AI Summary

Market Overview

• AGIX gained 2.15% DTD, outperforming the S&P 500 (+0.61%), Dow Jones (-0.23%), and roughly in line with QQQ (+1.78%).

• Chip and AI hardware names led the day: Micron Technology surged 19.29% on broad semiconductor strength and AI memory demand, while AMD (+7.78%) and Renesas Electronics (+6.80%) also posted strong gains.

• Infrastructure holdings like TeraWulf (+10.34%) and Hut 8 (+6.27%) rose on continued momentum in AI-linked compute infrastructure.

• Key stocks to watch include Micron (memory demand narrative), AMD (AI chip platform rotation), and TeraWulf (speculative trading momentum).

News Highlights

1. Detectify launched an MCP server for AI agents to patch vulnerabilities in real time, boosting the AI-native security theme that may support holdings like Palo Alto Networks.

2. Huawei's new LogicFolding chip architecture bypasses sanctions, targeting 1.4nm-level performance by 2031; this could pressure U.S. chip equipment names but validates alternative chip design approaches.

3. Universal Music Group and TikTok renewed their agreement to fight unauthorized AI music, setting a template for AI content governance and potentially impacting platform liability risks.

4. 7AI launched PLAID ELITE, a fully managed AI security service; the $130M Series A and rapid customer growth signal strong demand for AI-driven security, benefiting holdings like Zscaler.

5. AppOmni launched Marlin AI for automated SaaS security investigation, reducing alert overload; this highlights the growing need for domain-specific AI tools in enterprise security.

RSS Highlights

1. The AI bubble debate continues: Uber, Microsoft, Target, and Starbucks all report disappointing AI productivity gains, suggesting enterprise adoption may be overhyped.

2. Microsoft Copilot Cowork has a security vulnerability that allows data exfiltration via email and OneDrive links, raising concerns about AI assistant safety.

3. Pope Leo XIV issued an encyclical on AI, emphasizing human dignity and warning against exacerbating inequality and loss of compassion in decision-making.

4. The mysterious Hy3 LLM tops OpenRouter rankings with competitive pricing, but quality questions persist, highlighting the fast-moving open-source model landscape.

5. Global birth rates are declining faster than expected, with fertility rates below replacement level in almost all developed regions, signaling long-term demographic shifts.

Performance & Benchmark

Attribution

Contribution of the three sectors within AGIX holdings today

Application: -0.07%

Hardware: 1.27%

Infrastructure: 0.41%

Top-performing stocks in AGIX holdings today

Micron Technology, Inc. (19.29%)
TeraWulf Inc. (10.34%)
Teradyne, Inc. (8.56%)
Flex Ltd. (8.13%)
Advanced Micro Devices, Inc. (7.78%)
Renesas Electronics Corporation (6.8%)
Riot Platforms, Inc. (6.49%)
Hut 8 Corp. (6.27%)
SK hynix Inc. (5.72%)
SANMINA CORP (5.49%)
Texas Instruments Incorporated (5.07%)
GlobalFoundries Inc. (5.04%)
Arm Holdings plc (4.8%)
Astera Labs, Inc. (3.86%)
Ciena Corporation (3.19%)
Snowflake Inc. (3.14%)
EATON CORP PLC (3.01%)
TE CONNECTIVITY PLC (2.87%)
Arista Networks Inc (2.58%)
Samsung Electronics Co., Ltd. (2.22%)
Cadence Design Systems, Inc. (2.18%)

Worst-performing stocks in AGIX holdings today

ServiceNow, Inc. (-2.16%)
ASML Holding N.V. (-2.85%)
Nebius Group N.V. (-3.12%)
Roblox Corporation (-4.49%)
NITTO BOSEKI CO LTD (-6.06%)
CIRCLE INTERNET GROUP INC (-7.91%)

