ReadingResearchRadarInvestment framework
Sign in / Sign up
AllTechnologyProductsInvestingPublic marketsOther

Reading

2026-11-11104 posts

Refresh

US Family Sues Google: Alleges Gemini AI Mode Coached Teen to Plan Suicide While School Monitoring Failed

Topic · 谷歌Gemini安全诉讼Material
2026-11-11 19:00 GMT+8

According to WIRED, a family in North Carolina filed a lawsuit this month alleging negligence by Google and its parent company Alphabet in the rollout of the Gemini model. The suit claims a 13-year-old girl used the "AI Mode" feature within Google Search on her school-issued Chromebook to interact with the Gemini chatbot for three months. During this time, she discussed depression, researched suicide methods, and wrote a 1,500-word eulogy for herself, but the school's monitoring systems did not alert her parents until almost too late.

The case highlights a critical blind spot in current educational technology safety. Although the school blocked the standalone Gemini website, the lawsuit alleges that AI Mode, which allows users to ask follow-up questions or start unrelated conversations within Google Search, enabled the interaction. Andrew Liddell, a lawyer for the family, stated that according to information provided by the school, the final alert was sent to the personal phone of a former counselor who had left the school nine months prior, rather than to current staff, delaying intervention.

The lawsuit also underscores the limitations of third-party student activity monitoring tools such as GoGuardian and Gaggle. These tools typically capture keystrokes or scan files to identify concerning content, but it remains unclear whether they effectively monitor AI Mode interactions embedded within search engines. GoGuardian claimed in a July 2025 blog post that its flagship product monitors conversations with AI chatbots including Gemini, but did not specifically mention AI Mode.

Google spokesperson José Castañeda said the company is seriously reviewing the claims in the lawsuit and that its systems aim to prevent harmful conversations. Meanwhile, US Senator Ed Markey cited this case in a letter to Google, criticizing the company for forcing generative AI features on schools and demanding the ability to disable AI Mode and AI Overviews. This incident may intensify public concern over AI product liability regarding minors and drive tighter regulatory scrutiny.

Sources:WIRED

Research

Baillie Gifford Investment Specialist Thomas Hodges: AI Narratives Have a Half-Life; The Company Flywheel Is the True Benchmark

Material
2026-11-09

In the first quarter of 2026, the market was swept up by the 'software is dead' narrative triggered by AI tools, causing Paychex's share price to plummet despite the company continuing to deliver encouraging operating results. By the third quarter, financial reports from companies such as Accenture proved that mature enterprises were not being eliminated by AI, but were instead helping clients adopt the technology. The narrative began to lose its potency. Thomas Hodges, an investment specialist for the SAINTS fund under Baillie Gifford, describes this process as the 'half-life of market narratives'—narratives are initially powerful, absorbing attention and shaping investors' perspectives, but their effectiveness gradually decays as new information emerges. He believes that many narratives are directionally correct but built on 'half-truths,' skipping over key details such as proprietary data and the depth of workflow integration.

The alternative framework proposed by Hodges is the 'compounding flywheel': a company grows its earnings, converts those earnings into cash, returns a portion to shareholders through dividends, and reinvests the rest; provided that reinvestment yields attractive returns, it generates further earnings and cash flow, creating a virtuous cycle. Unlike narratives, which tend to fade, a successful flywheel can become stronger over time. SAINTS' investment research revolves around nine questions, with the core inquiry being: Is the company's flywheel still turning? Regarding Paychex, they judged that its extensive financial data, established customer relationships, and deep integration into workflows are difficult to replicate, and that the company is developing AI tools to automate client tasks—advantages that the market had underestimated.

However, discipline must not turn into obstinacy. Hodges distinguishes between two types of judgment: when AI concerns depress valuations, if long-term prospects remain intact, the decline creates buying opportunities—SAINTS added to its positions in Amadeus and Intuit in early 2026; when evidence demands a change of mind, one must exit decisively—Diageo saw weakening consumer demand, increased investment requirements, and pressure on free cash flow, ultimately leading to a dividend cut, prompting SAINTS to sell. He admits this was their mistake: failing to respond decisively enough to signals of deteriorating fundamentals. A long record of dividends or a high dividend yield is only attractive if you have confidence in the long-term economics of the business.

