Daily Issue
Vol. I — No. 8
05 · 05
Tuesday, 5 May 2026
Generated 2026-05-05 11:06
google/gemini-2.5-flash-lite-preview-09-2025
因为荒谬，我才相信。 — 德尔图良 48 items · 4 sections
§ 0

The Morning

Local weather 1
This morning in
London
Overcast
Today's range
17.0°↓11.3°
currently 15.0°
Feels
12.2°
Rain
34%
Wind
12 km/h
Humid
47%
Rise
05:24
Set
20:29
§ I

US Stocks

Pre-market signal radar 12
US pre-market radar
premarket 2026-05-05
1 Bullish
0 Bearish
11 Neutral
Sector Tape
Hyperscale Cloud 4 names
64 Top: GOOGL · Neutral · RS +1.9% Bullish 1 / Bearish 0 / 5d +3.8%
Compute Mining 4 names
63 Top: IREN · Neutral · RS -0.9% Bullish 0 / Bearish 0 / 5d +2.0%
Networking Equipment 4 names
62 Top: CIEN · Neutral · RS -0.7% Bullish 0 / Bearish 0 / 5d +0.3%
Servers and Thermal Management 2 names
60 Top: VRT · Neutral · RS -0.2% Bullish 0 / Bearish 0 / 5d +0.3%
Foundry 2 names
58 Top: INTC · Neutral · RS +5.2% Bullish 0 / Bearish 0 / 5d +5.9%
Battery and Energy Storage 3 names
56 Top: SLDP · Neutral · RS -8.9% Bullish 0 / Bearish 0 / 5d -7.2%
Manufacturing 4 names
56 Top: FLEX · Neutral · RS +5.8% Bullish 0 / Bearish 0 / 5d +5.4%
Energy Infrastructure 1 names
47 Top: VST · Neutral · RS -5.9% Bullish 0 / Bearish 0 / 5d -3.4%
Ticker Setup Move Score Evidence Quality
ORCL Oracle Hyperscale Cloud
Neutral Gap up + news Low confidence
+1.9% $183.70 5d +4.2%
66 sector positive RS +2.3%

Watchlist item from +1.9% vs previous close, positive sector tape, 3 recent headline(s).

I’m Calling A Bottom For Oracle (NYSE:ORCL) - Seeking Alpha Needs fresh price/news confirmation before becoming an actionable setup.
quote: delayed fallback news: fresh financials: fresh news: 3
WULF TeraWulf Compute Mining
Neutral Gap up + news Low confidence
+1.4% $22.59 5d +4.0%
63 sector positive RS +1.1%

Watchlist item from +1.4% vs previous close, positive sector tape, 3 recent headline(s).

TeraWulf executives head to four investor conferences in May - Stock Titan Needs fresh price/news confirmation before becoming an actionable setup.
quote: delayed fallback news: fresh financials: fresh news: 3
INTC Intel Foundry
Neutral Gap up + news Low confidence
+3.5% $99.18 5d +12.7%
62 sector flat RS +11.9%

Watchlist item from +3.5% vs previous close, 3 recent headline(s).

Intel Stock Drops. Should You Be Worried? - Barron's 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 24 24/24filings: 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:2605.02661v1Lead article

AcademiClaw: When Students Set Challenges for AI Agents

Junjie Yu, Pengrui Lu, Weiye Si, Hongliang Lu, Jiabao Wu

cademiClaw introduces a new bilingual benchmark sourced from real, complex, long-horizon academic workflows that students find current AI agents fail to solve. This benchmark features 80 challenging tasks across 25+ professional domains, including GPU-intensive work, executed in isolated sandboxes and scored using multi-dimensional rubrics and safety audits. Its core contribution is shifting evaluation from assistant-level tasks to assessing AI agents on genuine, high-level academic capabilities.

Read abstract →Full PDFLINNET · SCORE9RELEVANCE
Task complexity comparison: Claw-Eval vs. AcademiClaw. Claw-Eval focuses on assistant-level routines, whereas AcademiClaw targets tasks requiring deep academic expertise and sustained multi-step reasoning.
Task complexity comparison: Claw-Eval vs. AcademiClaw. Claw-Eval focuses on assistant-level routines, whereas AcademiClaw targets tasks requiring deep academic expertise and sustained multi-step reasoning.
Figure 1 . Distribution of Code Smell Counts. This box plot illustrates the distribution of counts for the most prevalent code smells, sorted in descending order by their mean value. Each box represents the interquartile range (IQR), with the central line denoting the median and the whiskers extending to 1.5 times the IQR. Points beyond the whiskers are plotted as individual outliers. The abbreviations for the code smells are as follows: TMB (Too Many Branches), PAU (Potential Improper API Usage), UD (Unstable Dependency), SF (Scattered Functionality), RFC (High Response for a Class), HCC (High Cyclomatic Complexity), TF (Temporal Field), and LCM (High Lack of Cohesion of Methods).
Figure 1 . Distribution of Code Smell Counts. This box plot illustrates the distribution of counts for the most prevalent code smells, sorted in descending order by their mean value. Each box represen…
cs.AIarxiv:2605.02741v1

AI-Generated Smells: An Analysis of Code and Architecture in LLM and Agent-Driven Development

Yuecai Zhu, Nikolaos Tsantalis et al.