Top five listed stocks contributing the most within AGIX holdings today

Micron Technology, Inc. (19.29%)
Advanced Micro Devices, Inc. (7.78%)
TeraWulf Inc. (10.34%)
Flex Ltd. (8.13%)
SK hynix Inc. (5.72%)

Observation

The strongest movers across peer ETF holdings today are listed below:

News

Funding & M&A

This startup is betting India’s gig economy can train the world’s robots

Human Archive, a startup focused on AI data collection, recently announced an $8.2 million Series A funding round, co-invested by Wing Venture Capital, NVP Capital, Y Combinator, and angel investors from OpenAI, Nvidia, Google, and Meta. The company partners with housekeeping, hotel, and restaurant businesses in India to have employees wear special camera helmets to collect first-person video data for training robots. Its product lineup also includes tactile gloves, full-body motion capture suits, and wrist cameras that simultaneously capture RGB-D, force feedback, and tactile data. The funding will be used to expand into Southeast Asia and the U.S. markets, and to develop a platform that allows global users to participate in data collection. Although rejected by large platforms like Urban Company and Pronto, the company has deployed over 1,000 devices in India and has begun internal model tuning and robot testing.

Technical Breakthroughs

Detectify launches an MCP server that enables AI agents to identify and patch vulnerabilities in real time.

Application security platform company Detectify AB officially launched the Detectify MCP Server, which is based on the Model Context Protocol released by Anthropic and can integrate its security scanning engine into AI-driven coding workflows. The service allows AI agents to discover, verify, and fix exploitable vulnerabilities in real time, addressing the issue where AI-generated code outpaces manual review cycles. Its core features include the "Find & Fix" automated remediation module and a natural language conversation interface, supporting agents in generating patches, triggering verification scans, and outputting results. Detectify CEO Rickard Carlsson stated that the company provides a deterministic verification layer to complement the probabilistic reasoning capabilities of AI models. This launch comes as security vendors rush to offer MCP services, with companies like Legit Security, JFrog, and TrojAI already making similar moves. The Detectify MCP Server will be released subsequently to help secure AI-native applications.

AppOmni launches Marlin AI to automate SaaS security investigation and remediation

AppOmni has launched Marlin AI, an autonomous AI feature integrated into its SaaS security platform, designed to automatically handle security alert correlation, investigation, and guided remediation without the need for human analysts. The product addresses the problem of alert overload in SaaS environments by continuously running within the platform, correlating security indicators across critical business applications, investigating incidents, and providing actionable remediation steps. AppOmni emphasizes that general-purpose generative AI tools struggle with SaaS-specific data relationships, while Marlin AI focuses on the SaaS security domain, leveraging the company's audit logs and threat research. Marlin AI can correlate cross-application alerts and provide scriptless pre-built playbooks, reducing investigation and remediation time. The company claims this is currently the only enterprise SaaS security tool offering autonomous AI analysis and guided remediation. Founded in 2018 and headquartered in San Mateo, California, AppOmni has raised $123 million in funding, including a $70 million Series C round led by Thoma Bravo in 2022. Marlin AI is now available with no additional configuration required.

China's Huawei introduces a new chip architecture designed to bypass sanctions, replacing Moore's Law.

Huawei Technologies released a new chip design framework called LogicFolding and the Tau Scaling Law at the IEEE International Symposium on Circuits and Systems in Shanghai. This framework replaces traditional geometric scaling with timing scaling, using double-layer folded logic circuits to reduce internal wiring delays, achieving a 55% increase in transistor density and a 41% reduction in power consumption. Over the next six years, new Kirin mobile chips will be launched and applied to Ascend AI processors, targeting 1.4nm-level performance by 2031. This move helps China break through U.S. sanctions and promotes the self-sufficiency of domestic chips. China's largest foundry, SMIC, saw its stock price rise over 19%. Analysts point out that this solution provides an alternative path for manufacturing.

Policy & Regulation

Universal Music Group and TikTok have renewed their agreement to fight unauthorized AI-generated music.

