Tech News
← Home  ·  All topics

Ai

5511 GoKawiil briefs on this topic · RSS for ai

OpenAI's computing chief to leave amid $750 billion infrastructure plan

OpenAI has announced the departure of its top computing executive, the person overseeing the company's infrastructure buildout. The company has outlined plans to spend roughly $750 billion on computing capacity through 2030 to support its AI operations.

Former OpenAI engineer Mohammad Alfozan charged over hidden recording device

According to a federal court complaint, Alfozan allegedly used an AI assistant to research meeting room schedules and queried a chatbot linked to his spouse about building and remotely accessing a recording device. CCTV reportedly placed him near a conference room used for confidential discussions on eight occasions in September, despite his having no authorization to be there. A hidden device was later found in that room.

Epic Games releases AI-coded preview launcher, claims 46% faster startup than Steam

Epic Games has launched a redesigned version of its game launcher that the company says was built using AI, or "vibe coded." Epic claims the new launcher starts up 46% faster than Steam and offers a 30x faster library, along with new features like dedicated game pages, a unified downloads page, and collections for organizing titles. The launcher is in a preview/open beta phase, with some social and account management features still missing, and Epic advises Unreal developers and competitive Fortnite and Rocket League players to stick with the old version for now.

Ukrainian drone strikes hit three Yandex data centers, disrupting Russian services

Ukrainian drones struck three data centers operated by Russian tech giant Yandex over four days, knocking facilities offline and disrupting services relied on by banks, retailers and consumers. One targeted site, in Sasovo, held tens of thousands of servers along with two supercomputers used to train Yandex's AI models.

Deezer to add toggle for filtering out AI-generated music

Deezer announced a new feature that will let users hide AI-generated tracks, albums, and related artists from search results, with a rollout expected before year's end. CEO Alexis Lanternier cited user demand and a company survey showing roughly 45% of customers wanted a way to exclude AI music from their listening experience.

Wharton study: managers split on who's accountable for AI agent errors

A Wharton School survey of over 1,000 business leaders found no consensus on who should be blamed when AI systems make consequential mistakes. Roughly a quarter pointed to the employee operating the tool, another quarter to that employee's manager, and a similar share to the company's risk team, while about 10% blamed the AI vendor and 10% were unsure. The study also cited an incident where a faulty AI-driven email bot sent false refund and cancellation notices to more than 18,000 customers at a major U.S. retailer within six hours.

Zenzap raises $44M to build AI-powered workplace messaging app

Zenzap, founded by Guy Weiss with co-founders Noam Neumann and Jonatan Bien, has raised $44 million from Bessemer, Lightspeed, The Garage and individual backers including Wix president Nir Zohar. The startup builds a private, IT-compliant messaging app with embedded AI meant to replace consumer tools like WhatsApp and iMessage for business communication. This marks Weiss's first time raising outside capital after previously bootstrapping and selling StuccoMedia in 2015.

Floqer to open-source its AI sales research platform in early 2027

B2B sales startup Floqer plans to open-source the technology behind its autonomous customer knowledge database, pending security reviews, with release targeted for early 2027. The company, pitching at TechCrunch Disrupt 2026, says it has surpassed $1.2 million in annual recurring revenue after roughly tenfold revenue growth over the past year. Founded by Shivam Mahajan and Zaaheda Islam, Floqer began as a Chrome extension called 'Super Chat' before pivoting to automate CRM data cleanup for enterprise clients.

Character.AI files for Chapter 11 bankruptcy amid mounting losses

Character.AI has filed for Chapter 11 bankruptcy protection, according to court documents reviewed by 404 Media. The filings show the company spent $56.1 million on compute costs and $19.7 million on salaries between January and August, while monthly revenue grew only from $2.8 million to $5.3 million and operating losses ran $8-9 million per month. CEO Colin Adams acknowledged the company's technology was expensive to run and that it struggled to turn a profit.

40 state AGs flag data risks in Character.AI's Chapter 11 sale to Disney

New York AG Letitia James and a bipartisan coalition of roughly 40 state attorneys general filed a notice in Character.AI's bankruptcy case, warning that any sale or transfer of its assets must address consumer protection and privacy concerns. Character.AI filed for Chapter 11 this week and is selling some assets to Disney. The filing discloses that several states had been investigating the company since 2024 over child safety and privacy issues, following a lawsuit tied to a 14-year-old's suicide after interacting with the chatbot.

AI tools accelerate game decompilation, enabling fast ports and mods

AI systems are now able to reverse-engineer compiled video games and software back into source code far faster than manual efforts, which previously took years and required large community projects like OpenGOAL and Ship of Harkinian. The resulting decompiled code can be modified, enhanced with features such as higher frame rates and dynamic resolution, and recompiled to run natively on other platforms.

AWS engineer tests encoder vs decoder designs for Strands Decider model

Marc Brooker, an AWS engineer, responded to questions about why the strands-decider-2B model uses a modified decoder (LLM) architecture rather than a bidirectional encoder like BERT. He ran comparisons across five model designs, including the existing Qwen3.5-based decoder (v19), two encoder-decoder variants built on T5Gemma 2 at different parameter sizes, and encoder-only models based on T5Gemma. The post presents these as empirical tests of whether bidirectional encoders outperform causal decoders for this classification-style task.