Daily Issue
Vol. I — No. 27
10 · 06
Wednesday, 10 June 2026
Generated 2026-06-10 12:47
google/gemini-2.5-flash-lite-preview-09-2025
见到你那一刻我心里有场海啸，可我静静站着，没有让任何人知道。 — 佚名 51 items · 4 sections
§ 0

The Morning

Local weather 1
This morning in
London
Overcast
Today's range
15.6°↓10.2°
currently 15.1°
Feels
12.3°
Rain
100%
Wind
13 km/h
Humid
55%
Rise
04:44
Set
21:16
§ I

US Stocks

Pre-market signal radar 12
US pre-market radar
premarket 2026-06-10
0 Bullish
2 Bearish
10 Neutral
Sector Tape
Battery and Energy Storage 3 names
37 Top: EOSE · Neutral · RS -17.6% Bullish 0 / Bearish 1 / 5d -24.1%
Foundry 2 names
38 Top: INTC · Neutral · RS +4.7% Bullish 0 / Bearish 0 / 5d -2.1%
Hyperscale Cloud 4 names
39 Top: AMZN · Neutral · RS +1.3% Bullish 0 / Bearish 0 / 5d -7.1%
Servers and Thermal Management 2 names
40 Top: VRT · Neutral · RS -5.2% Bullish 0 / Bearish 0 / 5d -12.9%
Networking Equipment 4 names
44 Top: CIEN · Neutral · RS -0.5% Bullish 0 / Bearish 1 / 5d -9.3%
Compute Mining 4 names
44 Top: IREN · Neutral · RS -1.1% Bullish 0 / Bearish 0 / 5d -12.8%
Manufacturing 4 names
44 Top: CLS · Neutral · RS -11.7% Bullish 0 / Bearish 0 / 5d -14.6%
Energy Infrastructure 1 names
46 Top: VST · Neutral · RS -7.0% Bullish 0 / Bearish 0 / 5d -7.4%
Ticker Setup Move Score Evidence Quality
EOSE Eos Energy Battery and Energy Storage
Bearish Risk watch Medium confidence
-6.4% $6.26 5d -33.5%
27 sector negative RS -27.1%

Bearish/risk setup from -6.4% vs previous close, negative sector tape, 3 recent headline(s).

Eos Energy (NASDAQ: EOSE) CFO discloses initial 12,114-share stake - Stock Titan Improves if price reclaims previous close and negative headlines are not confirmed.
quote: delayed fallback news: fresh financials: fresh news: 3
CIEN Ciena Networking Equipment
Bearish Risk watch Medium confidence
-0.5% $437.17 5d -29.9%
30 sector negative RS -21.1%

Bearish/risk setup from negative sector tape, 3 recent headline(s).

Ciena Stock 5-Day Losing Spree: Stock Falls -30% - Trefis Improves if price reclaims previous close and negative headlines are not confirmed.
quote: delayed fallback news: fresh financials: fresh news: 3
ANET Arista Networks Networking Equipment
Neutral News watch Low confidence
-2.1% $149.00 5d -13.2%
38 sector negative RS -4.4%

Watchlist item from -2.1% vs previous close, negative sector tape, 3 recent headline(s).

The Warning Sign Inside Arista Networks Stock's Great News - Yahoo Finance Needs fresh price/news confirmation before becoming an actionable setup.
quote: delayed fallback news: fresh financials: fresh news: 3
INTC Intel Foundry
Neutral News watch Low confidence
-1.2% $106.63 5d -0.0%
38 sector negative RS +6.8%

Watchlist item from -1.2% vs previous close, negative sector tape, 3 recent headline(s).

