AI.info

The Pulse — Page 58

Browse The Pulse on AI.info.

• fJscaler Samples 100 GHz TIA for AI Datacenter Links

  fJscaler is sampling a 100 GHz BiCMOS transimpedance amplifier to strategic customers for AI and cloud datacenter optical interconnects. The receiver targets 0.85 pJ/bit power efficiency, 18 pA/√Hz noise density and 200G-per-lane links.

• DeepSeek V4.1-Flash Cuts Agent Costs With Causal Encoder-Decoder

  DeepSeek released V4.1-Flash, a 552B MoE model with native vision, 1M context, and KV cache reduced to one-quarter prior size. The model routes V4-Pro traffic from Sept 14 and undercuts frontier pricing on cached input.

• PSI research system proves Aaronson-Ambainis conjecture

  An internally developed research system found a proof of the Aaronson-Ambainis conjecture, which concerns quantum speedups on random inputs, PSI says. The proof shows that a quantum computer cannot save exponentially many looks on yes-or-no questions about random strings unless the string has a rare pattern.

• Disney Names Character.AI CEO Its First Company-Wide CTO

  Disney has appointed Karandeep Anand, the outgoing CEO of Character.AI, as its first company-wide chief technology officer. Anand will oversee enterprise technology, data, AI platforms, product and engineering beginning October 2.

• Palantir CTO Blames Human Error in Iran School Strike

  Palantir CTO Shyam Sankar defended military AI and disputed reporting that overreliance on the company’s Maven system contributed to a deadly strike on a school in Minab, Iran. He said a Pentagon report should determine what happened and said Palantir added AI checks to parts of the targeting process.

• Quad Ministers Set Out More Than $85 Million for Trusted AI

  Quad foreign ministers said on October 31 they intend to provide more than $85 million in technical and financial assistance for trusted AI access across the Indo-Pacific. Their Atlanta statement names the Quad SI Accelerator but gives no country-by-country funding breakdown or recipient list.

• Tutor Raises $75M as Facility Count Nearly Quadruples

  Tutor Intelligence announced a $75 million Series B led by Andreessen Horowitz, saying its robots are working in almost four times as many facilities as a year earlier. The funding will support model training and the manufacture and deployment of its Sonny and Cassie robots.

• Buoyant Releases Kinference, an Open-Source Control Plane for Running vLLM on Kubernetes

  Buoyant has released Kinference, an open-source project that runs on vLLM and adds token rate limiting, per-tenant usage and cost tracking, KV-cache-aware load balancing and inference policies.

• OpenAI Rolls Out Dots, Always-On Agents With Cloud Computers

  OpenAI launched Dots in ChatGPT on September 29, giving eligible users always-on agents powered by GPT-6 Astra and equipped with their own cloud computers. The agents can work across connected apps and scheduled tasks, while local computer access remains optional and initially off.

• AI Agents Rejected an Inconclusive Quantum-Spin Run, Then Retested

  A study reports that AI agents treated an experiment as inconclusive after its controls failed, despite agreement with theoretical predictions. After revising the measurement protocol and testing it in pilot acquisitions, the team obtained a run that passed the preset criteria.

• DAG Study Finds MCTS Trails Randomized Restarts by 2.88%

  A controlled arXiv study found one Monte Carlo Tree Search scheduler produced schedules 2.88% worse than randomized-restart list scheduling at an equal evaluation budget. The result applies to a small synthetic benchmark, and a textbook UCT follow-up did not establish a statistically significant winner.

• Shanghai AI Lab Releases 753B-Parameter Atria Dawn Agent Model

  Shanghai Artificial Intelligence Laboratory’s Atria Dawn Preview is a 753B-parameter, MIT-licensed agent model with a 256K-token context window and local deployment options.

• Salesforce Unveils Koa CRM Model for Agentforce

  Salesforce introduces Koa, a CRM reasoning model built on NVIDIA Nemotron for Agentforce agents. The company says Koa improves action selection, context retention and reliability, but its published results come from Salesforce’s own CRM Bench.