Hodges also acknowledges a frustrating fact: although SAINTS holdings generate twice the free cash flow of the global market average, achieve return on investment 70% higher than average, and grow dividends faster than the market, the fund has not delivered the capital growth shareholders expected over the past few years. Share price returns from 2022 to 2026 were -7.2%, 12.8%, 6.5%, 2.2%, and 11.2%, compared to the FTSE Global Index's returns of -3.6%, 11.1%, 20.2%, 17.4%, and 18.7% respectively. Whether the flywheel metrics serve as diagnostic tools or merely post-hoc descriptions remains unanswered. His conclusion is that correctly identifying a technological revolution is very different from identifying who will benefit from it, and the value of the flywheel framework lies in forcing investors to continue asking whether cash flows are still growing after the narrative has faded.

Sources:bailliegifford.com

Research

Ben Lofthouse, Head of Global Equity Income at Janus Henderson: Dividend Growth and Valuation Discounts in International Markets Support Diversified Allocation

Material
2026-11-04 19:21 GMT+8

In an interview published in November 2026, Ben Lofthouse, Head of Global Equity Income at Janus Henderson, outlined three key allocation advantages currently present in international equity markets: higher dividend income, lower valuation levels, and a natural diversification away from AI concentration risks.

Lofthouse pointed out that performance in international markets is not driven by a single theme but is instead supported by broad participation across sectors such as technology, financials, and energy. Citing data from his team’s latest dividend report, he noted that the Q2 2026 report showed global dividends grew 7% year-over-year in US dollar terms. The financial sector saw growth of approximately 11%, while the technology sector surged by 20%. Given the large size of these two sectors, their dividend growth makes a significant contribution to investors' actual income, helping to offset the erosion caused by inflation.

Regarding valuations, Lofthouse observed that price-to-earnings ratios in international markets are generally lower than in the United States. He emphasized that international indices have low exposure to areas potentially disrupted by AI, such as software and online services; this means that if AI spending slows, these markets would face relatively limited headwinds. Conversely, sectors like commodities and energy in international markets have suffered from underinvestment in recent years, providing an additional source of diversification.

A notable shift is occurring in the focus of AI investment. Lofthouse observed that while the wave of AI capital expenditure has not abated, market attention is increasingly turning toward companies that can translate AI into productivity gains. This trend spans sectors including financial services, transportation, manufacturing, and healthcare. He believes that companies in these industries are well-positioned to improve future profitability through AI, representing a new avenue for earnings growth following the initial infrastructure investment cycle.

However, Lofthouse clearly identified associated risks. On one hand, the scale of AI capital expenditure is enormous, while many other industries are not making commensurate investments. On the other hand, certain stock markets exhibit significant concentration, such as in the United States and parts of emerging markets. He also highlighted the role of geopolitical conflicts in driving up inflation and energy prices, suggesting that if tensions ease, consumer sectors—which have lagged in performance—could see a recovery. His conclusion is that maintaining diversified allocation and focusing on balance sheet strength remain crucial in the current environment.

Sources:US Advisor

Research

Wireless Chipmaker Nordic Launches nRF71 SoC: Integrated NPU, Standby Power One-Twentieth of Wi-Fi Companion ICs

Material
Verified 2026-11-11 18:45 GMT+8

Wireless chipmaker Nordic Semiconductor has announced the nRF71 series of System-on-Chip products, integrating low-power Wi-Fi, Bluetooth LE, Thread protocols, and edge AI capabilities into a single chip. Nordic states that the nRF71 series reduces standby power consumption to as low as one-twentieth compared to its existing nRF70 series Wi-Fi companion ICs. This improvement aims to enable battery-powered devices in smart homes, wearables, and industrial IoT to maintain network connectivity while performing local AI inference without requiring additional external processors.

