AI-managed investment funds are quietly outperforming traditional stock market benchmarks, despite controlling only about $30 million of the nearly $17 trillion ETF market. The High Conviction US Equity AI-Managed ETF, launched in August by Tal Schwartz, gained 2.7 percent compared to the S&P 500's 1.1 percent gain. FinQ's AI ETFs posted even stronger returns, rising over 23 percent while the S&P 500 gained 11 percent. Pictet Asset Management's AI-enhanced ETF attracted $163 million in assets and returned 14.8 percent, though financial advisers caution that short track records and high fees may limit broader investor interest.
In other AI investment news, 91 percent of asset managers plan to expand AI portfolio tools in 2026, yet many remain cautious about fully automated decisions. A separate SS&C survey found that nearly half of organizations expect AI systems to participate in over 25 percent of work processes within three years. Two thirds of respondents named AI governance and legacy system integration as top priorities, but only 43 percent felt AI governance was applied consistently across their businesses.
On the safety and ethics front, researchers argue that fully aligning autonomous robots with human values is theoretically and practically impossible. Instead of maximizing alignment, the authors propose focusing on making misalignment incidents acceptable through a framework of normative requirements. Separately, Google DeepMind and Singapore's GovTech launched an on-device AI scam protection tool that uses a fine-tuned Gemma model running locally on phones to detect manipulative conversation patterns and trigger real-time fraud warnings.
LR-Guard introduced a faster approach to AI safety guardrails by using latent reasoning rather than explicit text generation. The method achieves a 19.4x inference speedup while outperforming the teacher model in most settings. In the blockchain space, AI agents spent over $138 million on the Base blockchain via the x402 protocol by Q3 2026, with over 87 percent of payments under 10 cents. The median time from funding to first payment dropped dramatically from 73 days to 18 minutes.
NOAA developed a machine learning system that predicts extreme heat events using weather station and satellite data, though translating predictions into protective actions for vulnerable communities remains a challenge. CM Labs added AI assistants and session review features to its Intellia training system for construction equipment. Colossal Biosciences launched GenoMorph, an AI platform that identified about 2,300 promising DNA editing targets for the dodo bird's cranial facial morphology. Researchers also found that Gene Expression-to-Image learning reveals a trade-off between spatial smoothness and gene-level fidelity when mapping biological data for neural networks.
Key Takeaways
- <ul><li>AI-managed ETFs like Tal Schwartz's High Conviction US Equity fund gained 2.7 percent versus the S&P 500's 1.1 percent, though they control only $30 million of the $17 trillion ETF market</li><li>Pictet Asset Management's AI-enhanced ETF attracted $163 million and returned 14.8 percent, while FinQ's AI ETFs gained over 23 percent</li><li>91 percent of asset managers plan to expand AI portfolio tools in 2026, but many remain cautious about fully automated decisions</li><li>Google DeepMind and Singapore's GovTech created an on-device AI scam detection tool using a fine-tuned Gemma model that runs locally on phones</li><li>LR-Guard achieves a 19.4x inference speedup for AI safety guardrails by using latent reasoning instead of explicit text generation</li><li>AI agents spent over $138 million on the Base blockchain by Q3 2026, with 87% of payments under 10 cents</li><li>Researchers argue full alignment of robots with human values is impossible and propose a framework for acceptable misalignment</li><li>NOAA developed a machine learning system to predict extreme heat events using weather station and satellite data</li><li>Colossal Biosciences' GenoMorph AI platform identified about 2,300 DNA editing targets for the dodo bird's cranial facial morphology</li><li>Nearly half of organizations expect AI to participate in over 25% of work processes within three years, but only 43% have consistent AI governance</li></ul>
Gene Expression-to-Image Learning Reveals Trade-Off in Spatial Mapping
Single-cell RNA sequencing allows cell-level biological analysis, but learning from gene expression data is difficult. Gene Expression-to-Image (GE2I) arranges gene values on pixel grids for neural networks. Researchers found a trade-off between spatial smoothness and gene-level fidelity in these mappings. One-to-one gene-to-pixel assignments improved predictions even when smoothness decreased. The study offers practical guidance for pixel allocation when applying vision models to gene expression data.
GeoVec-VLM Uses Vector Geometry to Improve Spatial Reasoning
Vision-language models (VLMs) struggle with spatial reasoning at large geographic scales because visual observations are limited. GeoVec-VLM introduces geospatial vector geometry as a native spatial modality. It uses structured geometry representation and a geometry interaction strategy that combines vector data with visual and language inputs. Across five spatial reasoning tasks, GeoVec-VLM improved mean accuracy by about 11 percent over the strongest baseline and showed greater geometric consistency.
AI-Managed Investment Funds Outperform Stock Market Benchmarks
A small group of fully AI-managed exchange-traded funds (ETFs) is outperforming major stock market benchmarks. Fewer than six such ETFs exist, controlling about $30 million of the nearly $17 trillion ETF market. The High Conviction US Equity AI-Managed ETF gained 2.7 percent since its August launch, beating the S&P 500's 1.1 percent gain. FinQ's AI ETFs gained over 23 percent while the S&P 500 rose 11 percent. Ninety-one percent of asset managers plan to expand AI portfolio tools in 2026, though many remain cautious about fully automated decisions.
