LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-11-11104 publications

Actualiser

Databricks Reports Genie Ontology Outperforms Frontier Agents by 1.6x Accuracy with Half the Tokens

Matériel
2026-11-11 23:00 GMT+8

Big data platform company Databricks released new benchmark results on November 11, 2026, comparing its Genie product—powered by the Genie Ontology context layer—against two major frontier AI agents. The benchmark was developed in collaboration with Turing to evaluate performance on business questions within a live Databricks workspace.

According to the report, Genie achieved a 91% pass rate, compared to 45% and 58% for the two frontier agents. Additionally, Genie used less than half the number of tokens and completed tasks three to four times faster. Databricks attributes this superior performance to Genie Ontology's ability to automatically extract and organize business knowledge from notebooks, dashboards, and SQL queries, providing AI agents with shared enterprise context.

The evaluation utilized a fictional UK energy company scenario, Calderwick Energy, featuring hundreds of tables and complex analytical assets. This marks the first time Databricks has published shareable quantitative benchmark data, addressing previous gaps where claims about Genie One MCP's real-world answer quality lacked specific verification. It is important to note that these results are self-reported by the vendor and have not yet been independently reproduced.

Sources :Databricks

Recherche

a16z Partners Olivia Moore and Justine Moore: AI Agents Have Become the Internet's Primary Users; Companies Must Compete for 'Machine Customers'

Sujet · B2A经济Structurel
2026-11-11 23:02 GMT+8

In an article published in November 2026, a16z partners Olivia Moore and Justine Moore present a judgment: the primary users of the internet are no longer humans, but AI agents. They refer to the resulting economic form as the 'B2A Economy'—Business-to-Agent, meaning businesses selling to agents. For decades, websites and platforms invested significant resources in preventing bot access; now, they must reverse course and compete for the business of these 'machine customers.'

The core mechanism of this argument is that as agents shift from answering questions to executing tasks, they will increasingly decide which products users employ, which services they subscribe to, and where they spend money. The authors draw parallels to historical platform shifts—the Web gave businesses a storefront accessible to anyone, while mobile put that storefront in everyone’s pocket. AI has already changed what people can do, but it is now changing how businesses reach users. Agents possess deeper user need data than traditional advertising and should theoretically serve as a stronger referral source than websites.

The data supporting this judgment comes primarily from company self-reports. Shopify stated that AI-driven traffic to its stores grew eightfold year-over-year in Q1 2026, with orders from AI-powered searches increasing nearly thirteenfold. Human Security observed that agentic browser traffic grew eightfold between August and October, with Muse accounting for 60% of activity and OpenAI products comprising a combined 30%. GitHub Octoverse data shows that as of September 2026, over 40% of GitHub Pull Requests began with an agent draft, compared to just 17% six months prior. The authors note that AI-guided consumer traffic exhibits higher interaction rates, better conversion, and lower bounce rates.

Coding agents serve as a leading case study for this transformation. Although developers constitute less than 1% of the global workforce, agent tools have fundamentally altered their workflows. Supabase reported that since June 2026, over 60% of new databases were created by AI tools; Neon stated that automated agents deployed 80% of its new databases; Stripe reported that agents accounted for nearly 40% of documentation traffic and 70% of API resource CLI requests; Vercel noted that agent-triggered deployments grew 28-fold within twelve months. The authors ask: What happens when the remaining 99% of the workforce begins using agents?

The article categorizes potentially benefiting enterprises into four types: agents themselves and agent models (such as ChatGPT, which has 1.2 billion weekly active users); products newly created by agents (analogous to Amazon for the internet or Uber for mobile); incumbent enterprises optimized for agents (expanding from Generative Engine Optimization/GEO to evaluating and executing purchase decisions); and agent infrastructure (payment, authentication, communication systems are already under strain at relatively low agent volumes). The authors simultaneously acknowledge that most enterprises still block agent traffic due to concerns about fraud and chargebacks.

