AIPULSEN - AI News
microsoft openai
Microsoft logged $10 billion from OpenAI in Q1—about 11% of its quarterly revenue—and still has $7.2 billion in outstanding receivables from the partnership.
Microsoft disclosed that its commercial ties with OpenAI generated $10 billion in revenue for the quarter ending 30 September 2026, roughly 11 percent of the company’s total quarterly earnings of $92.3 billion. The filing also revealed that OpenAI owes Microsoft $7.2 billion in outstanding receivables, up from $6 billion just three months earlier. The same regulatory document noted that Microsoft’s equity stake in OpenAI slipped to 24 percent from 25 percent following the October 2025 recapitali
agents benchmarks claude gemini google
Google unveiled Gemini 4 Carbon, a frontier model that tops the Artificial Analysis Intelligence Index v5.0 and promises major advances in coding, agentic tasks, and cybersecurity.
Google unveiled Gemini 4 Carbon on Wednesday, branding it as the company’s most powerful frontier model and a “significant leap” in cybersecurity. The announcement highlights three core advances: higher‑quality code generation, stronger performance on agentic (autonomous) tasks, and state‑of‑the‑art capabilities across the cybersecurity defense lifecycle—vulnerability detection, validation and remediation.
The model’s debut is underscored by its top ranking on the Artificial Analysis Intelligen
apple meta
Meta has quietly released an AI‑powered pet journal app, letting users document their pets’ lives with generative AI features.
Meta’s newest AI experiment has slipped onto app stores with little fanfare. The company has released **Waddle World**, a free “tamagotchi‑style” pet app that doubles as a daily journal. According to the Apple and Google Play listings, users converse with a virtual creature by voice rather than typing; the AI then generates doodles that illustrate the chat and offers light‑hearted activities such as grocery‑stacking mini‑games and word puzzles. The app is currently limited to adults in the Phili
benchmarks gemini google
Google unveiled Gemini 4 Carbon, its latest frontier model and the company's strongest AI to date.
Google has unveiled Gemini 4 Carbon, the second public model in its Gemini 4 series and the company’s “strongest model with state‑of‑the‑art cybersecurity and long‑horizon capabilities,” according to the firm’s October 28 announcement. The release follows Gemini 4 Argon, announced on 30 September 2026 and opened to paying users on 13 October, positioning Carbon as a modest upgrade that pushes the frontier further.
The new model immediately topped version 5.0 of the Artificial Analysis Intellige
bias training
Researchers evaluate the impact of truncating backpropagation through time on looped transformers, which reuse a shared layer block to boost depth without extra parameters.
A new scaling study has examined whether truncating back‑propagation through time (BPTT) is a worthwhile shortcut for “looped” transformers – models that recycle a single block of layers across multiple forward passes to deepen computation without adding trainable parameters.
The researchers trained almost 600 variants, systematically varying model size, token budget, forward‑loop count and the depth of BPTT. By fitting a single scaling law that captures the trade‑off between compute, loop co
agents multimodal training
A new approach called Zooming without Zooming v2 aims to boost spatial perception and multimodal long‑context understanding by linking detailed observation to evidence for more generalizable AI.
A team of researchers has released a new arXiv paper, **“Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception,”** together with open‑source code. The work builds on the earlier “Zooming without Zooming” approach, which used iterative “thinking‑with‑images” to zoom into regions of interest during inference. The v2 system, dubbed **ZwZ‑v2**, eliminates the need for runtime zooming tools by training on full‑image inputs while still learning
embeddings gemma multimodal
EmbeddingGemma 2, an open‑source multimodal embedding model, has been released under the Apache 2.0 licence. Built on the Gemma 4 architecture, the sub‑1 billion‑parameter model maps text, code, images, audio and video – and any combination of these modalities – into a single 768‑dimensional space. The authors highlight two technical refinements: a simplified per‑layer embedding scheme that reduces overhead, and a multi‑stage curriculum that progressively introduces modalities and objectives. To
google
Google is rolling out AI Search ads worldwide as its core ad growth decelerates, announced alongside Alphabet's Q3 earnings showing 25% YoY revenue rise to $127.6 billion.
Google announced on its October 28 earnings call that it is rolling out advertisements within its AI‑powered Search experience to ten additional markets, including Canada and India. The move marks the first time the company has extended “AI Mode” ads beyond its initial test regions, effectively taking its generative‑AI monetisation footprint global.
The expansion arrives as Alphabet reported third‑quarter revenue of $127.6 billion, up 25 % year‑over‑year, while growth in Google Search advertisi
agents reasoning
Automated theorem proving faces a bitter lesson as formal proof work leans increasingly on human‑crafted workflows—planning, decomposition and specialized orchestration—rather than raw model scaling.
A new study shows that a minimalist agentic loop, dubbed **h1**, can push formal theorem‑proving performance to the current state‑of‑the‑art using only Lean 4 feedback and raw test‑time compute. Unlike recent advances in informal mathematics, which have leaned on ever larger language models and sophisticated chain‑of‑thought prompting, the h1 system discards the elaborate, human‑crafted pipelines that have become standard in formal verification. Instead, it relies on a simple execution interface
huggingface nvidia open-source
The U.S. Justice Department has asked how Nvidia's $12.93 billion acquisition of Hugging Face could affect AI competition, as antitrust probes into the chipmaker continue.
