AI.info

The Pulse — Page 63

Browse The Pulse on AI.info.

• Thomas Dohmke's Entire Launches a Git Forge for AI Agents

  Entire, founded by former GitHub CEO Thomas Dohmke, launched a Git-compatible code forge for developers and AI agents on October 26, 2026. Its Trails review system, distributed hosting and session memory are designed to help teams manage agent-written code.

• Altman Says AI Benefits Justify Accepting Some Harm

  OpenAI CEO Sam Altman said society should accept some AI-related harms in exchange for the technology’s benefits and people’s ability to use it. He distinguished those harms from catastrophic risks, including a serious loss of control to AI.

• Turkey Opens EVREN AI Platform to Defense and Research Users

  Turkey’s Presidency of Defence Industries has opened EVREN, a national AI platform for defense firms, researchers, students and other users. Anadolu Agency reported that it combines data, AI models and high-performance computing, while an EVREN Platform post describes an additional language model beyond the 11 reported by Anadolu.

• Strata Brings Qwen’s 125B Model to 12GB GPUs

  Strata runs Qwen3.8-Flash-Next on a 12GB NVIDIA RTX 5070 by splitting model work across GPU memory, system RAM and SSD storage. Its October 4 release adds decoding and long-prompt changes, while an AMD ROCm port reports slower results on a 16GB Radeon.

• Stanford’s TRACER Turns ICU Video Into Hourly Patient Records

  Stanford researchers introduced TRACER, an AI system that combines ICU video with electronic health records to summarize patient activity over time. In a study spanning 280 patients and about 60,000 camera-hours, the team found an association between lower purposeful activity and positive delirium assessments.

• OpenAI Rolls Out GPT-Live-1.1 With New ChatGPT Voice Options

  OpenAI announced GPT-Live-1.1 and its smaller Quip version for ChatGPT Voice on October 15, 2026. The update adds voice options across six languages and English, though OpenAI’s release notes do not confirm a total of 13 new voices.

• Z.AI Raises $5 Billion for Models and Self-Training

  China’s Z.AI raises $5 billion through a Hong Kong share placement and zero-coupon convertible bonds. About 60% of the proceeds will fund next-generation models and the company’s fully self-training system.

• Insurance AI Company Kalepa Closes $75 Million Series B, Says Revenue Tripled

  Kalepa raised $75 million in Series B funding at a $470 million valuation. The company says its revenue and customer count each tripled over the past year, and it plans to expand its AI product suite, move into new markets and double its headcount in 2027.

• US and China Plan AI Safety Talks the White House Says Are Not Scheduled

  Two sources briefed on the planning told Reuters the United States and China are preparing a mid-September dialogue on AI safety, led by Treasury Secretary Scott Bessent. A White House official says no such meeting is currently planned. Proposed topics include monitoring AI-driven cyberattacks and model distillation.

• OpenAI Agents Reach Research-Intern Level, Company Says

  Coding agents can now complete defined research tasks that would take skilled researchers several days, OpenAI says. The company reports 3.1 agent-workdays for every human research workday and targets an automated AI researcher by March 2028.

• Baidu Releases ERNIE 5.5 Flash Base Weights

  Baidu released ERNIE-5.5-Flash-Base, a model with 280B total parameters and 14B activated per token. The pre-training checkpoint is available in FP8 and BF16 formats for further training and inference.

• UCLA Team Tests Optical Deepfake Detection on 15 Video Streams

  A UCLA study reports a hybrid optical-neural system that classified 15 video streams in parallel, reaching 97.79% average accuracy on the Celeb-DF deepfake dataset. The laboratory prototype uses digital processing alongside an optical stage and is presented as an initial screening tool, not a finished detector for online video feeds.

• Strands harness claims 28% lower token costs across six benchmarks

  Strands Agents has released Strands harness, an open-source general-purpose agent that runs locally or on cloud providers and reports lower token use with nearly equal benchmark scores.