Sources :a16z.news

Recherche

Siri Co-founder Backs Voice ID Startup Helix in $6 Million Funding Round

Sujet · Helix语音身份验证Prise rapide
2026-11-11 23:02 GMT+8

According to TechCrunch, voice identity verification startup Helix has closed a $6 million funding round led by Chicago Ventures, with participation from BAM Ventures, Aglaé Ventures, Corazon Capital, Tusk Ventures, Yonder Ventures, Rule30, and Brown Alumni Group. Dag Kittlaus, the co-founder of Siri, has backed the company and will serve as an advisor.

Founded in 2025 by Alex Oberg, Anna Dorofiyenko, and Mike Hall, Helix offers on-device voice technology designed to verify human users without collecting sensitive personal data. Oberg, who previously helped create Ticketmaster’s 'Verified Fan' system to block bots, argues that traditional identity proxies like email or IP addresses are failing as AI agents become more prevalent. He posits that voice, as a biometric signal, offers a more robust way to distinguish humans from machines.

The company claims its compressed model can run locally even on an iPhone 13, minimizing deployment costs. Zeel, an in-home healthcare provider, is already using Helix’s technology to verify visiting caregivers. Kittlaus noted that voice identification has reached a point of reliability, giving Helix a cost advantage through model compression. However, CEO Oberg acknowledged current limitations: the technology cannot reliably detect lies, does not identify individual speakers, and is not yet capable of detecting AI agents during live calls.

Sources :TechCrunch

Recherche

OpenAI Adds Business Profiles and Agents to ChatGPT

Sujet · OpenAI商业代理Matériel
2026-11-11 22:26 GMT+8

OpenAI has added a new feature for businesses within ChatGPT, similar to how companies claim and manage their presence on Google Maps. Verified business owners can now create a profile where they store essential information such as phone numbers, websites, opening hours, and details about products, prices, and availability.

When ChatGPT mentions a company during a conversation, users can access this profile via a link. There, they can start a chat with a "Business Agent," an AI assistant configured to answer questions based solely on the information the company has provided. This allows for direct interaction with brand-specific data without leaving the ChatGPT interface.

Companies can also view performance metrics in the "Business Hub," such as topics where their business is suggested to users. This data helps businesses update their descriptions, functioning similarly to SEO optimization but for AI chatbots. OpenAI emphasizes that these metrics are anonymous statistics and do not share individual user data.

The feature is currently available in beta for selected companies in the United States, with plans for a broader rollout. It represents a step towards integrating commercial entities more deeply into conversational AI platforms, competing with existing search-based discovery methods.

Sources :heise online

Recherche

Google DeepMind Releases AlphaGenome 2: Predicting Genetic Variant Effects on Specific Cell Types

Matériel
2026-11-11 23:00 GMT+8

Google's AI lab DeepMind released AlphaGenome 2 on November 11, 2026. The new genomic AI model is designed to unravel how changes in DNA affect different types of cells, giving scientists a more detailed view of genetic clues to disease causes.

Unlike its predecessor, AlphaGenome 2 predicts the effects of genetic variants on individual cell types and states. While nearly every human cell carries the same DNA, different cells—such as heart muscle, skin, or immune cells—switch different genes on and off, and these patterns change depending on what a cell is doing. Consequently, the same genetic variant can have very different effects depending on the cell type and state.

DeepMind stated that AlphaGenome 2 was trained on more than 100 million individual cell profiles across 68 tissues. According to the company, the model identifies nearly twice as many variants that can change gene expression in cells compared to its previous state-of-the-art model. Researchers also found that AlphaGenome 2 can make predictions about new types of cells and biological states without needing retraining, and it can identify gaps in experimental data to help guide future lab investigations.

AlphaGenome 2 is currently available for non-commercial research via the AlphaGenome API, with a limited "private preview" for commercial research. Žiga Avsec, Google DeepMind’s genomics lead, noted that while AlphaGenome 2 is a successor to AlphaGenome, its more detailed predictions come at a higher computational cost, so it will not replace the original model immediately.

Sources :The Verge

Recherche

Baillie Gifford Investment Manager Natzler: Can the Valuation Discount on ByteDance's AI E-commerce Closed Loop Narrow?

