pool pre-creation and selectable isolation layers (standard, gVisor, or Kata) to keep rogue code from compromising the host.
Chipu Mpya ya OLIX Inaacha HBM Ghali, Ikiahidi Kupunguza Gharama za Umeme kwa Kazi za AI
Mzunguko huu uliongozwa na Fundomo na ulihusisha Arm, Hudson River Trading, mfuko wa Sovereign AI wa Uingereza, na mwanzilishi mwenza wa Netflix, Reed Hastings, huku Nick McKeown wa Stanford akiungana na bodi. OLIX inapanga kuanza usafirishaji wake wa kwanza katika nusu ya pili ya mwaka 2027.
Half of BIRD Text-to-SQL Benchmark Gold Answers Are Wrong, UIUC Study Finds
The University of Illinois Urbana-Champaign team manually examined 238 BIRD-dev failures and discovered that 52.8% contain annotation errors—from wrong SQL to malformed questions—meaning many models are penalized for correct queries.
Injini ya AI ya The RealReal Inasababisha Ongezeko la 22% la GMV hadi $617M katika Robo ya Pili
Mfumo mpya wa Athena unachunguza na kuainisha bidhaa zinazowasilishwa, huku injini ya bei inayobadilika ikichambua pointi 100 za data kwa kila bidhaa, jambo linaloongeza thamani ya wastani ya oda kwa 13% hadi $659 na kupanua faida ya EBITDA iliyorekebishwa kwa pointi 290 za msingi.
DeepSeek Yachuma Dola Bilioni 8, Ikiongeza Thamani Yake hadi Dola Bilioni 74 ili Kushindana na OpenAI na Google
Mzunguko huo wa dola bilioni 8, ukiongozwa na NetEase, JD.com na IDG Capital, utagharamia uwezo mpya wa vituo vya data, chip za AI za kizazi kijacho, na uwezekano wa kuorodheshwa katika soko la hisa la China, ikimfanya DeepSeek kuwa "bingwa wa kitaifa" katika mashindano ya silaha ya AI.
Mistral AI’s Shieldstral 3B Model Matches 20B Guard Accuracy With One Token
Shieldstral treats moderation as a Q&A task, letting policies be updated via prompts instead of costly retraining, and returns a 0-1 probability score so teams can fine-tune thresholds for auto-block, human review, or pass.
Wakala Huru Huandaa Mfumo Kamili wa RAG Ndani ya Dakika 20, Tayari kwa PR
Kwa kutumia mzunguko wa hatua tano kati ya binadamu na wakala, AI iliandaa mpango, ikapata idhini, iliunda OpenSearch 3.7, Ollama LLM ya ndani, zana za FastMCP, na skripti za Helm, kisha ikafungua pull request—yote bila hata mstari mmoja wa kodi uliandikwa na binadamu.
AI-Powered Gifting Assistant Boosts Etsy’s Avg Order Value
Etsy’s AI-driven discovery tools convert a sub-1% stream of ChatGPT-origin traffic into higher-spending shoppers, lifting average order value and adding 350,000 active buyers in Q2—the first growth in active buyers since 2023.
Ditto Raises $9.2 Million Seed to Scrap Swipes for AI-Curated Blind Dates
New Dating Apps Bet Swiping Is Broken Swiping is broken. Gen Z is tired of endless profiles and small talk. Current dating apps profit when you keep browsing. If you find a partne…
ABSeeker’s 4B-Param Model Beats Bigger Rivals Using Step-by-Step Credit
ABSeeker’s Answer-Backtracked Credit Assignment (ABC) rewrites credit assignment by scoring each action against extracted answer clues, letting a 4 billion-parameter model match or surpass much larger agents on benchmark tasks.
OpenAI’s Astra Generates Formal Proofs for Ten Classic Math Problems Using Lean 4
Astra’s three-phase pipeline—generation, exposition, and formalization—feeds Lean 4 code to a compiler, which either accepts the proof or returns errors that the model iteratively fixes, offering a concrete template for trustworthy AI-assisted reasoning.