Why Is Intel Stock Sliding On Wednesday? - Benzinga 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:2606.11182v1Lead article

EEVEE: Towards Test-time Prompt Learning in the Real World for Self-Improving Agents

Weixian Xu, Shilong Liu, Mengdi Wang

EVEE introduces a novel test-time prompt learning framework designed for real-world, heterogeneous task streams, overcoming limitations of single-dataset methods. Its core method involves a router that clusters incoming inputs and assigns them to appropriate prompt configurations, optimized through a router-prompt co-evolution strategy. This approach significantly improves the robustness of LLM agents when handling diverse, interleaved data while preserving performance on individual tasks.

Read abstract →Full PDFLINNET · SCORE9RELEVANCE
Incremental multi-benchmark retention improvement as tasks are added in the order GPQA Diamond, Formula, TheoremQA, and HumanEval. Each bar stacks per-benchmark improvements for all tasks seen so far: solid upward blocks are positive gains, and hatched downward blocks are negative retention losses. The number above or below each bar is its final summed improvement after all blocks are added.
Incremental multi-benchmark retention improvement as tasks are added in the order GPQA Diamond, Formula, TheoremQA, and HumanEval. Each bar stacks per-benchmark improvements for all tasks seen so far: solid upward blocks are positive gains, and hatched downward blocks are negativ…
End-to-end illustration of a Word task in OfficeEval . The original document ( left ) is transformed according to the task instructions ( center ) into a styled brochure with header image, heading styles, and mail-merge labels ( right ). Only page 1 of the 2-page document is shown; several steps (e.g., 3-column layout, watermark) apply to page 2. The task is scored by 30 deterministic criteria across 6 skill categories. Instructions are translated from the original Chinese; additional examples across Word, Excel, and PowerPoint appear in the Appendix.
End-to-end illustration of a Word task in OfficeEval . The original document ( left ) is transformed according to the task instructions ( center ) into a styled brochure with header image, heading sty…
cs.AIarxiv:2606.10956v1

Mind the Gap: Can Frontier LLMs Pass a Standardized Office Proficiency Exam?

Tengchao Lv, Dongdong Zhang et al.

This paper introduces a rigorous benchmark, based on China's National Computer Rank Examination (NCRE), to evaluate frontier Large Language Models' (LLMs) ability to perform complex, multi-application Office automation tasks requiring long-horizon planning. Th…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2606.10989v1

Null-Space Constrained Low-Rank Adaptation for Response-Specified Large Language Model Unlearning

Bocheng Ju, Jianhua Wang et al.

This paper introduces Null-Space Constrained Response-Specified Unlearning (NSRU), a low-rank adaptation method for LLM unlearning. NSRU constrains the update parameters to the null space of estimated "retain subspaces" derived from benign data, ensuring adapt…

abstractpdfLINNET · SCORE9RELEVANCE
Motivation and core intuition of NSRU. (a) Suppression-only unlearning penalizes the undesired response y − y^{-} but leaves the safe replacement behavior unspecified and can induce under-constrained updates that perturb retained behavior. (b) NSRU specifies a safe target response y + y^{+} , explicitly suppresses y − y^{-} , and uses projected LoRA updates that act through retain-orthogonal components, redirecting forget queries while reducing retain-side interference.
Motivation and core intuition of NSRU. (a) Suppression-only unlearning penalizes the undesired response y − y^{-} but leaves the safe replacement behavior unspecified and can induce under-constrained …
An overview of the proposed ReasonAlloc framework. Left (I): Layer-wise allocation strategy based on offline architecture calibration, demonstrating the non-linear “Reasoning Wave” KV demand across layers. Right (II): Head-wise allocation strategy that dynamically routes KV budgets to distinct attention heads based on real-time importance and redundancy scoring during decoding.
An overview of the proposed ReasonAlloc framework. Left (I): Layer-wise allocation strategy based on offline architecture calibration, demonstrating the non-linear “Reasoning Wave” KV demand across la…
cs.AIarxiv:2606.11164v1

ReasonAlloc: Hierarchical Decoding-Time KV Cache Budget Allocation for Reasoning Models

Wenhao Liu, Hao Shi et al.