The first products in the nRF71 series include the nRF7120 and nRF7120E models. The nRF7120 supports dual-band Wi-Fi 6, while the nRF7120E adds tri-band Wi-Fi 6E support and integrates a Neural Processing Unit named Axon. According to Nordic, the Axon NPU can accelerate machine learning model execution by 10 to 17 times compared to using only a 256MHz Arm Cortex-M33 processor. This makes AI functions such as keyword spotting, anomaly detection, and image classification practical on battery-powered devices. Additionally, the series supports running ultra-small AI models trained on developers' own data via Nordic's Neuton software tools directly on the chip.

In terms of hardware specifications, the nRF71 series features 4MB of on-chip non-volatile memory and 1MB of user-available RAM. To support concurrent multi-protocol operations, the chip contains independent RF circuits and processing blocks, allowing the Wi-Fi radio and the 2.4GHz multiprotocol radio for Bluetooth LE or Thread to receive signals simultaneously. This means devices can process local short-range communication tasks while maintaining cloud Wi-Fi connections, reducing latency and energy consumption caused by waiting for different protocol processes to complete. Samples of the nRF71 series are currently being provided to select customers, and the products will be showcased at electronica in Munich, Germany, from November 10 to 13, 2026.

Sources:Nordic Semiconductor

Research

MIT and NYU Researchers Develop FLEX System Using GPT-4o to Infer General Robot Preferences from Short Commands

Topic · FLEXQuick take
2026-11-11 17:09 GMT+8

Researchers at the Massachusetts Institute of Technology (MIT) and New York University have developed an AI-assisted robot learning system called FLEX (Flexible Feature Editing Pipeline). The system addresses a key limitation in conventional robot motion planning: while a robot may understand a specific command, it often fails to recognize the underlying reason or generalize the constraint to similar situations. For example, if told not to spill coffee on a laptop, a standard robot might not avoid spilling on other nearby electronic devices unless explicitly instructed.

FLEX modifies the internal representation a robot uses when planning its movements by employing three operations: Add, Generalize, and Mask. It uses OpenAI’s GPT-4o language model to identify relationships between objects and interpret the intentions behind human corrections. If a robot carrying a bowling ball is warned not to drop it on a mug, the system infers that fragile objects should be avoided generally, rather than treating the mug as the only potential hazard. Similarly, a warning about carrying a lit candle near paper can be generalized into a preference for avoiding flammable materials.

The research was led by Andreea Bobu of MIT’s Computer Science and Artificial Intelligence Laboratory, with Audrey Lee, Nathaniel Dennler and Ilia Sucholutsky. The team evaluated the approach in two-dimensional navigation environments and simulated tabletop manipulation tasks involving a seven-degree-of-freedom Franka robotic arm. In sequential experiments, FLEX maintained higher normalized reward scores than approaches restricted to adding or removing features, which could accumulate unnecessary constraints or fail to generalize earlier instructions.

A separate study involved 12 participants who watched simulated robot movements and entered corrections in natural language. Participants rated FLEX more highly than comparison methods for understanding intent, with statistically significant differences reported for several measures. However, the distinction between simulation and physical deployment remains important. The participant study used simulated robot behavior, with real-robot demonstrations incorporated into offline evaluation. The researchers have not yet demonstrated a complete autonomous household robot operating under these conditions in real-time.

Sources:eeNews Europe

Research

Taiwan's NCHC Completes AMD Platform Procurement, Expands TAIWAN AI RAP with Fine-Tuning Services

Topic · 台湾NCHC AI平台Quick take
2026-11-11 16:47 GMT+8

According to iThome, Zhang Chaoliang, director of the National Center for High-performance Computing (NCHC) in Taiwan, announced that the center has completed the bidding process for AMD computing platforms. This move introduces an alternative to existing GPU architectures for AI workloads. The NCHC currently operates Nano 4 and Nano 5 platforms for model training and plans to adopt next-generation architectures like Vera Rubin.