Sujet · 字节跳动估值折价Matériel
2026-11-04

Robert Natzler, investment manager at Schiehallion Fund under Baillie Gifford, summarizes his core assessment of ByteDance with a set of figures: ByteDance generated estimated revenue of approximately $186 billion last year, nearly matching Meta’s $201 billion; however, its most recent round valuation reached as high as $600 billion, roughly 60% lower than Meta’s public market capitalization. The profit multiple is even more striking—investors are paying $12 for every $1 of ByteDance’s profit, compared to $22 for Meta. Natzler argues that this disparity is key to understanding why ByteDance resembles a 'coiled spring.'

The mechanism supporting this view lies in ByteDance’s platform integration strategy. Traditional e-commerce begins with user intent—you decide to buy a pair of running shoes and then open a search engine or shopping site. ByteDance can intervene earlier: the user may not have been looking for anything specific but sees a pair of shoes in a dance short video. Its recommendation engine links discovery, entertainment, and shopping into a tight feedback loop—inferring who might be interested, stimulating desire, facilitating the transaction, and using each transaction to improve future recommendations. Natzler emphasizes that no Western platform possesses such complete control over the entire path from entertainment to checkout. This means ByteDance can capture more value generated across sales stages, and its distribution logic follows engagement rather than follower count, lowering barriers for new creators and thereby enhancing the platform’s appeal.

The AI dimension provides additional momentum to this closed loop. Data cited by Natzler includes: ByteDance’s spending on data centers and AI infrastructure reaching up to $70 billion this year; in the 12 months leading to June, the total volume of tokens served by its proprietary foundation models exceeded the combined total of OpenAI and Anthropic; and the AI assistant Doubao has become China’s most popular chatbot by monthly active users. These investments have not yet fully translated into independent revenue streams, but Natzler’s logical argument is that as AI capabilities continue to strengthen, combined with ByteDance’s deep understanding of billions of users’ interests, commercialization pathways may further broaden.

So when will the spring release? Natzler identifies two convergence conditions. First, IPO obstacles are receding—TikTok’s US business has been placed into a joint venture structure with majority US ownership, significantly alleviating one key variable that long hindered listing. Second, even if the valuation does not converge with Meta, merely aligning with listed Chinese tech peers would yield considerable results for Schiehallion shareholders. However, he explicitly acknowledges China-related risks: 'We claim no special insight into the trajectory of US-China relations, but we see good reasons for both sides to find a new understanding, although we cannot assume the doves will prevail.' The boundary of this argument lies in the fact that all valuation figures are private market estimates, untested by public trading; the $600 billion valuation itself may deviate from actual transaction levels; and the convergence implied by the 'coiled spring' metaphor is not inevitable—it depends on the interplay of geopolitics, regulatory paths, and market sentiment.

Sources :bailliegifford.com

Recherche

Schneider Electric Cooling Architect: Liquid Cooling Orchestration Does Not Track Instantaneous GPU Load but Schedules Setpoints Across the Entire Heat Dissipation Chain

Sujet · 施耐德电气液冷编排架构Matériel
2026-11-11 20:05 GMT+8

AI workloads can fluctuate on a second-by-second basis, but the thermal, hydraulic, and mechanical systems supporting liquid cooling respond much more slowly. Amir Ibraheem, Cooling Solutions Architect at Schneider Electric, points out that the goal of liquid cooling orchestration is not to track every individual change in GPU load, but to continuously optimize the end-to-end cooling ecosystem over longer time scales, ensuring that cooling capacity is available when needed while minimizing energy, water, and infrastructure overhead.

This objective is achieved through a two-layer control architecture. The local CDU control loop is responsible for maintaining supply temperature and differential pressure within expected operational ranges—a fast, component-level response. The orchestration layer operates at a higher supervisory level, continuously resetting setpoints for heat rejection facilities, chilled water plants, CDUs, and secondary fluid networks to maintain adequate thermal headroom at the cold plate with minimal energy and water usage.