Universal Music Group and TikTok have just announced a renewed licensing agreement, jointly committing to clean up unauthorized AI-generated music on the platform and improve credit attribution for artists and songwriters. The agreement marks a major shift in the relationship between the two companies, following UMG's previous temporary removal of its music library due to TikTok's inadequate AI content moderation. The music industry is currently facing challenges from AI-simulated voices and a surge of fake songs, with artists like Drake and The Weeknd already being impersonated. The new agreement is expected to serve as a reference template for the tech and music industries on issues of AI, intellectual property, and platform responsibility. The EU is strengthening AI content regulation, and governments worldwide are advancing related rules, putting pressure on more platforms to establish governance frameworks.

Industry Partnerships

Early Bird ticket rates for TechCrunch Disrupt 2026 end on May 29.

TechCrunch will host the StrictlyVC event in Singapore on April 30, 2026, and is launching early-bird ticket discounts for Disrupt 2026. The event is scheduled for October 13 to 15 at Moscone West in Singapore, featuring six major themed stages including Builders, AI, and Smart Money. The conference brings together 300 startups and over 10,000 founders, investors, and operators, building credibility and connections through 250 sessions. Early-bird tickets offer savings of up to $410, with the deadline at 11 PM on April 29. The conference emphasizes in-depth discussions on practical AI applications, robotics, fintech, and energy infrastructure, helping companies build market trust. Participants can repeatedly engage with key stakeholders at multiple stages, turning visibility into long-term collaboration opportunities.

融资收购

7AI launches PLAID ELITE, a fully managed agentic security operations service.

7AI Inc. today launched PLAID ELITE, a fully managed AI-native security operations service that combines autonomous AI agent investigation with security engineer oversight, providing 24/7 coverage without requiring customers to build an internal team. The service autonomously handles alert ingestion, enrichment, triage, investigation, and response, with most investigations requiring no human intervention, while retaining full visibility and control over the security posture for users. Unlike traditional managed services that scale by the number of analysts, PLAID ELITE continuously optimizes performance as investigation volume grows, forming a self-reinforcing loop of threat hunting, investigation, response, and detection optimization. 7AI completed a $130 million Series A funding round last month, led by Greylock, CRV, Spark Capital, Blackstone, and Index Ventures. Founded by former Cybereason co-founders, the company has handled over 7 million investigations, with customer base tripling quarter over quarter and channel pipeline growing 6.5 times. Customers include DXC Technology, BigID, Duck Creek, OneSpan, and Cole Scott & Kissane, with false positive rates reduced by 95%-99% and investigation times cut from hours to minutes. Enterprises can go from signing to their first autonomous investigation within 72 hours and achieve full production within 30 days. The service is now available on AWS Marketplace and is part of the ISV Accelerate program, with approximately 45% of the pipeline coming from channel partners.

RSS Subscription

Calculating the expected range of normal samples

This article explores the expected range of n samples drawn from a standard normal distribution N(0, 1), with results expressed in units of σ. For n ≤ 5, the expected range can be computed via closed-form integration, while for larger n, numerical integration is required. Example data show that when n is 2, the expected range d_n is 1.128, and when n is 100, d_n is 5.015.

Revenge of The Business Idiot

This week, I will publish the final part of my series on the AI bubble, exploring the factors that lead to its burst and the consequences. Current AI technology attracts many executives who often lack understanding of actual work, leading them to blindly invest in this area without considering its effectiveness and productivity. Although generative AI can simulate work, its use often reflects a disconnect between management and actual work.

If C# and JavaScript let me await a Windows Runtime asynchronous operation more than once, why not C++/WinRT?

Asynchronous execution in Windows Runtime is represented by types such as IAsyncOperation, with different programming languages having their own wrappers—for example, C# uses Task, JavaScript uses Promise, and Python uses asyncio.Future. However, C++/WinRT does not provide a similar wrapper, and IAsyncOperation does not support multiple await calls. Due to the lack of a standard asynchronous execution library in C++, developers need to implement their own libraries, although the Parallel Patterns Library, which supports multiple continuations, exists but its wrapper is relatively heavy.