The U.S. Department of Justice has opened a formal inquiry into Nvidia’s $12.93 billion acquisition of the open‑source AI platform Hugging Face, asking how the deal could reshape competition in the fast‑moving generative‑AI market. The request arrives amid a broader antitrust sweep of Nvidia’s business practices, signalling that regulators are scrutinising the company’s expanding influence beyond its traditional graphics‑processing hardware.
Nvidia’s bid for Hugging Face marks a rare foray by a
chips funding startup
Seoul‑based AI chip startup FuriosaAI raised $620 million in a Series D round, boosting its valuation to $2.9 billion, with South Korea’s National Growth Fund contributing $280 million.
Seoul‑based AI‑chip designer FuriosaAI announced a $620 million Series D financing that lifts its valuation to $2.9 billion, a four‑fold increase from the $735 million level recorded in 2025. The round is anchored by South Korea’s National Growth Fund, which alone contributed $280 million, underscoring the government’s push to nurture a domestic semiconductor ecosystem capable of powering next‑generation artificial‑intelligence workloads.
The injection of capital comes at a time when global dem
funding startup
AI coding startup Cognition is weighing a new funding round before year‑end after offers raised its valuation to $70‑80 billion, up from $48 billion in September.
AI coding startup Cognition is weighing a fresh funding round before the end of the year after receiving investor offers that value the company at $70 billion to $80 billion – a steep jump from the $48 billion valuation it announced just weeks earlier in September. The new interest signals that capital markets remain eager to back firms that promise to accelerate software development with generative AI.
The rapid escalation in valuation matters for several reasons. First, it underscores the pre
Industry is exploring the balance between autonomous AI systems and human oversight to ensure safe, efficient operations.
A new approach to industrial automation is emerging that blends high‑level AI autonomy with direct human oversight. The concept, announced this week, proposes a layered control architecture where machine‑learning‑driven agents handle routine, high‑speed tasks while operators retain the ability to intervene, adjust parameters, or halt processes in real time. The framework is being piloted in manufacturing and logistics environments, allowing AI systems to make decisions within predefined safety e
Enterprises must align multiple AI platforms—from customer solutions to automation tools—with business context to unlock real value.
A new industry briefing underscores that the real lever for extracting value from AI in large organisations is not the sheer number of sophisticated tools deployed, but the ability to feed each of them a consistent, enterprise‑wide context. The analysis points out that many firms now run parallel AI‑powered customer platforms, conversational assistants and workflow‑automation engines. Each system can respond in milliseconds, yet when they operate in isolation the outcomes—whether in speed, accur
copilot microsoft
Microsoft will discontinue support for its Copilot AI assistant on all smart TVs this year, ending the feature introduced on LG and Samsung models announced at CES 2025.
Microsoft has announced that it will cease support for its Copilot AI assistant on all smart‑TV platforms later this year. The decision affects the LG and Samsung televisions that were unveiled at CES 2025 with built‑in access to Copilot, a move that had been billed as a way to bring conversational AI to the living‑room. Windows Central reports that the functionality will be disabled across the board, ending the partnership’s consumer‑facing phase.
The pull‑back matters because it signals a shi
Instagram head Adam Mosseri warned that negativity toward AI is increasing, noting that dislike extends beyond students and creatives.
Instagram chief Adam Mosseri warned that the backlash against artificial intelligence is set to intensify. Speaking on the Colin and Samir podcast, Mosseri was asked about the recent wave of college graduates booing tech CEOs during commencement speeches and the broader “anti‑AI” sentiment that has been bubbling across campuses and creative circles. He said the hostility is not confined to students or artists; “a lot of people really don’t like AI,” he told the hosts, and the mood is likely to g
nvidia speech voice
Nvidia has launched a free speech‑to‑text model that can transcribe recordings in up to 40 languages, aimed at meeting and medical transcription while real‑time use stays with rivals.
Nvidia has rolled out a free, open‑source speech‑to‑text model that can transcribe recordings in up to 40 languages. The tool is designed for batch processing rather than live streaming, meaning it can be applied to pre‑recorded meetings, lectures or medical dictations but does not yet support real‑time captioning. By making the model publicly available, Nvidia aims to lower the barrier for developers building multilingual transcription services, a segment that has traditionally relied on propri
anthropic
Anthropic introduced tools that reveal how companies' AI spending translates into usage of its models, ahead of its planned IPO.
Anthropic has introduced a new suite of analytics tools that let corporate users see exactly how their spending translates into model usage and output. The “frontier lab” feature breaks down costs by model, task and performance tier, giving finance and product teams a transparent view of the AI services they consume. The rollout comes as the company prepares for a public offering, signaling a push for greater accountability ahead of the IPO.
The move matters because enterprises have so far been
MicroZig announced it will enforce a strict policy prohibiting the use of large language models and artificial intelligence across its operations.
MicroZig announced that it will enforce a strict “no LLM / no AI” policy across its operations, barring the use of large language models and related artificial‑intelligence tools on company devices and internal systems. The decision, revealed in a brief corporate statement, marks a clear departure from the prevailing industry trend of integrating generative AI into everyday workflows.
The move matters because it underscores growing concerns about data security, model reliability and regulatory