• Equinix Opens Singapore’s SG6 With 8 MW in Its First Phase

  Equinix has opened SG6, its sixth International Business Exchange data center in Singapore, with 8 MW of capacity and approximately 1,550 cabinets in its first phase. The facility is designed for AI and other high-density computing workloads, with liquid cooling, renewable-energy features and a planned sustainability research partnership with the National University of Singapore.

• Kimi K3.1 Pro adds more ZDR providers on Vercel AI Gateway

  Kimi K3.1 Pro and K3.1 Pro Highspeed are now available on Vercel AI Gateway from more providers with Zero Data Retention agreements. The page describes their capabilities, ways to call them and the Gateway settings that require ZDR for requests.

• Researchers Train Microrobot Navigation Policies in Under 10 Minutes

  A vectorized simulator processed roughly 190,000 transitions per second across thousands of parallel environments, allowing researchers to train navigation policies in under 10 minutes. The team also tested zero-shot transfer on two magnetically actuated microrobots, including in channels, moving obstacles, fluid flow and a human brain vasculature model.

• OpenAI Tells Investors Revenue Run Rate Is Near $50B

  OpenAI’s annualized revenue was reported at $50 billion, about $20 billion below a figure circulated in September. The higher figure was an investor comparison that adjusted for differences in how OpenAI and Anthropic account for cloud-partner sales.

• Microsoft Finds North Korean Groups Using ARTEX AI Agents

  North Korean groups Onyx Sleet and Pearl Sleet used the open-source ARTEX agentic penetration-testing framework in operations, Microsoft says. Its October 29 assessment also describes AI-assisted phishing, fake identities and malware development, while saying the activity remains human-directed.

• AI Scientists Struggle to Reproduce Quantum Research

  A study of AI research agents found failures across computational quantum science, including cases where papers, code and results conflicted. Its Residual Agents system improved reproduction rates by learning from expert-verified workflow repairs.

• Pangram Puts Colleges in an AI-Detection Arms Race

  About 1,400 North American colleges have purchased Turnitin’s AI detector, while Pangram is gaining attention for its low reported false-positive rate. Professors and universities are split between expanding detection tools and redesigning assignments around process and human review.

• MemComposer Improves Agent Results While Using Fewer Memory Tokens

  MemComposer selects and orders retrieved memories for AI agents, improving task success in tests across three customer-service domains while reducing memory-token use. The study evaluates the method with two frozen agents and reports that its learned selection policy also transferred to other tested domains and an executor.

• AstroForge puts its Solo AI system in command of a space probe

  AstroForge is preparing to fly Solo, an in-house transformer-based control system, on spacecraft designed for deep-space missions. The company says the system could diagnose and correct anomalies without waiting for instructions from Earth.

• Google Beam Ships to Five New Countries With 18 Partners

  Google is expanding its Beam 3D video communication platform beyond the United States to Canada, the United Kingdom, France, Germany and Japan. The company is also partnering with Industrious to offer Beam trials at selected U.S. workspaces beginning in October.

• ByteDance in Advanced Talks to Buy MediaTek Data-Center Chips

  ByteDance is close to an agreement to buy custom data-center chips from MediaTek, according to a Bloomberg report citing people familiar with the talks. MediaTek’s July guidance set a 15%–20% target share of the 2027 custom AI-chip market, while the potential ByteDance deal remains unconfirmed.

• JPMorgan and Goldman Back Only $1.3B of Volta’s $5B AI Loan

  JPMorgan and Goldman Sachs are underwriting $1.3 billion of Volta Infrastructure’s $5.012 billion AI loan package, leaving a $3.712 billion GPU tranche without their underwriting. The Norwegian project’s indicative yield reaches 11.75% as lenders seek buyers for debt tied to a Bitdeer data center serving Anthropic.

• IBM Study Finds 60% of Employees Fear AI Skills Erosion

  An IBM Institute for Business Value study finds that 60% of employees worry AI is eroding their skills, with critical thinking cited most often. The survey also shows a gap between what CHROs say workers need and the skills employees prioritize.