This paper systematically audits technical debt in AI-generated software, revealing that LLMs introduce a distinct "machine signature" of defects rather than eliminating flaws. The core finding is a **Reasoning-Complexity Trade-off**: more capable models produ…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.02592v1

Foundation-Model-Based Agents in Industrial Automation: Purposes, Capabilities, and Open Challenges

Vincent Henkel, Felix Gehlhoff et al.

This paper systematically surveys the literature to examine the current state, capabilities, and challenges of foundation-model-based agents in industrial automation. The core contribution is synthesizing findings from 88 relevant studies, revealing that most …

abstractpdfLINNET · SCORE9RELEVANCE
The steps of our approach: (1) assign different personas to the language models (LMs) using default, benevolent or malicious system prompts, (2) conduct pre-game persona assessment and identify core implicit traits, (3) agents compete in multi-turn social dilemma games, (4) post-game assessment quantifies effects of misalignment contagion and our steering with implicit traits (SIT) intervention.
The steps of our approach: (1) assign different personas to the language models (LMs) using default, benevolent or malicious system prompts, (2) conduct pre-game persona assessment and identify core i…
cs.AIarxiv:2605.02751v1

Mitigating Misalignment Contagion by Steering with Implicit Traits

Maria Chang, Ronny Luss et al.

This paper investigates "misalignment contagion," the spread of undesirable behavior between language models (LMs) in multi-agent, multi-turn interactions, observing that LMs become more anti-social after playing social dilemma games. The core contribution is …

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.02572v1

On Training Large Language Models for Long-Horizon Tasks: An Empirical Study of Horizon Length

Sunghwan Kim, Junhee Cho et al.

This paper empirically investigates the impact of task horizon length on training Large Language Models (LLMs) for long-horizon tasks. By controlling for decision rules and reasoning structures, the authors demonstrate that increasing horizon length alone sign…

abstractpdfLINNET · SCORE9RELEVANCE
A summary of our contributions. In this work, we study the training of long-horizon LLM agents from a horizon-centric perspective and identify horizon length as a fundamental bottleneck. We show that horizon reduction stabilizes RL and strengthens the tendency toward horizon generalization on longer tasks with similar reasoning difficulty.
A summary of our contributions. In this work, we study the training of long-horizon LLM agents from a horizon-centric perspective and identify horizon length as a fundamental bottleneck. We show that …
№06
cs.AI
9

ORPilot: A Production-Oriented Agentic LLM-for-OR Tool for Optimization Modeling

Guangrui Xie

ORPilot is an agentic LLM system designed to translate ambiguous, real-world business problems with raw data into solver-ready optimization models for production use. Its core cont…

abstractpdf
№07
cs.AI
9

Strategy-Aware Optimization Modeling with Reasoning LLMs

Ruiqing Zhao, Fengzhi Li et al.

This paper introduces SAGE, a framework that explicitly incorporates modeling strategies into the training of Large Language Models (LLMs) for optimization programming. SAGE utiliz…

abstractpdf
№08
cs.LG
9

Beating the Style Detector: Three Hours of Agentic Research on the AI-Text Arms Race

Andreas Maier, Moritz Zaiss et al.

This paper demonstrates the efficiency of modern agentic research tools by reproducing and extending a recent NLP study in just three hours, with the human acting only as a reviewe…

abstractpdf
№09
cs.LG
9

Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models

Inoussa Mouiche

The paper introduces **Gradient-Gated Preference Optimization (Gate-DPO)** to stabilize Direct Preference Optimization (DPO) training, which suffers from a "squeezing effect" causi…

abstractpdf
№10
cs.CL
9

ContextualJailbreak: Evolutionary Red-Teaming via Simulated Conversational Priming

Mario Rodríguez Béjar, Francisco J. Cortés-Delgado et al.

ContextualJailbreak introduces an evolutionary red-teaming strategy to automatically discover multi-turn jailbreak attacks that exploit contextual priming in LLMs. It performs evol…

abstractpdf
See all 20 papers →
§ III

The Town Square

Hacker News 7
410
pts
Top story
How OpenAI delivers low-latency voice AI at scale

OpenAI details the technical strategies, including optimized model architectures and efficient infrastructure, used to achieve low-latency, high-quality voice AI performance at a massive scale.

openai.com4 May discuss on HN →
288
Google Chrome silently installs a 4 GB AI model on your device without consent
thatprivacyguy.com5 May
270
Train Your Own LLM from Scratch
github.com5 May
198
Securing a DoD contractor: Finding a multi-tenant authorization vulnerability
strix.ai4 May
162
Let's talk about LLMs
b-list.org4 May
117
OpenAI, Google, and Microsoft Back Bill to Fund 'AI Literacy' in Schools
404media.co4 May
105
Sierra Raises $950M at $15B Valuation
sierra.ai4 May
compiled overnight by google/gemini-2.5-flash-lite-preview-09-2025 · end of issue no. 8 · thank you for reading