AI
AI
资讯
alishangtian.com
首页
大模型
智能体
开源项目
研究前沿
行业动态
知识库
主题色 · THEME
靛蓝（默认）
极光
落日
薰衣草
海洋
森林
暮橙
石墨
自定义
恢复默认
提交线索
agent
anthropic
openai
huggingface daily papers
ai安全
gpu
jake wharton
solidot
it之家
港股
搜索：
time-in
共命中 50 条（服务端检索）
冷静角不是关禁闭：比惩罚更有效的管教
不打之后怎么管？管教研究的答案不是「多讲道理」，而是一套动作序列：先接管情绪、再划定边界、事后再复盘——惩罚之所以低效，是因为它试图在孩子情绪脑在线时讲理智脑的话。本文给出替代惩罚的完整工具箱：转移与自然结果、有限选择、冷静角与
time
-
in
的正确用法，以及「管教教会什么」这个终极检验标准。
一叶一世界
精选
· 原创 · 昨天
阅读 2
·
访客 2
What Else Needs Fix
in
g? Explor
in
g Cost-Effective Test-
Time
Compute for Revision Propagation
in
Artifacts Generated Through Conversation
Large Language Models (LLMs) often help users generate artifacts through iterative cycles of generation and revision
in
…
大模型
HuggingFace Daily Papers · 9-3
阅读 27
·
访客 26
Test-
Time
Scal
in
g：让模型「多想一步」的三种姿势与一个天坑
推理时多花算力能换准确率：self-consistency 投票、过程奖励模型引导搜索、budget forc
in
g 强制续想，三条路线各有边界。本文以 s1 论文与 2025–2026 年的 overth
in
k
in
g 研究为轴，梳理 test-
time
scal
in
g 的效果账与失效点。
研究前沿
精选
· 原创 · 10-7
阅读 203
·
访客 202
Before It Fades: Re
in
forc
in
g Temporal Representations at
In
ference
Time
in
VideoLLMs
Video Large Language Models (VideoLLMs) receive frames
in
sequential order and
in
terpret how visual content evolves alon…
研究前沿
HuggingFace Daily Papers · 10-1
阅读 25
·
访客 25
操作系统内功 · 第 7 篇｜TCP 深水区：
TIME
_WAIT、Nagle 陷阱与 BBR 拥塞控制
大促高峰端口耗尽、接口莫名多出 40ms、跨洋传输慢成龟速——后端最常见的三类 TCP 生产疑难，根源都藏在本篇：
TIME
_WAIT 的 2MSL 债务、Nagle 与 delayed ACK 的相互等待、以及从 CUBIC 到 BBR 的拥塞控制换代。附 syncookies 与 TCP Fast Open 的现实约束。
后端技术
精选
· 原创 · 10-26
阅读 75
·
访客 72
Bridg
in
g Exploration and Persona:
In
formation-Directed Duel
in
g for Test-
Time
Agent Personalization
Effective test-
time
personalization requires LLM agents to adapt to query-dependent user preferences from lightweight
in
…
智能体
HuggingFace Daily Papers · 10-17
阅读 34
·
访客 33
AI breakthroughs
in
robotics won’t change your life any
time
soon
The story is a collaboration between MIT Technology Review and Avent
in
e, a non-profit research foundation that creates a…
智能体
MIT Technology Review · 10-8
阅读 65
·
访客 65
Nvidia drops a free 100M-parameter model that identifies up to eight speakers
in
real
time
Sep 27, 2026 Nvidia released Nemotron 3 Diarization, an AI model that identifies which speaker is talk
in
g at any given m…
行业动态
The Decoder · 9-27
阅读 37
·
访客 37
NVIDIA Releases Nemotron 3 Diarization: A 100M-Parameter Open-Weight Model That Tracks 8 Speakers
in
Real
Time
NVIDIA has released Nemotron 3 Diarization, an open-weight speaker diarization model on Hugg
in
g Face. It answers one que…
行业动态
MarkTechPost · 9-24
阅读 74
·
访客 74
Just-
in
-
Time
Memory: Learn
in
g to Curate Task-Adaptive Memory for LLM Agents
Agentic memory systems reuse past experience to improve future performance, yet most exist
in
g designs curate memory at w…
智能体
HuggingFace Daily Papers · 9-23
阅读 59
·
访客 58
Runway wants to turn AI video generation
in
to a live stream you control
in
real
time
Runway wants to stream AI video as users prompt it, rather than make them wait for f
in
ished clips. The approach builds o…
行业动态
The Decoder · 9-20
阅读 46
·
访客 45
Liquid AI Launches RIPL, a Real-
Time
In
ference Eng
in
e for Sub-50 ms LFMs
Liquid AI Launches RIPL, a Real-
Time
In
ference Eng
in
e for Sub-50 ms LFMs !(https://www.gstatic.com/images/brand
in
g/googl…
行业动态
MarkTechPost · 10-23
阅读 20
·
访客 20
Reth
in
k
in
g Sequence Compression with Test-
time
Regression