• Google Announces Offline Foresight Meeting Companion for Mac

  Google’s experimental Foresight app uses local AI models to assist with meeting notes and search across transcripts and personal files on Mac, including when offline. The announcement also details two mobile search demos and tools for developers building on-device retrieval features.

• Berget AI’s 8.5× Test Shows Where AI Runs Matters

  A typical Kimi K3 request produces an estimated 0.4 grams of CO₂ in Sweden and 3.2 grams in Texas on identical hardware, Berget AI says. Its request-level estimate includes hardware manufacturing emissions but excludes model training and the water and energy footprint of chip manufacturing.

• OpenAI Launches LifeVal to Measure AI on Everyday Tasks

  OpenAI introduced LifeVal on November 2, 2026, an evaluation of AI performance on practical tasks such as childcare, budgeting, shopping and accessing government services. The company says GPT-6 Astra earned 100% on several listed tasks, including booking a car inspection and ordering groceries within a budget.

• Australia’s House Passes Digital Duty of Care Bill

  The House of Representatives passed Australia’s broad Digital Duty of Care bill on Wednesday, 28 October, sending it to the Senate for further scrutiny. The bill would set safety obligations for online services, with an amendment covering chatbot conduct that simulates emotional intimacy and fosters a child’s dependence.

• Audit Finds Benchmark Graders Shift Scores by 22 Points

  A study of multimodal AI benchmarks finds that grading rules can change accuracy scores by up to 22 percentage points on identical model outputs. The authors recommend reporting non-compliance rates and results under at least two scoring rules.

• NVIDIA Makes cuDSS Generally Available for Simulation Software

  NVIDIA has made cuDSS generally available, giving developers a GPU-accelerated library for sparse linear systems used in engineering and scientific simulations. The announcement names six companies integrating the library and reports performance results for specific workloads.

• CFTC Chair Selig Raises ‘Know-Your-Agent’ Question for AI Trading

  CFTC Chair Michael S. Selig raised questions about responsibility and oversight as autonomous software begins to act in financial markets. He said SI systems could help exchanges identify suspicious activity and that CFTC staff are using SI to police markets.

• Samsung Electro-Mechanics Commits $4.9B to AI Substrates

  Samsung Electro-Mechanics approved 6.78 trillion won in investments to expand package-substrate capacity in South Korea and Vietnam. The projects target FC-BGA substrates used in AI accelerators and server CPUs, with new Sejong production lines scheduled to start in 2028.

• OpenAI Promises a Daybreak Blue Training Opt-Out After Backlash

  OpenAI’s updated data-controls page says it may use personal-account Daybreak Blue sessions to improve its models, even when the standard training toggle is off. After users objected, the company promised a separate opt-out for those sessions, with more details expected closer to November 29.

• Apple Adds M6 to iMac and MacBook Pro, Pitches Local AI

  Apple announced M6 versions of the iMac and 14-inch MacBook Pro, with preorders beginning October 28 and availability starting November 6. The company emphasizes on-device AI performance and says some server-based Apple Intelligence features have daily usage limits, with expanded access planned for a future fee.

• Enveda Raises $311M to Advance AI-Discovered Drugs

  Enveda has raised $311 million in Series E funding at a $2 billion valuation to advance three AI-discovered drug candidates. The Boulder biotech will use the money to move programs for eczema, metabolic health and inflammatory bowel disease into later-stage trials.

• FCC weighs Club for Growth bid to permit AI political robocalls

  The FCC is considering a Club for Growth petition to allow unsolicited political calls to cellphones using prerecorded or AI-generated voices. Replies in the agency’s comment process are due October 19, ahead of the November midterm elections.

• Meta’s 1,300-Line Router Made Its AI Agents More Reliable, It Says

  A 1,300-line Meta orchestrator routes more than 20 data workflows to 14 specialized AI agents. The system cuts new-column configuration work from days to under an hour, the company reports.

• RRSI Paper Describes a Method for Improving Agent Harnesses

  RRSI improves performance on both the split used for evolution and five out-of-distribution benchmarks while reducing policy-token use, according to the arXiv abstract.