Daily Issue
Vol. I — No. 20
28 · 05
Thursday, 28 May 2026
Generated 2026-05-28 12:54
google/gemini-2.5-flash-lite-preview-09-2025
生活不止眼前的苟且，还应该有诗和远方的田野。 — 许巍 48 items · 4 sections
§ 0

The Morning

Local weather 1
This morning in
London
Partly cloudy
Today's range
31.7°↓15.7°
currently 31.2°
Feels
31.7°
Rain
10%
Wind
11 km/h
Humid
32%
Rise
04:52
Set
21:03
§ I

US Stocks

Pre-market signal radar 12
US pre-market radar
premarket 2026-05-28
9 Bullish
0 Bearish
3 Neutral
Sector Tape
Compute Mining 4 names
73 Top: WULF · Bullish · RS +6.4% Bullish 3 / Bearish 0 / 5d +31.8%
Servers and Thermal Management 2 names
73 Top: DELL · Bullish · RS +6.5% Bullish 1 / Bearish 0 / 5d +14.4%
Battery and Energy Storage 3 names
70 Top: EOSE · Bullish · RS +17.8% Bullish 1 / Bearish 0 / 5d +23.8%
Networking Equipment 4 names
69 Top: CIEN · Neutral · RS +9.8% Bullish 1 / Bearish 0 / 5d +16.2%
Manufacturing 4 names
68 Top: FLEX · Neutral · RS +3.3% Bullish 2 / Bearish 0 / 5d +9.7%
Foundry 2 names
68 Top: INTC · Neutral · RS -2.7% Bullish 1 / Bearish 0 / 5d +8.8%
Hyperscale Cloud 4 names
63 Top: ORCL · Neutral · RS -1.6% Bullish 0 / Bearish 0 / 5d +2.3%
Energy Infrastructure 1 names
51 Top: VST · Neutral · RS +21.5% Bullish 0 / Bearish 0 / 5d +18.9%
Ticker Setup Move Score Evidence Quality
DELL Dell Technologies Servers and Thermal Management
Bullish Gap up + news High confidence
+4.8% $319.93 5d +29.8%
80 sector positive RS +21.8%

Bullish setup from +4.8% vs previous close, positive sector tape, 3 recent headline(s).

5-Day Rally Sends Dell Technologies Stock Up 30% - Trefis Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
CIFR Cipher Mining Compute Mining
Bullish Sector tailwind Medium confidence
-1.8% $24.72 5d +33.8%
72 sector positive RS +8.4%

Bullish setup from -1.8% vs previous close, positive sector tape, 3 recent headline(s).

Cipher Digital Stock Surges 34%, With A 5-Day Winning Spree - Trefis Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
HUT Hut 8 Compute Mining
Bullish Sector tailwind Medium confidence
+0.3% $118.00 5d +26.1%
72 sector positive RS +0.6%

Bullish setup from positive sector tape, 3 recent headline(s).

Hut 8 Stock Rockets 26% With 5-Day Winning Streak - Trefis Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
INTC Intel Foundry
Bullish Sector tailwind Medium confidence
-1.3% $120.18 5d +9.9%
70 sector positive RS -1.6%

Bullish setup from -1.3% vs previous close, positive sector tape, 3 recent headline(s).

Why Intel Stock Dropped Today - The Motley Fool Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
FLNC Fluence Energy Battery and Energy Storage
Neutral Sector tailwind Medium confidence
+0.6% $21.34 5d +18.5%
70 sector positive RS +12.4%

Watchlist item from positive sector tape, 3 recent headline(s).

Why Fluence Energy (FLNC) stock is trading up today - MSN Needs fresh price/news confirmation before becoming an actionable setup.
quote: delayed fallback news: fresh financials: fresh news: 3
quotes: nasdaq 24 24/24news: google_news_rss 22, gdelt 1 23/24filings: sec 24 24/24, fallback 24

Generated from public market data and news for research and education. Not financial advice; data may be delayed, incomplete, or wrong.