Fund Managers Bet AI Bots Can Pick Stocks Better Than Humans
A handful of fund managers are using AI-powered exchange-traded funds (ETFs) to pick stocks automatically. Fewer than six ETFs relying entirely on AI manage about $30 million of the $17 trillion ETF industry. Tal Schwartz launched the High Conviction US Equity AI-Managed ETF in August, which uses a proprietary machine-learning algorithm and has gained 2.7 percent, outperforming the S&P 500. FinQ's two AI ETFs gained 25.35 percent and 23.98 percent respectively. Pictet Asset Management launched an AI-enhanced ETF that attracted $163 million and returned 14.8 percent. Financial advisers warn that short track records and high fees may limit investor interest.
Researchers Argue Full Alignment of Robots With Human Values Is Impossible
Designers of autonomous robots face the value alignment problem, which is how to ensure robots act in line with human values when many deployment unknowns exist. The paper argues that full alignment is both theoretically and practically impossible, meaning misalignment incidents will keep occurring. The authors introduce the concept of acceptable misalignment and propose normative requirements for alignment mechanisms. They present a new methodological framework focused on making misalignment incidents acceptable rather than trying to maximize alignment.
NOAA builds AI system to predict and protect against extreme heat
NOAA has developed a machine learning system that predicts extreme heat events using weather station and satellite data. The system can identify when and where heat waves will occur. However, experts still need to figure out how to turn those predictions into actions that protect vulnerable people. Targeted interventions like cooling centers and prevention programs are needed. Social and economic factors must also be included in the system.
AI agents spend over 138 million on Base blockchain, Chainalysis reports
AI agents are increasingly making automated purchases using crypto on the Base blockchain. Cumulative payments through the x402 protocol grew from about 1,500 in mid-2025 to over 138 million by Q3 2026. Developers fund AI bot wallets with USDC stablecoins so bots can pay for services without human approval. Over 87% of payments were under 10 cents, making blockchain cheaper than credit cards. The median time from funding to first payment dropped from 73 days to 18 minutes.
Enterprises expect AI in over a quarter of work processes within three years
A survey by SS&C found that nearly half of organizations expect AI systems to participate in over 25% of work processes within three years. Two thirds of participants named AI governance and legacy system integration as top priorities. Only 43% of organizations felt AI governance was applied consistently across their business. One in five respondents said measuring AI performance and ROI will be critical to success.
CM Labs adds AI tools and session review to Intellia training system
CM Labs added new features to its Intellia Workforce Training System for construction equipment training. The After-Action Review records training sessions and lets instructors replay them in a 3D environment to review trainee actions and errors. AI assistants answer questions based on CM Labs product documentation and help plan courses. A Performance Leaderboard displays live training results so instructors can track group performance.
Colossal Biosciences launches AI platform to aid dodo de-extinction
Colossal Biosciences announced GenoMorph, an AI platform using 3D morphology and comparative genomics to identify gene editing targets for de-extinction. The platform helped identify about 2,300 promising DNA editing targets for the dodo bird's cranial facial morphology. Colossal has only one dodo genome compared to 100 woolly mammoth genomes, making the AI tool essential. The company also reported its Nicobar pigeon colony has grown to 36 birds with cells that become eggs and sperm successfully created.
Google DeepMind and Singapore launch on-device AI scam protection
Google DeepMind and Singapore's GovTech created a new AI tool to detect scams on personal devices. The system uses a fine-tuned Gemma model that runs locally on phones to analyze conversations for manipulative patterns. It triggers real-time warnings when it detects potential fraud. The approach preserves privacy by keeping all analysis on-device rather than scanning messages in the cloud. Training used anonymized scam data and took place within GovTech's secure environment.
LR-Guard delivers faster AI safety guardrails with latent reasoning
Researchers introduced LR-Guard, a new method for AI safety guardrails that uses latent reasoning instead of explicit text generation. The system learns compact representations from a frozen teacher model's analysis of safe and unsafe prompts. It reconstructs human-readable explanations on demand without generating intermediate text for every decision. LR-Guard achieves a 19.4x inference speedup at batch size one while outperforming the teacher model in most settings. The method also produces reconstructed traces with high semantic similarity to the teacher's reasoning.
Sources
- Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity
- Geospatial Vector Geometry as a Native Spatial Modality for Vision-Language Models
- AI-powered investment funds outperform markets as managers test automated stock picking
- AI-Powered Stock Funds Bet Bots Can Pick Stocks Faster and Better Than Humans
- Acceptable Misalignment
- AI's next heat challenge: Protecting people
- AI agents are on a shopping spree: Total payments on Base blockchain eclipse 138 million, Chainalysis finds
- Nearly half expect AI systems to take part in over a quarter of work processes within three years.
- CM Labs Adds AI, Session Review Tools to Intellia Training System
- Colossal Biosciences announces new AI platform to help accelerate de-extinction efforts
- Google DeepMind and Singapore Government announce new AI tool to protect citizens against scams
- Efficient Safety Guardrails through Latent Dual-Hypothesis Reasoning with On-Demand Trace Reconstruction
Comments
Please log in to post a comment.