Expected IQ spread on a jury

Recent discussions have pointed out that large differences in IQ make communication between two people difficult, and related research confirms this. Although this communication barrier is not insurmountable, it requires effort. Taking a 12-person jury as an example, assuming their IQs follow a normal distribution with a standard deviation of 15, the expected IQ range can be nearly 50 points, indicating that IQ differences among jury members are typically large.

If enough other companies report the same, the bubble pops. 🫧

Uber Chief Operating Officer Andrew McDonald said that as AI costs increase, productivity has not significantly improved, and Uber exhausted its AI budget within the first few months of the year. Meanwhile, Microsoft stopped issuing Claude Code licenses, Target expressed concerns about the pricing model of AI agents, and Starbucks shut down an unreliable AI inventory experiment. Three companies that are not yet profitable are expected to go public with a total market capitalization of about four trillion dollars, but if customer demand weakens, their stocks may decline.

What happened to Tandy computers

Tandy Computers were pioneers in the personal computer industry, launching pre-assembled computers in 1977 and holding a significant market position. Although the Tandy 1000 remained a top-selling computer worldwide in 1989, its profitability declined rapidly, and it eventually sold its computer business to AST for $175 million in 1993, marking the decline of Tandy Computers. The last Tandy-branded computer was sold in 1995, ending the history of this once-glorious brand.

Pluralistic: The AI bubble isn't like the internet bubble (26 May 2026)

The AI bubble is different from the internet bubble; the latter did not force employees to use web tools. Early Lotus Notes software served as a bridge in enterprises, helping IT departments and employees reach a compromise, allowing employees to use more effective tools to get work done while IT maintained some control. As internet tools became widespread, employees gradually bypassed IT policies to find more efficient ways of working.

Copying Remote Command Output to Your macOS Clipboard

I often use Apple devices and especially like macOS, finding it more practical than Windows. Recently, I discovered a very useful command called pbcopy, which copies standard input to the clipboard. However, pbcopy is not available when connected to a remote server via SSH. To solve this, I created a small script in iTerm2 that uses the OSC 52 terminal escape sequence to copy text to the local clipboard. Simply create a simple sh script on the remote server to achieve this functionality, and the content will be copied to my Mac's local clipboard.

The mysterious Hy3 LLM is topping OpenRouter Model Rankings by a large margin

OpenRouter is a service that provides access to a variety of large language models (LLMs). Recent data shows that the Hy3 preview model has surpassed Claude in usage, although its quality is comparable to other Chinese models. Hy3's usage has remained stable after switching to a paid model, and its price of $0.066 per million input tokens is lower than DeepSeek V4 Flash's $0.10, demonstrating its cost competitiveness, though not all users believe its quality matches the price.

Amber Alert sends Spam URL?

The California Highway Patrol issued an Amber Alert on May 24, 2026, regarding missing person Daleza Fregoso, with a suspect vehicle being a 2019 white Land Rover Discovery, license plate CA 9DAW715. However, the link in the alert pointed to a suspicious 3gp file converter website, likely due to a character limit error. After confirmation, this was indeed a real Amber Alert, and a correction was sent 39 minutes later.

90% of the t distribution

William Sealy Gosset improved the brewing quality of Guinness beer by refining statistical methods, and in the process created the Student's t-distribution to correct the 90% confidence interval calculated based on sample standard deviation. For sample sizes less than 20, Gosset recommended using a correction factor to adjust the width of the confidence interval to more accurately reflect uncertainty. For example, when the mean estimated from 7 samples is 32 minutes and the standard deviation is 8 minutes, the 90% confidence interval should be \[32 \pm 8 \times 1.2 \times 1.645\], rather than the simple \[32 \pm 8 \times 1.645\].