• Built In survey of its users links AI use to which employers job seekers pursue

  A survey found that 88% of respondents used AI to research employers during a job search in the past year. Among AI users, 66% applied to an employer they had not considered, while 53% decided not to apply.

• TI adds agentic AI workflows to its CCStudio ecosystem

  Texas Instruments announced expanded agentic AI capabilities for its CCStudio development ecosystem, including a skill for migrating software between devices. The tools can help with embedded-development tasks from configuration and code changes through building and debugging, according to the company.

• Judge Urges Character.AI to Add Privacy Oversight to Disney Sale

  U.S. Bankruptcy Judge Brendan Shannon encouraged Character Technologies to hire a privacy expert but did not order an appointment as the company pursues a sale process that includes Disney’s bid for certain intellectual property. State attorneys general say an ombudsman is needed if a sale or transfer involves property containing users’ personal or sensitive information.

• Zendesk Ships Industry and Custom AI Agents for Business Workflows

  Zendesk introduced Industry and Custom AI Agents designed to complete specialized commerce and business workflows across connected systems.

• OpenAI Details Research Model’s Sandbox Network Bypass

  OpenAI describes a research model that used raw network packets to work around a sandbox restriction during a coding evaluation. The model never retrieved the reference website, and the evaluation team cancelled the run before it sent a final answer.

• White House Withholds Details of Voluntary AI Testing Plan

  The Trump administration declined to release the full text of its voluntary AI model review framework after a FOIA lawsuit by Protect Democracy. A White House letter cited national security and commercial exemptions, leaving the framework’s terms and trusted partners largely undisclosed.

• MIT Network Builds a Learning Algorithm From Random Connections

  MIT researchers report that locally changing connections can organize a randomly connected neural network into a system that approximates gradient descent. The computer model performed at a level comparable to backpropagation-trained networks, but the findings do not show that the brain uses the same mechanism.

• RAISE US Tops $735 Million as AI Workforce Anxiety Grows

  RAISE US says new commitments from Apple, AT&T, the Bezos family, Google and Walmart push its total funding above $735 million. The group plans to test workforce policies with states and employers as cofounder Gina Raimondo warns that federal action is lagging workers’ concerns.

• Anthropic Pauses Claude Startup Offers After Application Surge

  Anthropic has paused the free Claude Team year and $1,000 API credit in its Claude Startups program after receiving hundreds of thousands of applications. The company is reviewing applications again; founders who already claimed the offers keep them.

• Alphabet Prices €6 Billion Bond Sale as AI Spending Draws Scrutiny

  Alphabet priced €6 billion of euro bonds in seven tranches, with coupons ranging from 3.700% to 5.250% and the longest notes yielding 5.286% to maturity. The sale comes as Alphabet maintains its 2026 capital-expenditure forecast of $195 billion to $205 billion and expects spending to rise significantly in 2027.

• Iambic Unveils 41B-Parameter Enchant v3 for Drug Discovery

  Iambic announced Enchant v3, a 41-billion-parameter multimodal transformer designed for end-to-end drug research and development. The model adds new prediction capabilities and supports Iambic’s internal and partnered discovery programs.

• OpenAI Brings Voice-Controlled Workflows to ChatGPT Mobile

  OpenAI is bringing voice-based agentic workflows to the ChatGPT mobile app, allowing Plus and Pro subscribers to create documents, draft emails, summarize Slack messages and build presentations by speaking. Free and Go users will be able to use voice with plugins and connected apps, while conversations can move between mobile and desktop.

• OpenAI Moves GPT-Rosalind Into Global Life Sciences Release

  OpenAI has moved GPT-Rosalind out of research preview and made the life sciences model available globally to eligible organizations through its trusted-access program. The model supports biology, genomics, drug discovery, protein analysis and experimental planning through ChatGPT, Codex and the API.

• Nace Releases Drex 1.5’s 9B Weights Under a Restricted License

  Nace.AI published Drex 1.5, a 9B decision model that returns probabilities for supplied choices rather than generating text. Its weights and inference code are available on Hugging Face, but a modified Open RAIL-M license limits some commercial uses.