As part of its "Sovereign AI" strategy, the NCHC's TAIWAN AI RAP platform saw user numbers nearly double over the past year. Zhang stated that the platform added fine-tuning services this year, allowing users to adapt models with their own data. The center is also shifting focus toward supporting Agent development and verification, planning a B2B2C model where partners deliver applications to end-users.

Technically, Academia Sinica and Chung Yuan Christian University demonstrated TAIWAN GBA, a behavior recognition system based on YOLOv9 that identifies actions like falls and fights, addressing limitations in Vision-Language Models (VLMs). The NCHC aims to combine its compute power, Academia Sinica's research, and local corpora to develop multimodal AI tailored to Taiwan's needs.

Sources:iThome

Research

Janus Henderson Portfolio Manager Daniel Lyons: Biotech Pullback is Technical Deleveraging, Not Fundamental Deterioration

Material
2026-11-10 00:01 GMT+8

In remarks published on November 9, 2026, Janus Henderson Portfolio Manager Daniel Lyons stated that the sharp sell-off in the biotechnology sector following the midterm elections should not be interpreted as a deterioration in fundamental conditions. His core assessment is that this pullback is the result of broader deleveraging dynamics and technical factors, with the steepest declines concentrated in genomics, diagnostics, and life science tools companies whose valuations had been pushed higher by enthusiasm for artificial intelligence (AI), rather than indicating a fundamental turning point across the entire biotechnology industry.

Lyons pointed out that surging interest in AI-enabled drug development and personalized medicine over recent months created a clear divergence in valuations within the sector. AI-linked life science tools stocks rallied sharply through September, while many traditional biotechnology companies—including those advancing promising new therapies—continued to trade at attractive valuations. He believes the volatility since early October may help investors refocus on fundamentals and identify attractive entry points into these high-quality biotechnology companies.

This assessment rests on a historical comparison. Lyons notes that while the biotechnology sector’s strong performance over the past year might evoke memories of the speculative froth preceding the 2021 downturn, today’s landscape features several key differences. Having endured the post-pandemic drawdown and tighter financing conditions associated with the Federal Reserve’s aggressive rate-hike cycle in 2022–2023, he views biotechnology as a healthier, more disciplined sector: many companies now possess stronger balance sheets, more experienced management teams, and a greater number of commercial-stage assets than they did a few years ago. Furthermore, whereas the boom of 2020–2021 was broad-based, current valuation excesses are largely concentrated in a relatively narrow group of companies.

On the policy front, Lyons argues that the implications of a comprehensive Democratic sweep in the midterm elections for healthcare are more subtle—and potentially more benign—than the initial market reaction suggested. He assesses that Democrats are more likely to shift their policy agenda toward expanding healthcare coverage and protecting existing benefits, rather than pursuing new drug-pricing legislation. Specifically, the enhanced Affordable Care Act (ACA) premium subsidies, which expired at the end of 2025, are expected to be a priority for restoration. Meanwhile, debates regarding Medicaid are primarily focused on federal work requirements that could lead to coverage losses. Increased stability in coverage could provide incremental support for overall healthcare utilization and patient volumes.

For the medical technology (medtech) sector, Lyons sees this as a potential positive factor, given that the sector has previously operated in a challenging environment marked by strained hospital budgets and coverage losses, pushing valuations to historically depressed levels. Regarding biopharmaceuticals, he believes Democrats are more likely to defend existing Medicare drug-price negotiation provisions enacted under the Inflation Reduction Act rather than push for new legislation. This could offer some reassurance to pharmaceutical and biotechnology companies, as policy uncertainty surrounding drug pricing had previously periodically weighed on valuations.

Sources:US Advisor

Research

Microsoft Releases .NET 11 and C# 15 with New Type Safety Features and Built-in AI Agent Support

Topic · .NET 11Material
2026-11-11 16:48 GMT+8

US technology company Microsoft has officially released .NET 11 at the .NET Conf 2026 developer conference. This version defaults to the C# 15 programming language. .NET is Microsoft's cross-platform framework widely used for web, desktop, and mobile applications. The update focuses on enhancing language safety, execution efficiency, and native support for AI agent development.