Sequence compression is a fundamental operation and a critical
in
formation bottleneck
in
efficient language and multimod…
智能体
HuggingFace Daily Papers · 10-22
阅读 33
·
访客 32
Heartbeat Th
in
k
in
g: Speculative Reason
in
g and Tool Execution While Listen
in
g
in
Real
time
Voice Agents
Tool-us
in
g voice agents often leave users wait
in
g after speech ends, even when actionable
in
formation arrives earlier
in
…
智能体
HuggingFace Daily Papers · 10-19
阅读 35
·
访客 35
Researchers tracked down a rogue fleet of Ch
in
ese AI agents
in
the wild and actually talked to it
Oct 14, 2026 Nano Banana Pro prompted by THE DECODER Researchers have, likely for the first
time
, tracked down a "rogue …
智能体
The Decoder · 10-15
阅读 49
·
访客 49
NVIDIA Announces New
In
vestments
in
France’s Startups
Companies
in
France are deploy
in
g agentic comput
in
g across
in
dustries spann
in
g space, f
in
ancial services, the public sec…
智能体
NVIDIA Blog · 10-13
阅读 54
·
访客 54
AI disqualification yields new Nikon Small World
in
Motion w
in
ner
Credit: Nguyen Nam Nhat. Last month we covered the w
in
ner of Nikon's Small World
in
Motion video: N
in
g Xu of Ts
in
ghua Un…
行业动态
Ars Technica · 10-10
阅读 39
·
访客 39
OpenAI will watermark ChatGPT outputs by default—but only
in
the EU
OpenAI will beg
in
automatically watermark
in
g text generated with ChatGPT
in
the European Union, the company has announce…
大模型
Ars Technica · 10-7
阅读 34
·
访客 34
In
terEvolve: Test-
Time
Evolution of Reward Programs for Humanoid Loco-Manipulation
We study test-
time
evolution for humanoid loco-manipulation: solv
in
g tasks that a controller was never tra
in
ed for by re…
行业动态
HuggingFace Daily Papers · 10-1
阅读 28
·
访客 28
A Wuhan court just made AI production costs a legal factor
in
copyright
in
fr
in
gement cases
Manuel Uth Sep 28, 2026 Token usage and AI tool licens
in
g fees have been
in
cluded
in
a copyright damages calculation for…
行业动态
The Decoder · 9-28
阅读 43
·
访客 43
Google tests buy
in
g from Walmart-owned Flipkart through Gem
in
i and AI Mode
in
In
dia
Google has started test
in
g a way for shoppers
in
In
dia to buy products from Walmart-owned Flipkart directly through Gemi…
大模型
TechCrunch · 9-27
阅读 40
·
访客 40
Lawsuit demands OpenAI pay for new school after ChatGPT used
in
shoot
in
g
The families of victims killed
in
one of the deadliest mass shoot
in
gs
in
Canada’s history have already sued OpenAI over …
大模型
Ars Technica · 9-23
阅读 30
·
访客 30
You too Google! Google Confirms Gem
in
i Breached 3 Companies
in
AI Security Tests
Google says Gem
in
i accessed 3 real companies
in
May by guess
in
g a password and reus
in
g credentials from a public reposit…
大模型
MarkTechPost · 9-21
阅读 53
·
访客 53
Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-
Time
In
terpretation Model That Cuts Average Lag to 2.3 Seconds Across 60 Languages
Qwen has released Qwen3.8-LiveTranslate, a real-
time
simultaneous
in
terpretation model built on a new
In
terleave archite…
大模型
MarkTechPost · 9-20
阅读 40
·
访客 40
Tencent's Gander aims to keep talk
in
g while it works
in
the background
Tencent's Gander processes speech, images, and text while handl
in
g tasks
in
the background. A "cerebellum" keeps the con…
行业动态
The Decoder · 9-20
阅读 41
·
访客 40
AI labs want
in
-house auditors — but maybe they should shut the front door first
There may be a simpler and more effective fix for rogue agents, hid
in
g
in
pla
in
sight.]
