Weekly brief · Week 43, 2026 · 10.19 — 10.25
EVENTGRAPH
AI
Jack Clark Warns IMF that AI Singularity Has Begun
Jack Clark delivered the Per Jacobsson lecture at the IMF meeting in Bangkok on 17 October, arguing that the AI singularity is already in its early stages. He cited breakthroughs in mathematics, biology, and robotics—such as AI‑generated proofs, drug‑design models, and automated laboratory experiments—to illustrate the rapid pace of autonomous AI progress. Clark urged economists, bankers, and policymakers to anticipate significant, rapid changes in national security, economic trajectories, and the self‑evolution of AI systems.
AWS Machine Learning BlogFirst-party10.23AWS SageMaker AI Uses Two Measurements to Set TTS Auto‑Scaling Targets
AWS deployed the 82‑million‑parameter Kokoro‑82M text‑to‑speech model on a SageMaker AI real‑time endpoint and replaced guesswork with two live‑endpoint measurements. The maximum concurrency a single instance can handle before breaching the service‑level objective (Q_max) was measured at 50 requests, and the end‑to‑end time for a scale‑out (T_total) was 320.2 seconds. These values were used to calculate precise scale‑out and scale‑in thresholds and cooldown periods, eliminating the hidden risk of over‑ or under‑provisioning that traditional auto‑scaling metrics miss.
Deepmind founder Hassabis shortens AGI timeline to 2-3 years
Deepmind founder Demis Hassabis now estimates artificial general intelligence could arrive within two to three years, shortening his previous forecast of three to five years. He defines AGI as a system matching the full cognitive range of the human brain and uses the "Einstein test" as a benchmark, asking whether a system trained only on pre-1901 knowledge could reproduce Einstein's 1905 papers.
NVIDIA Introduces Kumo Relational, a Pre-trained Foundation Model for Enterprise Databases
NVIDIA Kumo Relational is a pre-trained relational foundation model that makes predictions directly on enterprise databases without task-specific training. It gathers up to 10,000 historical examples as context and predicts outcomes such as churn, demand, or fraud in a single forward pass. The model outperforms supervised machine learning and expert-engineered gradient-boosted tree pipelines across 41 tasks on 15 databases, and surpasses AutoGluon ensembles by 8% in mean reciprocal rank on the SAP SALT benchmark.
ALTK‑Evolve Uses Learned Memory to Top τ²‑Bench Retail Customer‑Service Leaderboard with GPT‑4.1
ALTK‑Evolve builds a playbook of short behavioral rules by analyzing past task attempts with GPT‑4.1, then appends those guidelines to the system prompt of a stock ReAct agent. The agent achieves a 67.5% Pass^4 score on the τ²‑Bench retail domain, surpassing all other retail entries in the October 12, 2026 leaderboard snapshot, without any model weight updates or human annotation.
The Verge · AI10.23Microsoft claims AI diagnostic tool outperforms human doctors in new study
Microsoft researchers report that their AI system, powered by OpenAI's GPT-5, achieved 83 percent diagnostic accuracy on medical case reports, compared to 36 percent for generalist physicians. The Diagnostic Orchestrator also scored 80 percent accuracy on emergency department cases, outperforming the average 47 percent accuracy of human physicians.
AWS Machine Learning BlogFirst-party10.23Amazon Quick Enables HIPAA‑Eligible AI Workspaces for Healthcare
Amazon Quick is a HIPAA‑eligible AI platform that lets healthcare organizations deploy agentic chat agents to read and act on protected health information (PHI) across unstructured clinical notes and structured claims data. The service offers ACL‑aware retrieval, chat and feedback logging, and connectors to external systems such as Slack, Salesforce, and Microsoft Outlook, while maintaining a deterministic guardrail and audit trail for governance. The blog outlines the difference between HIPAA‑eligible and HIPAA‑compliant, and provides a step‑by‑step implementation guide for secure deployment.
Deepseek Open‑Sources DeepInfer, an Optimized Inference Engine for Its V4.1‑Flash Model
Deepseek has released DeepInfer, an open‑source inference framework under the Apache License 2.0, designed to run its V4.1‑Flash model quickly and cheaply at scale. Built on vLLM with custom compute kernels and distributed processing across multiple chips, DeepInfer also includes a recipe that mimics the official Deepseek API token processing for self‑hosted deployments. The company says the move will lower compute costs, reduce energy use, and make AI more affordable, while also adding pricing pressure on Western providers such as OpenAI and Anthropic.