The connection between the two layers relies on thermal buffering. Fluid volume, metal mass, and allowable temperature rise provide temporary buffers, allowing slower equipment time to respond. Orchestration software learns from historical and real-time data, using predictive models to identify sustained thermal demand trends and proactively adjusting setpoints before heat reaches the cooling loop.

On the GPU side, the orchestration software considers NVIDIA’s MAX-LPS, MAX-P, and MAX-Q configuration profiles, matching just sufficient cooling capacity to meet performance requirements rather than designing for peak loads.

It should be noted that this is the manufacturer's explanation of its own product architecture. There is no public empirical data quantifying the specific efficiency gains brought by the orchestration layer; the aforementioned setpoint scheduling logic represents design objectives rather than verified operational results. However, it clearly clarifies the control architecture of liquid cooling orchestration: instead of reacting to each load fluctuation, it schedules ahead across the entire heat dissipation chain.

Sources :Schneider Electric Blog

Recherche

Texas Instruments Unveils Three New Chips for AI Data Centers: Nanosecond Telemetry and 800V Power Protection

Sujet · 德州仪器AI数据中心硬件Matériel
2026-10-12

Texas Instruments announced three new analog solutions for AI data centers on October 12, 2026, at the Open Compute Project (OCP) Global Summit. These products aim to address challenges such as power instability, inadequate real-time monitoring, and signal degradation that arise as AI workloads increase in volume and complexity.

The flagship product, the ADC34DSO75, is a 3-gigasample-per-second (GSPS) analog-to-digital converter. TI claims it is the industry's highest-performance ADC of its kind, enabling 24/7, low-power, event-based remote monitoring at nanosecond resolution. According to the company, this is thousands of times faster than current solutions, allowing for the real-time detection and capture of transient fault events that cause costly downtime. As the first solution to offer fully remote, event-based telemetry, it eliminates the need for on-site debugging and oscilloscopes, providing hyperscalers with unmatched remote diagnostic capabilities at a fraction of the size and cost of conventional methods.

The LMI800PS is the industry's first 800V isolated hot-swap controller. It replaces a complex discrete solution requiring five integrated circuits with a single high-density package that delivers in-rush current and power path protection. By integrating an isolated bias supply and PMBus telemetry, the controller reduces board area by more than 60% and simplifies high-voltage designs. It meets safety standards with 5kVrms reinforced isolation and more than 8mm creepage, supporting both ±400V and 800V distributions for diverse rack architectures.

The third product, the LMK6B, is a bulk acoustic wave (BAW)-based oscillator that delivers best-in-class jitter performance and power-supply rejection ratio (PSRR). As data center transmission rates approach 3.2Tbps, the margin for timing error shrinks significantly, and even minute jitter can increase bit error rates. The LMK6B achieves 9.6 femtoseconds of jitter at 625MHz, ensuring reliable performance and preserving signal integrity for next-generation AI-driven networking infrastructures.

Sources :Texas Instruments

Recherche

Adobe Max 2026 Previews Four Experimental AI Tools for Creative Workflows

Prise rapide
2026-11-11 22:00 GMT+8

US software company Adobe (Nasdaq: ADBE) unveiled four experimental AI tools, dubbed "Sneaks," at its annual Max conference, aiming to reshape how creative professionals interact with generative AI. The tools—Project Screen Test, Project Sight Span, Project Sound Bender, and Project Rinse Repeat—are currently technology previews with no guarantee of integration into Adobe's Creative Cloud or Firefly platforms.

Project Screen Test allows users to develop AI-generated characters by conversing with them in an audition-like environment, defining personality traits and backstory through dialogue rather than just text prompts. Project Sight Span transforms flat images into explorable 3D environments, enabling users to reposition viewpoints and view content outside the original frame while preserving artistic styles.

Additionally, Project Sound Bender offers a visual interface for audio mixing, where users adjust AI-generated sound effects by dragging a cursor across a canvas. Project Rinse Repeat enables users to copy photo edits applied to one subject and automatically adapt them to other people or objects in the same image within Photoshop. Gautham Mysore, a senior principal research scientist at Adobe, stated that these experiments aim to make complex workflows more intuitive, giving experienced creatives more precise control over the output.