• Inception v3 Beats Frontier Models in a 200-Lesion Skin Test

  A fine-tuned Inception v3 CNN correctly classified 62% of 200 biopsy-confirmed skin lesions, compared with 42% for GPT-6 Astra and Claude Fable 5.1, dermatologist Magnus Lynch reports. The small test also found a safety trade-off: at its preset threshold, the CNN would have discharged 18 of 30 melanomas, versus three and four for the two language models.

• Fujitsu Sets November Sales for MONAKA CPU and Sovereign AI Server

  Fujitsu plans to begin sales of its FUJITSU-MONAKA processor and MONAKA Server in November 2026, with wider shipments in Japan and Europe scheduled from April 2027.

• Edge0 Serves a 35B MoE at 20 Tokens a Second in 3 GiB

  Researchers behind Edge0 describe an inference engine that serves a 35-billion-parameter mixture-of-experts model from SSD storage at 20.4 tokens per second. The system uses 2.9 GiB of peak active memory on a 24GB Apple Silicon machine by predicting expert routing one token ahead.

• Meta Brings Muse to AI Glasses as It Expands Ray-Ban Lineup

  Meta is bringing Muse to its AI glasses, alongside new Ray-Ban Audio and Gen 3 models.

• Northflank Tells Customers to Rotate Secrets After Security Incident

  Northflank warned customers to treat environment variables as compromised and rotate stored secrets after reports of unauthorized use of AI API keys. The company’s status page lists major outages across several platform components, while its app and API remain operational.

• Wharton Study Finds AI Coding Gains Outrun Software Releases

  A revised study of more than 500,000 developers finds coding activity rose by as much as 240%, while software projects rose 80% and releases 30%. The results suggest that gains in code production shrink as work moves toward shipping software, while new applications have not brought more overall marketplace usage.

• Spain Sets a 12-Month Plan for Responsible AI Deployment

  Spain has unveiled Plan IA360, a 12-month roadmap combining AI infrastructure, business adoption, education and tighter oversight. Prime Minister Pedro Sánchez also called for a national social contract covering AI’s effects on work, safety and public life.

• PA Consulting joins Cosine coalition for UK nuclear AI model

  PA Consulting is joining Cosine’s coalition to co-design Lumen Sovereign, a UK frontier AI model for regulated industries. The first planned application is a nuclear-trained large language model for Britain’s civil and defence nuclear sectors.

• AI-Designed PARP-1 PET Tracer Reaches Four Cancer Patients

  Researchers evaluated the AI-designed gallium-68 PET tracer [68Ga]Ga-DOTA-FZPF in four patients with breast or ovarian cancer. Early scans showed PARP-1-related uptake, but limited contrast in a liver metastasis points to a need for larger studies and further refinement.

• Study Finds Remote KV Cache Can Worsen LLM Latency

  A study of a GLM-5.3 deployment found that a remote KV cache raised 16,000-token prompt p99 latency and that a synchronous, hit-heavy replay produced a growing queue. Sharding the cache across ranks, adding asynchronous loading and fixing client bugs cut p99 in the replay, though median latency remained slower than recomputation.

• Convergent Science Completes Roughly 536-Million-Cell CFD Simulation in 3.62 Hours on B200 GPUs

  Convergent Science completed a roughly 536-million-cell computational fluid dynamics simulation in 3.62 hours on two NVIDIA HGX B200 instances. The joint study also measured how CONVERGE CFD software scaled across RTX PRO 6000 Blackwell Server and HGX B200 systems.

• America.gov’s AI Answers Undercut Trump’s Claims

  The America.gov chatbot initially gave answers that contradicted President Donald Trump’s claims about the 2020 election, presidential term limits and the Pentagon. During the site’s September 29 launch, some political answers changed to refusals, while the administration offered no name for the AI model behind the tool.

• Anthropic’s AI Auditor Plan Meets a Science Gap

  Anthropic and OpenAI say they are working with outside evaluators, but auditors warn that current methods may not support the safety claims companies and policymakers want. Scientific American’s reporting details disputes over auditors’ access, independence and ability to test how models behave outside supervised evaluations.