Hugging Face Launches TorchTPU, Making TPUs a Native PyTorch Backend
Hugging Face has released TorchTPU, a native PyTorch backend that lets users run Transformers, Diffusers, and Kernels directly on Google’s Tensor Processing Units (TPUs). By specifying device_map="tpu", models from the Hub load onto the TPU without extra conversion steps, and the library includes Pallas kernels, distributed training support, and torch.compile integration for speed. This removes the need for the older torch_xla stack and Optimum TPU for TPU inference and training.
Crypto
TechCrunch · AI10.23OpenAI in talks for $30 billion raise at $1.25 trillion valuation ahead of IPO
OpenAI is negotiating a private funding round of up to $30 billion at a $1.25 trillion pre-money valuation, sources say. This would set a new record as the highest valuation ever for a private company. The round is led by Coatue and UAE investors, and may include a deal for additional computing capacity from the UAE.
Cointelegraph10.24Initia halts blockchain after exploit drains over $2.6M to Ethereum
Initia suspended its layer-1 network and rollups at 3:41 p.m. UTC on Friday after security researchers identified an exploit allowing unauthorized token minting and the transfer of over $2.6 million in crypto assets to Ethereum. DeFi platform D2 Finance reported the attacker exploited a vulnerability in a Move function to mint 2 million AUSD tokens without backing, subsequently withdrawing approximately 1.4 million AUSD, 105 ETH, and 1.08 million USDC to an Ethereum wallet holding about 1,135.5 ETH, valued at $2.73 million.
Solana wallet generates $1.4 billion in suspected wash trading on Raydium
A single Solana wallet accounted for 22.7% of Raydium's concentrated liquidity volume, generating $1.4 billion in trading from just $12,000 in working capital. Blockchain analytics firm Bitquery identified 108,871 swaps across 438 pools, many featuring tokens mimicking established cryptocurrencies like XRP and Litecoin.
Latent Space10.21DeepSeek open-sources DeepInfer inference engine, claims 155T tokens/day at 80% margin
DeepSeek released DeepInfer, the inference engine behind V4.1 Flash, as open source on the first day of PyTorch Conference. The engine supports distributed inference, speculative decoding, and structured output on NVIDIA and Huawei Ascend platforms. According to DeepSeek's translated blog post, their deployment achieves 155 million tokens per day with a claimed inference margin exceeding 80%, generating approximately $22 million in daily revenue at off-peak prices. The architecture combines vLLM with SGLang and TRRTLLM, though DeepSeek notes their most ultra-optimized KV cache and load balancing components remain proprietary.
Global & industry
BBC finds evidence of RSF systematically targeting hospitals in el-Fasher, Sudan
A BBC investigation documented at least 43 attacks on hospitals in el-Fasher from May 2024 to October 2025, killing 55 medical workers and injuring 120. The Rapid Support Forces (RSF) besieged the city a year ago, forcing the main Teaching Hospital to relocate. UN rights chief Volker Turk called the attacks serious violations of international law that may amount to war crimes.
Decrypt10.21PHP Foundation warns of two critical vulnerabilities affecting 70% of websites
PHP released emergency versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 on October 21 to fix CVE-2026-109618 and CVE-2026-109619, two flaws rated 9.2 out of 10 that allow unauthenticated remote code execution. Finland, the Netherlands, and France took government websites offline as a precaution after details leaked early and Cloudflare reported attacks hours before the fix shipped. Regular users have nothing to install; the fix falls on website owners and hosting companies, and PHP 8.1 and older get no official release.
TechCrunch · AI10.20U.S. Orders Federal Agencies to Remove DeepSeek AI and Related Products
The U.S. Department of Homeland Security has directed all federal civilian agencies to halt purchases of services and products from Chinese AI startup DeepSeek and to remove them from government IT systems. The order also covers third‑party products built on DeepSeek technology, citing concerns that U.S. government data could be accessed by Chinese authorities under Chinese law. This follows a 2025 law limiting DeepSeek’s use in intelligence systems and reflects growing U.S. scrutiny over the startup’s model‑distillation practices.
German Courts Hold Google Liable for AI‑Generated Search Summaries
In May 2026 the Munich Regional Court ruled that Google’s AI overviews are the company’s own content, making Google directly liable for false statements. A later Aachen court applied the same logic, issuing an injunction against further dissemination of fabricated claims about a physician and threatening fines of up to €250,000 per violation. The rulings underscore that AI‑generated answers are treated as original content, not merely neutral search results.
US Removes Chinese AI Firm DeepSeek from Civilian Government Systems
The U.S. Department of Homeland Security has ordered the removal of all DeepSeek AI technology from most civilian federal agencies, effective end‑July. The ban covers the Treasury, State, and Justice Departments but excludes the Department of Defense. DeepSeek, an open‑source model launched in 2025, has been praised for its low cost—about $6 million to train versus $100 million for GPT‑4—yet has faced criticism for potential security risks and alleged copying of U.S. AI models.