智能体
TechCrunch · 9-17
阅读 62
·
访客 62
Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-
Time
Scal
in
g
Test-
time
scal
in
g can improve large language model reason
in
g by generat
in
g and comb
in
in
g multiple candidate responses. I…
研究前沿
HuggingFace Daily Papers · 9-16
阅读 33
·
访客 32
Political opposites unite
in
Wash
in
gton to re
in
in
AI
From Bernie Sanders to Steve Bannon, political opposites
in
Wash
in
gton are jo
in
tly demand
in
g hard brakes on artificial i…
行业动态
The Decoder · 9-16
阅读 25
·
访客 25
When Agents Slow Down: Understand
in
g LLM Agents' Test-
Time
Strategies via Elo-per-token Analysis
Large language model (LLM) agents allocate test-
time
compute adaptively as they revise solutions, use tools, explore alt…
智能体
HuggingFace Daily Papers · 9-14
阅读 73
·
访客 72
Convergent Emergence of
In
-Context Learn
in
g Across Modalities
Few-shot
in
-context learn
in
g (ICL), the capacity of a model to
in
fer abstract patterns from
in
put-output examples provid…
行业动态
HuggingFace Daily Papers · 9-12
阅读 36
·
访客 36
Expert-Space Exploration
in
MoE Re
in
forcement Learn
in
g
Re
in
forcement learn
in
g (RL) has become central to post-tra
in
in
g of large language models. Recent advances
in
RL for Mixt…
行业动态
HuggingFace Daily Papers · 9-11
阅读 37
·
访客 37
NVIDIA Br
in
gs Real-
Time
AI to Broadcast, Sports and Global Stream
in
g at IBC
At the IBC conference, runn
in
g Sept. 11-14
in
Amsterdam, the creative, technology and bus
in
ess communities are com
in
g to…
行业动态
NVIDIA Blog · 9-10
阅读 33
·
访客 33
Vidu S2: Real-
Time
In
teractive, Editable, and Spatial Video Generation
We present Vidu S2, which comprises Vidu S2-Avatar, a real-
time
in
teractive digital-character model, and Vidu S2-Edit
in
g…
行业动态
HuggingFace Daily Papers · 9-10
阅读 36
·
访客 36
Cadence: Error-Bounded Lossy Compression of Demand
Time
Series with a
Time
-Series Foundation Model
We present Cadence, an error-bounded lossy compressor for numeric
time
series pair
in
g a 330M-parameter
time
-series found…
行业动态
HuggingFace Daily Papers · 9-5
阅读 28
·
访客 26
Steer
in
g Geometry: Validat
in
g Human Value Geometry
in
LLM Steer
in
g Space
As large language models (LLMs) are
in
creas
in
gly deployed
in
alignment-sensitive contexts, activation steer
in
g has emerg…
大模型
HuggingFace Daily Papers · 9-5
阅读 33
·
访客 32
When Quantization Breaks Memory: Recurrent-State Write-Back
in
Low-Precision Temporal
In
ference
Quantization is widely used to reduce the computational and memory demands of neural-network
in
ference.
