Daily Issue
Vol. I — No. 10
07 · 05
Thursday, 7 May 2026
Generated 2026-05-07 11:30
google/gemini-2.5-flash-lite-preview-09-2025
从长远看，我们都已经死了。 — 凯恩斯 43 items · 4 sections
§ I

US Stocks

Pre-market signal radar 12
US pre-market radar
premarket 2026-05-07
12 Bullish
0 Bearish
0 Neutral
Sector Tape
Servers and Thermal Management 2 names
74 Top: DELL · Bullish · RS +8.2% Bullish 2 / Bearish 0 / 5d +16.7%
Manufacturing 4 names
73 Top: FLEX · Bullish · RS +12.6% Bullish 4 / Bearish 0 / 5d +19.6%
Hyperscale Cloud 4 names
72 Top: GOOGL · Bullish · RS +3.2% Bullish 3 / Bearish 0 / 5d +8.6%
Foundry 2 names
72 Top: TSM · Bullish · RS +1.6% Bullish 2 / Bearish 0 / 5d +12.9%
Compute Mining 4 names
70 Top: CIFR · Neutral · RS +4.5% Bullish 2 / Bearish 0 / 5d +38.5%
Battery and Energy Storage 3 names
68 Top: FLNC · Neutral · RS +3.4% Bullish 1 / Bearish 0 / 5d +5.9%
Networking Equipment 4 names
68 Top: CRDO · Neutral · RS -3.2% Bullish 1 / Bearish 0 / 5d +3.7%
Energy Infrastructure 1 names
62 Top: VST · Neutral · RS +4.6% Bullish 0 / Bearish 0 / 5d +2.9%
Ticker Setup Move Score Evidence Quality
DELL Dell Technologies Servers and Thermal Management
Bullish Sector tailwind Medium confidence
-1.0% $236.47 5d +16.1%
74 sector positive RS +7.7%

Bullish setup from -1.0% vs previous close, positive sector tape, 3 recent headline(s).

5-Day Rally Sends Dell Technologies Stock Up 16% - Trefis Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
FN Fabrinet Manufacturing
Bullish Sector tailwind Medium confidence
-0.8% $673.86 5d +5.5%
74 sector positive RS -1.5%

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

Fabrinet $FN Shares Sold by Danske Bank A S - MarketBeat Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
GOOGL Alphabet-A Hyperscale Cloud
Bullish Gap up + news Medium confidence
+1.2% $402.82 5d +13.8%
74 sector positive RS +8.4%

Bullish setup from +1.2% vs previous close, positive sector tape, 3 recent headline(s).

Why Is Alphabet Stock Soaring? - AOL.com Weakens if price fades below previous close or sector benchmarks roll over.
quote: delayed fallback news: fresh financials: fresh news: 3
ORCL Oracle Hyperscale Cloud
Bullish Gap up + news Medium confidence
+2.8% $199.50 5d +18.4%
74 sector positive RS +13.1%

Bullish setup from +2.8% vs previous close, positive sector tape, 3 recent headline(s).

How Much Upside Can ORCL Stock Deliver? - 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% $21.51 5d +29.5%
71 sector positive RS -4.4%

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

Strs Ohio Increases Stake in Cipher Mining Inc. $CIFR - MarketBeat Weakens if price fades below previous close or sector benchmarks roll over.
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:2605.05090v1Lead article

Automatically Finding and Validating Unexpected Side-Effects of Interventions on Language Models

Quintin Pope, Ajay Hayagreeve Balaji, Jacques Thibodeau, Xiaoli Fern

his paper introduces an automated, contrastive evaluation pipeline to audit the behavioral impact of interventions on language models by comparing generations from a base model ($M_1$) and an intervention model ($M_2$). The method generates statistically validated, natural-language hypotheses describing model differences and summarizes recurring themes. This approach reliably surfaces both intended and unexpected side-effects across various real-world interventions like reasoning distillation and knowledge editing.

Read abstract →Full PDFLINNET · SCORE9RELEVANCE
VerTQ Physical Layout in 4-die XCVU29P-3 FPGA. 3x SLR dies shown. Conductor 2.0 optimized the architecture to minimize inter-die signal crossings.
VerTQ Physical Layout in 4-die XCVU29P-3 FPGA. 3x SLR dies shown. Conductor 2.0 optimized the architecture to minimize inter-die signal crossings.
cs.AIarxiv:2605.05170v1

Design Conductor 2.0: An agent builds a TurboQuant inference accelerator in 80 hours

The Verkor Team, Ravi Krishna et al.