§ II

From the arXiv

arXiv preprints 10 of 20
cs.AIarxiv:2605.28632v1Lead article

Blind PRNG Hijacking: An Undetectable Integrity-Preserving Attack Against LLM Watermarking

Ziyang You, Huilong He, Xiaoke Yang, Xuxing Lu

his paper introduces **SeedHijack**, a novel, undetectable attack against LLM watermarking that targets the underlying Pseudo-Random Number Generator (PRNG) in the supply chain. The core method replaces the PRNG to bias green-list selection without altering the output tokens or requiring knowledge of the watermark key or detector. This results in an integrity-preserving attack that amplifies the watermark signal while remaining statistically independent of content-side detection statistics.

Read abstract →Full PDFLINNET · SCORE9RELEVANCE
Dual-flow comparison of watermarked LLM inference. Top : benign watermarked generation where the watermark adds bias + δ +\( \delta \) to green-list tokens G G in logit space. Bottom : SeedHijack attack where a malicious PRNG replaces the honest one at the supply-chain layer, biasing sampling toward a target set T T in probability space. Because G G and T T are statistically independent (green-list orthogonality), the watermark z z -score is preserved while the attacker gains content control.
Dual-flow comparison of watermarked LLM inference. Top : benign watermarked generation where the watermark adds bias + δ +\( \delta \) to green-list tokens G G in logit space. Bottom : SeedHijack attack where a malicious PRNG replaces the honest one at the supply-chain layer, bia…
(a) Numbers of reasoning and answer tokens. Qwen-4B, Qwen-32B, and Qwen-235B refer to Qwen3-VL-4B, Qwen3-VL-32B, and Qwen3-VL-235B-A22B. (b) Accuracy and speedup of Qwen3-VL-32B under different decoding methods. Original denotes standard decoding. SR-Q2B, SR-Q4B, SR-M7B, and SR-R4B denote SpecReason (Pan et al. , 2025 ) using Qwen3-VL-2B, Qwen3-VL-4B, MiMo-VL-7B-RL, and Qwen3-VL-R1-VL-4B as draft models, respectively. Speedup is normalized to Original.
(a) Numbers of reasoning and answer tokens. Qwen-4B, Qwen-32B, and Qwen-235B refer to Qwen3-VL-4B, Qwen3-VL-32B, and Qwen3-VL-235B-A22B. (b) Accuracy and speedup of Qwen3-VL-32B under different decodi…
cs.AIarxiv:2605.28678v1

DREAM-R: Multimodal Speculative Reasoning with RL-Based Refined Drafting, Precise Verification, and Fully Parallel Execution

Yunhai Hu, Zining Liu et al.

DREAM-R enhances speculative reasoning in multimodal models using a novel reinforcement learning objective, Speculative Alignment Policy Optimization (SAPO), to train draft models for generating concise and faithful reasoning steps. It incorporates a Threshold…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.28721v1

LiveBrowseComp: Are Search Agents Searching, or Just Verifying What They Already Know?

HuiMing Fan, Xiao Wang et al.

This paper introduces the **LiveBrowseComp** benchmark to diagnose whether LLM search agents genuinely search or merely verify their intrinsic knowledge. The core method involves analyzing agent behavior on the original BrowseComp dataset, revealing significan…

abstractpdfLINNET · SCORE9RELEVANCE
Overview of LiveBrowseComp. As models iterate, the knowledge required by a static benchmark is gradually absorbed into their parameters, so the effective difficulty of its questions collapses over time. By being constructed from up-to-date knowledge, LiveBrowseComp can effectively mitigate this erosion.
Overview of LiveBrowseComp. As models iterate, the knowledge required by a static benchmark is gradually absorbed into their parameters, so the effective difficulty of its questions collapses over tim…
Framework for automatic diagnosis of LLM memory systems. We first execute a memory system to construct an execution graph. Given a failed case, MemTrace performs step-by-step tracing over this graph to locate the faulty operation. This framework is general across different memory systems and enables faster failure attribution than human experts.
Framework for automatic diagnosis of LLM memory systems. We first execute a memory system to construct an execution graph. Given a failed case, MemTrace performs step-by-step tracing over this graph t…
cs.AIarxiv:2605.28732v1

MemTrace: Tracing and Attributing Errors in Large Language Model Memory Systems

Xinle Deng, Ruobin Zhong et al.