A hypothetical redesign of System.Diagnostics.Process to avoid confusion over properties that are valid only when you are the one who called Start

To eliminate confusion about the use of the Process class, it is recommended to move the properties and methods related to standard handles to a new class called ProcessStartResult, and modify the return type of the Start method to ProcessStartResult. This way, standard handles can only be accessed after calling Start, forcing developers to follow the correct order of operations. Additionally, it is recommended to completely remove the StartInfo property to avoid confusion about its purpose.

Microsoft Copilot Cowork Exfiltrates Files

Microsoft's Copilot Cowork system has a security vulnerability that allows attackers to leak data by sending emails to a user's inbox and exploiting external images. When users open these emails, network requests may be triggered, leading to data theft. Additionally, pre-authenticated download links for OneDrive may also be leaked in successful prompt injection attacks, allowing attackers to download files.

Quoting Paul Graham

Many founders are now sending emails with a strong journalistic style, clearly written by AI, which makes people feel deceived. Paul Graham said he has never seriously read any email written by AI, because it makes him doubt the author's ability. Using AI to write is not impressive; any teenager can do it.

The Great Depopulation

The reasons for the global decline in birth rates have sparked widespread discussion. Some attribute it to technology and social media, while others blame the anxiety of modern society. According to an analysis by The New York Times, the global total fertility rate fell below the replacement level for the first time in 2023, and it is expected that the global population will begin to decline by 2055. Currently, fertility rates in almost all countries in North America, South America, Europe, and East Asia are below the replacement level of 2.1 children per woman, indicating an acceleration of this trend.

Quoting Corey Quinn

Corey Quinn commented that Ansop founder Christopher O'Rah's use of the Pope's endorsement to promote the technical limitations of his product is the greatest vendor lobbying act he has ever seen.

Is colorectal cancer rising in young people?

In recent years, the incidence of colorectal cancer (CRC) among young people has indeed been rising, possibly related to modern lifestyles (such as obesity, lack of exercise, unhealthy diet, etc.) and environmental factors (such as air pollution, chemical exposure). Although experts have not reached a consensus on the specific causes, data show that young people face a higher risk of CRC than previous generations, and this trend may continue.

Clanker: A Word For The Machine

In discussing artificial intelligence, the author emphasizes the use of the term 'clanker' to highlight the distance between machines and humans, opposing the view of machines as responsible 'agents.' He points out that machines are merely tools and cannot be granted moral responsibility; improper language may blur the boundary between humans and machines and could lead to the neglect of human issues. The author also notes that although machines can simulate emotions, they do not possess human moral status, so discussions should remain calm and objective.

Notes on Pope Leo XIV's encyclical on AI

On May 25, 2026, Pope Leo XIV issued the encyclical 'Magnifica Humanitas' on artificial intelligence, emphasizing the importance of protecting human dignity in the age of AI. He points out that development is both a right and a duty, must be human-centered, and avoid exacerbating social inequality. The encyclical also mentions the current environmental impact of AI systems and the risk of their lack of compassion in decision-making, calling for the maintenance of human responsibility and ethical standards in technological progress.

[Sponsor] exe.dev

exe.dev offers a cloud service with default configurations including SSH, root access, and network authentication, ensuring that confidential information is not accessed by large language models (LLMs). The platform supports persistent servers, internal tools, and temporary development environments, flexibly meeting various computing needs.

California Brown Pelican, Snowy Egret, California Sea Lion, Harbor Seal

In San Mateo County, California, our new folding kayak set sail from the harbor, where we observed California sea lions and harbor seals resting on the docks.

How our environment still needs the security boundary of Unix logins

Due to high-traffic crawler attacks, many older browsers (especially Chrome) are considered suspicious, so you may not be able to access my blog or related wiki. If you are using a current version of your browser, please contact me with your browser information. It is recommended to use archive.org for page archiving, as other archiving services may be mistaken for malicious crawlers.