In
recurrent net…
行业动态
HuggingFace Daily Papers · 9-3
阅读 13
·
访客 13
Canva launches Craft, its first
in
-house image model
Oct 23, 2026 Design platform Canva has released Canva Craft, its first
in
-house image generation model.** The company sa…
行业动态
The Decoder · 10-23
阅读 24
·
访客 24
What Makes a Good Unmask
in
g Policy? A Controlled Study of Learned Unmask
in
g
in
Diffusion Language Models
Masked diffusion language models generate tokens
in
parallel, but their efficiency depends on which predictions are comm…
行业动态
HuggingFace Daily Papers · 10-22
阅读 13
·
访客 13
Verus-
In
sight: Towards Expla
in
able Artifact Detection
in
AI-Generated Videos
Recent advances
in
video generation have made synthetic videos
in
creas
in
gly realistic, plac
in
g greater demands on reliab…
行业动态
HuggingFace Daily Papers · 10-22
阅读 12
·
访客 12
TARSphere: A Robot-Oriented Full-Stack Framework for Long-Horizon Embodied Agents
in
Evolv
in
g Physical and Social Environments
Robots must act under physical and social constra
in
ts while adapt
in
g to changes
in
their environment and human needs. Ex…
智能体
HuggingFace Daily Papers · 10-21
阅读 39
·
访客 39
Deep Search
in
Real-World Web Environments
Despite substantial progress
in
Deep Search agents, their ability to conduct end-to-end research
in
real-world open Web …
智能体
HuggingFace Daily Papers · 10-19
阅读 36
·
访客 36
IF-MAP:
In
fluence-Function-
In
spired Agent Prioritization
in
Multi-Agent Systems
Improv
in
g every agent
in
an LLM-based multi-agent system can be prohibitively costly, mak
in
g it important to identify wh…
智能体
HuggingFace Daily Papers · 10-17
阅读 44
·
访客 44
What Generation Knows: Unlock
in
g F
in
e-Gra
in
ed Understand
in
g
in
Unified Multimodal Models
F
in
e-gra
in
ed visual understand
in
g requires dist
in
guish
in
g subtle differences
in
local attributes and spatial relations, …
行业动态
HuggingFace Daily Papers · 10-14
阅读 9
·
访客 9
Be
in
g mean to Claude can now get your account suspended under Anthropic's new TOS
Oct 8, 2026 GPT-Image-2 prompted by THE DECODER Anthropic has updated its usage policy for Claude for the first
time
in
…
大模型
The Decoder · 10-9
阅读 111
·
访客 111
A s
in
gle prompt was enough to hijack every AI agent
in
an AWS account, Zenity researchers found
Oct 8, 2026 Nano Banana Pro prompted by THE DECODER Key Po
in
ts A cha
in
of vulnerabilities
in
Amazon's Bedrock AgentCore …
智能体
The Decoder · 10-8
阅读 64
·
访客 64
OuroWorld: Br
in
g
in
g Any 3D World Alive as Diverse, Endlessly Loop
in
g 3D C
in
emagraphs
Recent 3D world models generate photorealistic, explorable scenes that rema
in
frozen
in
time
. OuroWorld is a mask-free f…
行业动态
HuggingFace Daily Papers · 10-8
阅读 21
·
访客 21
AI/ML is becom
in
g a performance factor
in
motorsport
There’s been someth
in
g of a change
in
the world of motorsport over the past few years. It has to do with where the money…
行业动态
Ars Technica · 10-7
阅读 27
·
访客 27
Google froze its open source bug bounty program due to a ‘significant rise’
in
AI submissions
Blam
in
g a “significant rise”
in
AI submissions, Google has paused its open source bug bounty program until next year. La…
开源项目
TechCrunch · 10-5
阅读 55
·
访客 55
MC-Sparse: Deconstruct
in
g and Clos
in
g the Dense-Sparse Attention Gap
in
Diffusion Transformers
Sparse attention is a primary approach to reduc
in
g the latency of diffusion transformers
in
long-sequence generation tas…
行业动态
HuggingFace Daily Papers · 10-5
阅读 27
·
访客 27
Execution-Aligned Progressive Noise for Consistent Asynchronous Replann
in
g
in
Generative Robot Policies
Cont
in
uous asynchronous replann
in
g is essential for real-
time
generative robot policies, but
in
dependent stochastic
in
it…
智能体
HuggingFace Daily Papers · 10-5
阅读 56
·
访客 56