MemTrace introduces a novel framework to trace and attribute errors in large language model memory systems by transforming memory pipelines into executable memory evolution graphs. This allows for fine-grained tracking of information flow and systematic analys…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.28805v1

OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibration

Xinchen Zhang, Bowei Liu et al.

This paper introduces OmniVerifier-M1, a multimodal meta-verifier that uses symbolic outputs (like bounding boxes) as effective rationales for training, outperforming textual explanations. The core method involves decoupling the reinforcement learning objectiv…

abstractpdfLINNET · SCORE9RELEVANCE
Pipeline of two key findings. Left: the advantage of symbolic bounding boxes over textual explanations, enabling rule-based rewards to inherently prevent reward hacking and accelerate training. Right: the comparison between joint training and decoupled training.
Pipeline of two key findings. Left: the advantage of symbolic bounding boxes over textual explanations, enabling rule-based rewards to inherently prevent reward hacking and accelerate training. Right:…
№06
cs.AI
9

Position: Retire the "Positive Backdoor" Label -- Secret Alignment Requires Strict and Systematic Evaluation

Jianwei Li, Jung-Eun Kim

This paper argues for retiring the term "positive backdoor" and replacing it with "Secret Alignment" to describe trigger-activated hidden behaviors in AI models. The core contribut…

abstractpdf
№07
cs.AI
9

Rethinking Memory as Continuously Evolving Connectivity

Jizhan Fang, Buqiang Xu et al.

This paper introduces **FluxMem**, a novel memory framework for LLM agents that models memory as a **continuously evolving, heterogeneous graph**. FluxMem dynamically refines its t…

abstractpdf
№08
cs.AI
9

Technical Report: Exploring the Emerging Threats of the Agent Skill Ecosystem

Luca Beurer-Kellner, Aleksei Kudrinskii et al.

This paper analyzes 3,984 AI agent skills to uncover emerging security threats within the agent skill ecosystem. The core contribution is the identification of 76 confirmed malicio…

abstractpdf
№09
cs.AI
9

The Importance of Being Statistically Earnest: A Critical Re-evaluation of GSM-Symbolic

Dominika Agnieszka Długosz, Arlindo Oliveira et al.

This paper critically re-evaluates the GSM-Symbolic benchmark, arguing its conclusion of widespread LLM reasoning failure is statistically unsound. Using Generalised Linear Mixed M…

abstractpdf
№10
cs.AI
9

TRACER: Turn-level Regret Matching with Inner Reinforcement Credit for Cooperative Multi-LLM Reasoning

Chusen Li, Zhou Liu et al.

TRACER is a novel turn-level reinforcement framework designed to integrate reinforcement learning with multi-LLM cooperation. It uses a controller-regret layer employing regret mat…

abstractpdf
See all 20 papers →
§ III

The Town Square

Hacker News 7
1000
pts
Top story
YouTube to automatically label AI-generated videos

YouTube is implementing automatic labeling for videos that are identified as being substantially altered or generated by artificial intelligence to enhance transparency for viewers and creators.

blog.youtube27 May discuss on HN →
925
DuckDuckGo search saw 28% more visits after Google said people love AI mode
pcgamer.com27 May
665
Tech CEOs are apparently suffering from AI psychosis
techcrunch.com27 May
572
Canada to order military plane fleet from Sweden in shift from US suppliers
theguardian.com27 May
204
Training our own AI models
posthog.com27 May
139
Lombardy increases charges for the construction of data centres in green areas
en.ilsole24ore.com27 May
108
A Eureka machine that thinks like nature and explores what AI cannot
iisc.ac.in28 May
compiled overnight by google/gemini-2.5-flash-lite-preview-09-2025 · end of issue no. 20 · thank you for reading