Sources :The Verge

Recherche

OpenAI reports over 20,000 US veterans used free ChatGPT Plus

Prise rapide
2026-11-11 16:00 GMT+8

OpenAI announced on November 11, 2026, that more than 20,000 transitioning service members and veterans have claimed a year of free ChatGPT Plus in the past year. The milestone, released ahead of Veterans Day, highlights how artificial intelligence is being used to help this group prepare for civilian careers.

According to OpenAI, the initiative started with veterans working at the company who understood the challenges of transition. Beyond free access, OpenAI Academy and community partners have hosted more than a dozen trainings for nearly 1,000 members of the military community, with seven more scheduled before the end of the year.

Many veterans are using ChatGPT to organize job searches, rehearse for interviews, and explore new career paths. For instance, Air Force veteran Tilak Bhatnagar used his free year to prepare for multiple interviews and articulate his military experience to civilian employers. Another Air Force veteran, Sadarian Bell, utilized voice mode to practice interview answers, describing it as a career coach.

Sources :openai.com

Recherche

German Translation Firm DeepL Launches Aure Model to Preserve Tone and Emotion in Real-Time Translation

Sujet · DeepL Aure模型Matériel
2026-11-11 21:50 GMT+8

German translation technology company DeepL announced on November 11, 2026, the launch of a new voice model called Aure, designed to improve the naturalness of its real-time translation service, DeepL Voice. According to The Decoder, the model carries over tone, emotion, emphasis, and speaking pace from the original speaker's voice into the target language.

Previously, DeepL Voice generated translated speech from a static voice sample captured at the start of a conversation, meaning it could not pick up on mood shifts or sarcasm. Aure works differently by continuously analyzing both the translated text and the original audio as the conversation unfolds. During training, DeepL separates voice identity, speech melody, and content so each component can be transferred into the new language independently, using reinforcement learning to keep accents out of the output.

Aure runs automatically in DeepL Voice for Online Meetings, In-Person Conversations, and the DeepL API. At launch, the model supports English, German, Spanish, French, Italian, Japanese, Portuguese, and Korean. This follows DeepL's initial rollout of voice preservation features in September.

Sources :The Decoder

Recherche

SAP and VF Corp Open 'Intelligent Fashion' Pilot Store in Germany to Test AI-Driven Autonomous Retail

Sujet · SAP与VF智能时尚试点Prise rapide
2026-11-11 21:00 GMT+8

German software giant SAP SE and US apparel company VF Corporation (owner of brands like The North Face and Vans) opened S.Fashion on November 11, 2026, at SAP's headquarters in Walldorf, Germany. This retail innovation space serves as a physical test environment for the "Autonomous Enterprise" vision, exploring how Industry AI can connect the shopper journey with back-end retail operations.

The store integrates SAP applications, automation, and connected hardware. Item-level RFID data allows AI agents to coordinate inventory, order management, pricing, and replenishment. Consumer experiences include navigation via Joule (part of the SAP Business AI Platform), holographic virtual try-ons, and personalized embroidery. Notably, during selected hours, the store operates autonomously with no staff on site, testing different retail models under live conditions.

Christian Klein, CEO of SAP, stated that the "Autonomous Enterprise becomes real when AI can understand business context and act across connected processes." Sujay Kumar, Vice President of Global ERP Technology at VF Corporation, emphasized that this live environment helps both companies learn what resonates with consumers and improves team efficiency. Currently, S.Fashion is open to SAP employees and their guests, with arranged visits for customers and partners, rather than being fully open to the general public.

Sources :prnewswire.com

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement

It is worth noting that this article was written by a venture capital firm with substantial investment positions in related fields, aligning its commercial interests with the narrative direction depicted. All core data consists of self-reported agent traffic or deployment metrics from companies, rather than independently audited revenue or profit figures. The judgment that 'agents have become the primary users' currently finds its strongest evidence in developer workflows; the scale of diffusion to general consumers remains to be verified. The article itself is positioned as the first in a series, with subsequent pieces planned to supplement data on consumer agent usage, agent decision-making mechanisms, and transaction completion hurdles.