The paper introduces **Design Conductor 2.0**, an advanced multi-agent system capable of autonomously designing complex hardware, handling tasks 80 times larger than its predecessor. Its core contribution is demonstrating this capability by designing **VerTQ**…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.04960v1

EP-GRPO: Entropy-Progress Aligned Group Relative Policy Optimization with Implicit Process Guidance

Song Yu, Li Li et al.

This paper introduces EP-GRPO to address credit assignment failures in Group Relative Policy Optimization (GRPO) for LLM reasoning. EP-GRPO integrates entropy-gated modulation to prioritize informative decision points and uses implicit process guidance derived…

abstractpdfLINNET · SCORE9RELEVANCE
Conceptual illustration of the fundamental limitations in standard GRPO. The top panel demonstrates Uniform Granularity , where the model fails to distinguish between critical high entropy decision pivots and deterministic low entropy derivations. The middle panel shows Uniform Polarity , where sequence-level rewards lead to the indiscriminate reinforcement or penalization of both correct and incorrect intermediate steps. The bottom panel illustrates Zero-Variance Collapse , where identical rewards within a group cause the learning signal to vanish.
Conceptual illustration of the fundamental limitations in standard GRPO. The top panel demonstrates Uniform Granularity , where the model fails to distinguish between critical high entropy decision pi…
cs.AIarxiv:2605.05138v1

Executable World Models for ARC-AGI-3 in the Era of Coding Agents

Sergey Rodionov

This paper introduces a coding agent system for ARC-AGI-3 that employs an **executable Python world model** to simulate and plan actions. The core method involves **verifying the model against observations and refactoring it for simplicity** (as an MDL proxy) …

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2605.05003v1

Misaligned by Reward: Socially Undesirable Preferences in LLMs

Gayane Ghazaryan, Esra Dönmez

This paper introduces a framework to evaluate whether Large Language Model (LLM) reward models capture socially desirable preferences by converting social evaluation datasets into pairwise preference data. The core method tests if these reward models prefer so…

abstractpdfLINNET · SCORE9RELEVANCE
№06
cs.AI
9

SoK: Robustness in Large Language Models against Jailbreak Attacks

Feiyue Xu, Hongsheng Hu et al.

This paper systematically surveys jailbreak attacks and defenses against Large Language Models (LLMs) by proposing a taxonomy to structure the field. Its core contribution is the i…

abstractpdf
№07
cs.AI
9

Uno-Orchestra: Parsimonious Agent Routing via Selective Delegation

Zhiqing Cui, Haotong Xie et al.

Uno-Orchestra introduces a unified reinforcement learning (RL) policy that jointly learns when to decompose a task and which specific model/primitive pair should handle each result…

abstractpdf
№08
cs.LG
9

On the Hardness of Junking LLMs

Marco Rando, Samuel Vaiter

This paper investigates the "junking" of LLMs, focusing on the hardness of finding naturally occurring, instruction-free token sequences (natural backdoors) that trigger harmful ou…

abstractpdf
№09
cs.LG
9

Self-Induced Outcome Potential: Turn-Level Credit Assignment for Agents without Verifiers

Senkang Hu, Yong Dai et al.

This paper introduces **Self-Induced Outcome Potential (SIOP)** to provide turn-level credit assignment for long-horizon LLM agents without relying on external verifiers or final a…

abstractpdf
№10
cs.CL
9

Detecting Hallucinations in Large Language Models via Internal Attention Divergence Signals

Gijs van Dijk

This paper introduces a lightweight, single-pass method to detect LLM hallucinations by analyzing internal attention dynamics. The core technique measures the Kullback-Leibler dive…

abstractpdf
See all 20 papers →
§ III

The Town Square

Hacker News 2
466
pts
Top story
Higher usage limits for Claude and a compute deal with SpaceX

Anthropic has increased usage limits for its Claude AI model and secured a significant compute deal with SpaceX.

anthropic.com6 May discuss on HN →
141
Learning the Integral of a Diffusion Model
sander.ai6 May
compiled overnight by google/gemini-2.5-flash-lite-preview-09-2025 · end of issue no. 10 · thank you for reading