ReasonAlloc addresses KV cache bottlenecks in LLM reasoning by introducing a hierarchical, training-free budget allocation framework. It combines an offline layer-wise preallocation strategy, capturing the "Reasoning Wave" demand pattern, with an online head-w…

abstractpdfLINNET · SCORE9RELEVANCE
cs.AIarxiv:2606.10917v1

Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution

Xucong Wang, Ziyu Ma et al.

The Role-Agent framework bootstraps LLM agent learning by having a single LLM concurrently act as both the agent and the environment. It uses a dual-component system: World-In-Agent (WIA) generates a process reward based on state prediction accuracy, while Age…

abstractpdfLINNET · SCORE9RELEVANCE
(a): Static environments provide sparse and non-specific feedback that limits the agent’s exploration; (b): Synthetic environments incur high labor and runtime costs; (c): The proposed Role-Agent enables one model to switch roles between agent and environment to achieve bootstrapped co-evolution.
(a): Static environments provide sparse and non-specific feedback that limits the agent’s exploration; (b): Synthetic environments incur high labor and runtime costs; (c): The proposed Role-Agent enab…
№06
cs.AI
9

T1-Bench: Benchmarking Multi-Scenario Agents in Real-World Domains

Genta Indra Winata, Amartya Chakraborty et al.

T1-Bench is introduced as a high-fidelity benchmark designed to evaluate LLM-based agents in complex, realistic, multi-domain customer-facing scenarios. Its core contribution is pr…

abstractpdf
№07
cs.AI
9

What Fits (Into Few Tokens) Doesn't Overfit: Compression and Generalization in ML Research Agents

Martin Andres Bertran, Aaron Roth et al.

This paper investigates the hypothesis that successful machine learning strategies are highly compressible, even when adaptively reused on held-out benchmarks. The authors test thi…

abstractpdf
№08
cs.AI
9

Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic tasks in Real-World Professional Fields

Liya Zhu, Jingzhe Ding et al.

Workflow-GYM is introduced as a novel benchmark to address the lack of evaluation for AI agents performing long-horizon, high-value professional workflows using graphical user inte…

abstractpdf
№09
cs.LG
9

Flow-DPPO: Divergence Proximal Policy Optimization for Flow Matching Models

Bowen Ping, Xiangxin Zhou et al.

Flow-DPPO addresses limitations in applying standard PPO to flow matching models by replacing noisy ratio clipping with a direct divergence constraint. Leveraging the Gaussian natu…

abstractpdf
№10
cs.CL
9

Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models

Prajakta Kini, Avinash Reddy et al.

This paper investigates whether converting instruction-tuned Large Language Models (LLMs) into reasoning models via post-training preserves their original alignment behaviors (safe…

abstractpdf
See all 20 papers →
§ III

The Town Square

Hacker News 10
734
pts
Top story
CEOs who think AI replaces their employees are just bad CEOs

The core argument is that CEOs viewing AI solely as a tool for employee replacement, rather than augmentation, demonstrate poor leadership.

techdirt.com9 Jun discuss on HN →
694
German ruling declares Google liable for false answers in AI Overviews
the-decoder.com10 Jun
404
Apple decided not to roll out Siri in EU after denied request for exemption
reuters.com9 Jun
245
Ultrafast machine learning on FPGAs via Kolmogorov-Arnold Networks
aarushgupta.io9 Jun
219
Surprise, Pay $1000
forestwalk.ai9 Jun
195
'Sloppenheimer:' Amazon employees mock the company's AI on Slack
404media.co9 Jun
169
AWS Bedrock to require sharing data with Anthropic for Mythos and future models
10 Jun
156
Where is the AI jobs crisis?
apollo.com9 Jun
156
Ask HN: Are you still using a Vision Pro?
9 Jun
152
Rich Sutton on AI creativity and discovery
twitter.com10 Jun
compiled overnight by google/gemini-2.5-flash-lite-preview-09-2025 · end of issue no. 27 · thank you for reading