汉字News

← 汉字NewsChinese News for Learners
technology
kē jì
科技
·HSK 6

Google
Gǔgē
谷歌
to launch / to announce
tuīchū
推出
TurboQuant (AI compression algorithm)
TurboQuant
TurboQuant
algorithm
suànfǎ
算法
，
to claim / to assert
shēngchēng
声称
can / be able to
kě
可
used to move object before verb
jiāng
将
artificial intelligence
réngōng zhìnéng
人工智能
model
móxíng
模型
memory / RAM
nèicún
内存
requirement / demand
xūqiú
需求
to reduce
jiǎnshǎo
减少
six times
liù bèi
六倍
。

Google
Gǔgē
谷歌
to launch / to release
tuīchū
推出
named / called
míng wéi
名为
TurboQuant (AI compression algorithm)
TurboQuant
TurboQuant
(structural particle)
de
的
compression
yāsuō
压缩
algorithm
suànfǎ
算法
，
to claim
shēngchēng
声称
under (certain conditions)
zài
在
without losing
bù sǔnshī
不损失
accuracy
zhǔnquèxìng
准确性
(structural particle)
de
的
on the premise of
qiántí xià
前提下
，
can
kě
可
used to move object before verb
jiāng
将
large-scale
dàxíng
大型
language model
yǔyán móxíng
语言模型
(structural particle)
de
的
memory / RAM
nèicún
内存
requirement
xūqiú
需求
to reduce
jiǎnshǎo
减少
six times
liù bèi
六倍
。
news / information
xiāoxi
消息
to announce / to release
gōngbù
公布
after / following
hòu
后
，
SK Hynix (South Korean memory chipmaker)
SK Hǎilìshì
SK海力士
and
yǔ
与
Samsung (South Korean technology conglomerate)
Sānxīng
三星
stock price
gǔjià
股价
by a large margin
dàfú
大幅
to drop / to fall
xiàdiē
下跌
。

March 25, 2026 at 6:00 PM

Google
Gǔgē
谷歌
in / during (formal preposition)
yú
于
this week
běn zhōu
本周
to release / to announce
fābù
发布
(completion particle)
le
了
one (measure word for items or projects)
yī xiàng
一项
named / called
míng wéi
名为
TurboQuant (AI compression algorithm)
TurboQuant
TurboQuant
(structural particle)
de
的
new type / novel
xīnxíng
新型
compression
yāsuō
压缩
algorithm
suànfǎ
算法
，
can
kě
可
under / in
zài
在
without losing
bù sǔnshī
不损失
accuracy
zhǔnquèxìng
准确性
(structural particle)
de
的
on the premise of
qiántí xià
前提下
，
used to move object before verb
jiāng
将
large-scale
dàxíng
大型
language model
yǔyán móxíng
语言模型
(structural particle)
de
的
memory / RAM
nèicún
内存
consumption / footprint
zhànyòng
占用
to reduce
jiǎnshǎo
减少
six times
liù bèi
六倍
。
this / the said
gāi
该
algorithm
suànfǎ
算法
(structural particle)
de
的
core / key point
héxīn
核心
lies in / consists in
zàiyú
在于
used to move object before verb
jiāng
将
key-value (computing)
jiàn zhí
键值
cache
huǎncún
缓存
to compress
yāsuō
压缩
to / down to
zhì
至
per value
měi zhí
每值
only / merely
jǐn
仅
3 bits
3 bǐtè
3比特
，
far lower than
yuǎn dī yú
远低于
standard
biāozhǔn
标准
(structural particle)
de
的
16 bits
16 bǐtè
16比特
，
thereby / thus
cóng ér
从而
by a large margin
dàfú
大幅
to reduce / to lower
jiàngdī
降低
(completion particle)
le
了
to run / to operate
yùnxíng
运行
large model (LLM)
dà móxíng
大模型
required / needed
suǒ xū
所需
(structural particle)
de
的
memory
nèicún
内存
bandwidth
dài kuān
带宽
。
Google
Gǔgē
谷歌
researchers
yánjiū rényuán
研究人员
to state / to indicate
biǎoshì
表示
，
on
zài
在
NVIDIA (semiconductor company)
NVIDIA
NVIDIA
H100 (AI accelerator chip)
H100
H100
accelerator
jiāsùqì
加速器
on
shàng
上
，
TurboQuant (AI compression algorithm)
TurboQuant
TurboQuant
(structural particle)
de
的
4-bit
4 bǐtè
4比特
implementation
shíxiàn
实现
method / approach
fāngshì
方式
can
kě
可
used to move object before verb
jiāng
将
attention (neural network mechanism)
zhùyìlì
注意力
computation
jìsuàn
计算
performance
xìngnéng
性能
to boost / to improve
tíshēng
提升
eight times
bā bèi
八倍
。
this time / the present
cǐ cì
此次
breakthrough
tūpò
突破
to combine / comprehensive
zōnghé
综合
to employ / to apply
yùnyòng
运用
(completion particle)
le
了
two items
liǎng xiàng
两项
technology / technique
jìshù
技术
：
quantization
liànghuà
量化
method
fāngfǎ
方法
PolarQuant (quantization sub-method)
PolarQuant
PolarQuant
，
as well as / and
yǐjí
以及
training
xùnliàn
训练
and
yǔ
与
optimization
yōuhuà
优化
framework
kuàngjià
框架
QJL (training and optimization method)
QJL
QJL
。
Google
Gǔgē
谷歌
to plan / plans to
jìhuà
计划
at
zài
在
in the year 2026
2026 nián
2026年
ICLR (International Conference on Learning Representations)
Guójì Xuéxí Biǎoshì Dàhuì
国际学习表示大会
at the conference
shàng
上
formally / officially
zhèngshì
正式
to publish / to present
fābiǎo
发表
related / relevant
xiāngguān
相关
research
yánjiū
研究
results / achievements
chéngguǒ
成果
。
news / information
xiāoxi
消息
to announce / to release
gōngbù
公布
on that day / the same day
dāng rì
当日
，
global
quánqiú
全球
the two largest
liǎng dà
两大
memory chip
cúnchǔ xīnpiàn
存储芯片
manufacturer
zhìzào shāng
制造商
SK Hynix (South Korean memory chipmaker)
SK Hǎilìshì
SK海力士
and
yǔ
与
Samsung (South Korean technology conglomerate)
Sānxīng
三星
(structural particle)
de
的
stock price
gǔjià
股价
respectively
fēnbié
分别
to drop / to fall
xiàdiē
下跌
approximately
yuē
约
6 percent
6%
6%
and
hé
和
5 percent
5%
5%
。
analysts
fēnxī rénshì
分析人士
to point out
zhǐchū
指出
，
if / should
ruò
若
TurboQuant (AI compression algorithm)
TurboQuant
TurboQuant
to obtain / to gain
huòdé
获得
widespread / broad
guǎngfàn
广泛
deployment
bùshǔ
部署
，
with / using
yǐ
以
high-bandwidth memory
gāo dài kuān nèicún
高带宽内存
as / being
wéi
为
core
héxīn
核心
product
chǎnpǐn
产品
(structural particle)
de
的
chip
xīnpiàn
芯片
enterprise / company
qǐyè
企业
will / shall
jiāng
将
to face / to confront
miànlín
面临
sustained
chíxù
持续
(structural particle)
de
的
revenue
yíngshōu
营收
pressure
yālì
压力
。
however / nevertheless
rán ér
然而
，
also (literary)
yì
亦
there are
yǒu
有
viewpoint / perspective
guāndiǎn
观点
to believe / to consider
rènwéi
认为
，
as / along with
suízhe
随着
AI (artificial intelligence)
AI
AI
model
móxíng
模型
parameter
cānshù
参数
scale / scope
guīmó
规模
continuously
chíxù
持续
to expand
kuòzhāng
扩张
，
global
quánqiú
全球
memory
nèicún
内存
demand
xūqiú
需求
long-term
cháng qī
长期
will still
réng jiāng
仍将
to grow / growth
zēngzhǎng
增长
，
market
shìchǎng
市场
(structural particle)
de
的
single-day
dān rì
单日
reaction / response
fǎnyìng
反应
perhaps / possibly
huò
或
amounts to / constitutes
shǔ
属
excessive
guòdù
过度
。