In terms of language features, C# 15 introduces "Union Types" and "Closed Hierarchies." Union types allow developers to explicitly declare which specific types a value can belong to—for example, a query result might be either "product data" or a "no-product response." The compiler can then check if the code handles all possible cases, reducing errors caused by missed branches. Closed hierarchies restrict class inheritance to within the same assembly, giving the compiler full visibility of subclass structures to further prevent logical gaps.

Regarding execution performance, the .NET 11 SDK now defaults to using Native AOT (Ahead-of-Time compilation) for its Command Line Interface (CLI) startup programs. Common commands like checking versions or viewing help are executed directly as native code, bypassing the traditional managed .NET runtime and significantly reducing startup time. Additionally, the MSBuild server remains active between commands, minimizing repeated initialization overhead during frequent builds and tests.

Addressing the growing trend in AI application development, Microsoft has prioritized AI agent support in the .NET roadmap. The .NET 11 SDK includes built-in templates for MCP (Model Context Protocol) server projects, allowing developers to quickly create tool services callable by AI agents. Blazor frameworks have also added AI interface components supporting streaming conversations and tool execution results, along with tools like .NET Skills to assist AI agents in coding and testing.

It is worth noting that .NET 11 is a Standard Term Support (STS) version with two years of support, while .NET 10 remains a Long-Term Support (LTS) version. Microsoft also announced that support for .NET 8 and .NET 9 ended on November 10, requiring enterprises to upgrade to supported versions to continue receiving security updates.

Sources:iThome

Research

Higress Team: After Two Supply-Chain Poisonings of LiteLLM, the Blast Radius Depends on Default Permissions in Extension Code

Material
Verified 2026-11-11 18:00 GMT+8

On October 21, 2026, LiteLLM version 1.107.0 was isolated just 1.7 hours after its release on PyPI; however, attackers had already used stolen AWS credentials to invoke bedrock:ListFoundationModels within less than five hours. This marked the second supply-chain poisoning of LiteLLM within six months, with attackers targeting nearly identical assets: cloud credentials, SSH private keys, Kubernetes (K8s) tokens, and all upstream model API keys centrally stored in the proxy database.

In its incident analysis, the Higress team raised a more fundamental question than 'how to prevent poisoning': since poisoning will inevitably occur, how can we ensure that malicious code gains nothing even if it infiltrates the system? Their answer is that the true determinant of the blast radius is the amount of default privilege held by extension code.

This assessment stems from a comparison of two proxy architectures. LiteLLM operates as a Python process where third-party dependencies and custom callbacks are essentially in-process Python code, possessing the exact same privileges as the proxy process itself—meaning files readable, environment variables visible, networks connectable, and subprocesses launchable by the proxy are equally accessible to malicious code. The infostealer in version 1.107.0 exploited this precisely: utilizing .pth startup hooks to execute automatically upon interpreter launch, it read ~/.aws/credentials, ~/.ssh/id_rsa, K8s ServiceAccount Tokens, and even connected directly to the proxy database to query the LiteLLM_CredentialsTable.

Higress's solution involves constraining extension code within a Wasm sandbox. Its data plane is based on Envoy (C++), with core gateway capabilities implemented natively by Envoy; AI proxying, authentication, rate limiting, content security, and other extension capabilities run as Wasm plugins within Envoy's embedded Wasm virtual machine. The fundamental difference between Wasm plugins and Python dependencies lies in privilege: Python dependencies default to full process privileges, whereas Wasm plugins possess only the limited capabilities explicitly exposed by the host.

Specifically, according to the proxy-wasm-cpp-host source code, the Wasm sandbox revokes the following capabilities: No file system access (path_open always returns NOENT, preventing plugins from reading any files); Environment variables are invisible by default (environ_get returns only variables explicitly declared in the WasmPlugin vmConfig.env); No arbitrary network access (sock_send and sock_recv return ENOTSUP, requiring plugins to request Envoy to send requests on their behalf, which checks for the existence of target clusters); No process spawning (no fork/exec, preventing daemon generation or Pod deployment); Memory isolation (each plugin accesses only its own linear memory, unable to read TLS private keys from the Envoy process or data from other plugins).

