OpenAI Creating A Recent Jobs Board

OpenAI is developing its first in-house AI chip with Broadcom, slated for 2026 and fabricated at TSMC, in a bid to cut back reliance on Nvidia GPUs and control spiraling compute costs. Broadcom’s CEO flagged a $10B AI infrastructure order from a “latest client,” almost definitely OpenAI, signaling industrial-scale deployment.

The corporate can be preparing the OpenAI Jobs Platform, an AI-driven hiring service to rival LinkedIn, launching mid-2026. It’s going to use machine learning to match candidates with employers and features a dedicated track for small businesses and governments. Through OpenAI Academy, the corporate will even pilot “AI fluency” certifications in late 2025 with Walmart as a key partner and a goal of certifying 10 million Americans by 2030.

OpenAI’s push now extends beyond tech. Recently, It has subpoenaed advocacy groups Encode and CANI over funding, calling it transparency. Critics, nonetheless, call it intimidation. Some filings verge on absurd, similar to linking a nonprofit leader to Elon Musk via a property owned by “Tesla Place, LLC,” later revealed to belong to an unrelated retired couple. At the identical time, its Super PAC, “Leading the Future,” has now raised over $100M to beat back on regulations it argues could slow U.S. AI dominance.

On the patron side, OpenAI has opened up ChatGPT Projects, once paywalled, to free users, with tiered file uploads and customization nudging upgrades.

Google is now turning devices and apps into fully capable AI platforms. EmbeddingGemma, a 308M-parameter model built on Gemma 3 and trained in 100+ languages, now tops the Massive Text Embedding Benchmark for sub-500M systems. Designed for on-device RAG and semantic search, it leverages Matryoshka Representation Learning to flex between speed and precision, while integrations with llama.cpp, LangChain, LMStudio, and Ollama make deployment frictionless.

On the patron side, Google Photos now lets U.S. users turn still images into four-second silent clips with Veo 3, offering “Subtle movements” or random “I’m feeling lucky” effects. Once locked behind the $250/month AI Ultra plan, it’s now free with limited each day access.

Meanwhile, Gemini’s Nano Banana has driven 200M+ image edits and, since its launch last week, was liable for 10+ million first-time Gemini app users. The model, now integrated across AI Studio, the Gemini prompt bar, and social media, has quickly develop into a pacesetter on LMArena’s Image Edit leaderboard, outperforming competitors by a 171-point margin.

Android features now include on-device Gboard AI editing, Quick Share with live previews, LE Audio group syncing, and generative Androidify avatars. Flow video generation expands interactive storytelling via Flow Sessions, supported by recently joined filmmaker-in-residence Henry Daubrez.

RoboBallet, built by UCL, Google DeepMind, and Intrinsic, is popping robotic arms into precision production-line assemblers. Using reinforcement learning and graph neural networks, it maps tasks and obstacles as network nodes, letting fleets of arms solve as much as 40 jobs in seconds, even in layouts it’s never seen.

Traditional programming can take weeks for a handful of robots; RoboBallet scales immediately, generating plans tons of of times faster than real-time while adapting to layout changes or failures on the fly. Current focus is welding, however the architecture can extend to pick-and-place, painting, or strict sequence workflows. Open-sourced code positions it as a brand new benchmark for autonomous industrial robotics. Read more.