New Words

tǐyù
体育
sports
yùndòng
运动
movement/exercise
jìngsài
竞赛
competition
xuǎnshǒu
选手
athlete
chéngjì
成绩
achievement
jiǎngpái
奖牌
medal
guànjūn
冠军
champion
duìwu
队伍
team
xùnliàn
训练
training
bǐsài
比赛
match

News in English

Science & Technology

Google Unveils TurboQuant Algorithm That Reduces AI Memory Use by Six Times

Google unveiled TurboQuant, an AI compression algorithm claimed to reduce large language model memory requirements by six times without accuracy loss; following the announcement, shares in SK Hynix and Samsung fell sharply.

Sports news reporting on international competitions and athletic achievements.

More in Science & Technology

China
Zhongguó
中国
AI
réngong zhìnéng
人工智能
models
móxíng
模型
captures
zhànlǐng
占领
market
shìchǎng
市场

Chinese AI Models Gain Market Share as GLM 5.2 Rivals OpenAI and Anthropic at Low Cost

UBTech
Yōubìxuǎn
优必选
U1
U1
U1
humanoid robot
rénxíng jīqìrén
人形机器人
launches
fābuō
发布
price
jiàgé
价格

UBTech Launches U1 Ultra-Bionic Humanoid Robot Series Starting at 17600 Dollars

Apple (company)
Píngguǒ
苹果
to file a lawsuit / to sue
qǐsù
起诉
OpenAI (AI company)
OpenAI
OpenAI
，
to accuse / to charge
zhǐkòng
指控
it / the company (formal)
qí
其
to steal / to misappropriate
qièqǔ
窃取
hardware
yìngjiàn
硬件
trade secrets / confidential business information
shāngyè jīmì
商业机密
。

Apple Sues OpenAI Alleging Hardware Trade Secret Theft

Tencent (tech company)
Téngxùn
腾讯
in
zài
在
WeChat (messaging app)
Wēixìn
微信
to launch / roll out
tuīchū
推出
artificial intelligence
réngōng zhìnéng
人工智能
assistant
zhùshǒu
助手
Xiaowei (AI assistant name)
Xiǎowēi
小微
。

Tencent Begins Rolling Out AI Agent Xiaowei to WeChat Users