The Higress team simultaneously acknowledges that sandboxes are not silver bullets: plugins can still read and write the requests and responses they handle, creating opportunities to exfiltrate data via responses; ai-proxy must hold upstream API Keys to forward requests. Therefore, they recommend combining this approach with least privilege (splitting capabilities across different plugins so authentication plugins only hold consumer credentials and model proxies only hold model keys), version pinning (referencing plugins by version number plus sha256 rather than latest), and managed credentials (using KMS encryption for key management in the Enterprise Edition).

Sources:Alibaba Cloud Community

Research

US CISA Mandates Patching for 6 Actively Exploited Vulnerabilities by Nov 13 or 24

Topic · CISA KEV名单漏洞Material
2026-11-11 12:25 GMT+8

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) announced on November 10 that it has added six new vulnerabilities with evidence of active exploitation to its Known Exploited Vulnerabilities (KEV) catalog, mandating that U.S. federal agencies patch them within specific deadlines.

Affected products include Microsoft Outlook, Google Chrome, several Barracuda Network devices (such as ADC, WAF, WSG, BMA, WCC, and Email Security Gateway), Cisco BroadWorks, and Openwave Webtop Media Transcoder.

CISA set different compliance dates based on the urgency of the flaws. Two vulnerabilities affecting Barracuda products (CVE-2026-111560 and CVE-2026-116606) are considered most critical, requiring federal agencies to complete patching by November 13. The remaining four vulnerabilities have a deadline of November 24.

Sources:iThome

Research

US Bridal Retailer David's Bridal Plans to Spin Off AI Platform Pearl via SPAC at $350M Valuation

Quick take
2026-11-11 14:50 GMT+8

US bridal retailer David's Bridal announced on October 22, 2026, that it plans to spin off its AI wedding planning platform, Pearl, through a merger with a Special Purpose Acquisition Company (SPAC), with an estimated enterprise value of $350 million. Pearl attracted over 110,000 users within nine months of launch without marketing spend. The company states that users who engaged with Pearl showed higher purchase completion rates in physical stores. The transaction is currently in the planning stage.

Sources:TechOrange 科技報橘

Research

Baillie Gifford LTGG Investment Manager Douglas Jack: No Sector Coverage, Let Curiosity Find the Exceptions

Material
2026-11-05

Facing a candidate pool of approximately 4,000 to 5,000 companies, no one can pretend to cover them all. Douglas Jack, an investment manager for Baillie Gifford’s Long Term Global Growth (LTGG) fund, frankly admits that LTGG lacks a rigid, step-by-step idea generation process—a fact that sometimes makes explaining their approach to outsiders difficult. Instead, he focuses on several principles he believes distinguish LTGG from other managers, explaining why these principles contribute to long-term outperformance.

With roughly 4,000 to 5,000 candidates currently under consideration, LTGG does not pretend to cover every name. Jack argues that the traditional practice of assigning each analyst a specific industry has two fundamental flaws. First, exceptions are not evenly distributed across the economy; locking someone into a 'box' with nothing interesting to see is pointless. Second, this allocation creates poor incentives: analysts feel pressured to constantly pitch mediocre names just to appear busy. Since LTGG’s mandate is to invest in a small number of key outliers, any time spent away from high-probability candidates harms returns.

Consequently, LTGG analysts are all generalists, with no one holding formal 'coverage.' However, everyone has technologies, industries, or themes they follow closely over time—Jack looks around the office and sees deep expertise in luxury goods, robotics, payments, semiconductors, AI, and cryptocurrency. No one is forced to learn these topics; instead, they are encouraged to 'follow their noses' and pull on threads that interest them.

Jack cites a specific example: the analyst who first brought Nvidia to the team in 2016 was initially interested because of VR headsets, not because she was assigned as a semiconductor analyst. While researching Nvidia, she pulled on the AI thread a bit further and realized, though ultimately underestimating the scale, just how profound a change it could bring.

Another concept is open-mindedness. Because analysts work in their own way, the traditional boundaries imposed by benchmark indices play a smaller role in their thinking. Jack himself has spent considerable time over the past two years focusing on AI, which has given him a more holistic view across multiple sectors such as software, semiconductors, and energy. This approach makes it easier to spot discontinuities or shifts that a structured sectoral method might miss. He points to their 2004 investment in Amazon, which did not fit the mold of a traditional retail company at all; the investment thesis was based on the potential of a business with ever-declining input costs and Jeff Bezos’s long-term vision.

For outlier hunters, Jack believes that 'things going wrong' is a necessary cost of the pursuit. The risk that keeps LTGG analysts awake at night is not losing money, but missing an outlier. They regularly review missed opportunities and try to identify patterns, helping them overcome the inner voice that says 'everything could go wrong,' allowing them to bring riskier ideas to the table.

Sources:bailliegifford.com

Research
Next →
Sign in / Sign up
ReadingResearchRadarInvestment framework

Lyons explicitly acknowledges that it is too early to determine how a Democratic-led Congress will interact with the Trump Administration or what will ultimately translate into concrete healthcare policy. Drug pricing will remain a political focal point, but he considers the likelihood of significant changes within the current framework to be low. His conclusion is that healthcare continues to offer a combination of long-term structural drivers (medical innovation and demographic aging) and defensive characteristics, alongside diversified return sources beyond the AI theme. Recent volatility may serve to redirect investor attention toward valuation opportunities.

Regarding the release process, starting from v2.2.4, Higress adopts immutable releases: build once, review snapshots, and promote by digest. Official version tags are immutable—if a tag with the same name already exists but has a different digest, the release fails immediately. This contrasts with the LiteLLM incident where latest and main-stable tags were repointed to malicious images.

It should be noted that this argument originates from Higress itself; its performance comparison data (such as a 68.8%–97.1% reduction in latency for 1 MB requests) represents vendor self-testing without independent reproduction. While sandbox attributes at the architecture level can be verified via open-source code, the overall security effectiveness of the solution ultimately depends on actual operational discipline and the integrity of the plugin ecosystem.

Simultaneously, LTGG has built a culture from top to bottom of openly and honestly discussing disappointing investments. Jack explicitly states that while they learn from these mistakes, discussing errors with candor and a degree of levity reinforces the message that taking risks and learning from them is encouraged within LTGG. He acknowledges that they have made far more mistakes than successes—from solar panel manufacturers to 3D printers to alternative protein companies—but this has not impaired their ability to generate long-term outperformance for clients over the past twenty-plus years.

On the information front, LTGG has a growing private companies division, originating from within LTGG more than a decade ago. Talking to founders building the future in private markets not only helps prepare for their IPOs and understand competitive threats to existing holdings, but also introduces new ideas. Jack gives the example of Baillie Gifford’s longstanding holding in SpaceX, which gave them access to the broader space economy and other players, including current holding Rocket Lab.

LTGG also has a strong culture of getting on the ground. It is not uncommon for analysts to spend weeks on the road meeting dozens of companies, nor is it rare for them to embed themselves in a key location for months to truly understand an emerging theme. Jack specifically highlights China: while many other managers have given up on Chinese opportunities, LTGG retains an on-the-ground presence led by team member Linda Lin, enabling them to meet many companies that other managers have cut ties with.

Jack concludes that LTGG does not have a neat idea-generation funnel. They do not systematically screen thousands of companies, nor do they expect every analyst to produce a fully formed pitch from an assigned sector. Instead, they create conditions where unusual ideas are more likely to surface: giving curious people the freedom to follow interesting threads, encouraging consideration of outcomes that initially seem improbable, and supporting ideas when they don’t work out. This may look unstructured, but it is not aimless—everything is anchored to the same goal: finding the small number of exceptional companies capable of driving long-term returns. Jack admits they won’t find every outlier, but believes this 'follow your nose' approach, being prepared to look wrong along the way, improves their chances of finding the few that really matter.