tsinghua's achievements have been implemented, leading the domestic ai large model!
2024-09-23
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
among the domestic enterprises that independently develop large models
knowledge engineering laboratory, department of computer science, tsinghua university
the transformed “smart spectrum”
it is undoubtedly a bright spot
its research and developmentlarge model glm-4
many performances are comparable to the large model gpt-4
the number of global downloads of the series of open source models exceeds 20 million
is the current download volume and open source influence
one of the tallest domestically produced large models
from "experiment" to "market"
“zhipu” is launched to the public
generative video function "clear shadow"
software "qingyan" with video calling function
transforming cutting-edge science and technology into
ai assistant for everyone
ceo of zhipu
tsinghua university computer science department
1998 undergraduate, 2002 master's alumni
zhang peng
let's get closer
"tsinghua people serving the country" series of reports
share the experience of tsinghua people in all walks of life
a sincere desire to serve the country and the people!
climbing over the technical wall
“china needs to have its own pre-training model framework”
1998
zhang peng from hubei
admitted to the department of computer science, tsinghua university
through in-depth participation in laboratory projects
he soon found his research direction.
"text data mining and semantic analysis"
zhang peng climbing a mountain
during school
he has published more than 10 papers in top conferences
designedthe first in china
a cross-language knowledge graph system that balances chinese and english
after graduation, zhang peng chose to join
knowledge engineering laboratory, department of computer science, tsinghua university
continue to explore cutting-edge technology
since 2006
the laboratory began to study big data mining and knowledge graphs
hatched
science and technology intelligence big data mining and service system platform
—— aminer system
as the system matures
the team began to consider commercialization
aminer was separated in june 2019
established a new company "zhipu"
in 2020, open ai released
gpt-3 model with 175 billion parameters
attracting world attention
under the impact of foreign artificial intelligence products
“china needs to have its own pre-training model framework”
become a consensus
zhang peng attended the china computer conference
GPT、BERT、T5
these are all basic technologies proposed by western scientists.
following the existing path is undoubtedly the most labor-saving
however
“core technology cannot be begged for or bought.”
as a tsinghua student
we must do something for the country, the nation and the society
want to master advanced technologies in cutting-edge fields
some walls must be crossed
zhang peng and his team members took to
difficult but reliable road of independent research and development
in the big language model (llm) wave
the team was pleasantly surprised to find
self-developed pre-training framework - glm
human input intention and some common sense
or the ability to comprehend knowledge and information
an unexpected performance
breaking the foreign monopoly
there is hope for achieving "overtaking on the curve"
the team immediately decided
dedicated to the development of glm pre-training framework
high training cost, large manpower investment, unstable process
facing industry-recognized challenges
zhang peng believes
“although the situation is complex and difficult
but we wanted to do something of our own.”
by continuously optimizing and quantifying the algorithm
while ensuring accuracy and inference speed
bilingual pre-trained model with 130 billion parameters
GLM-130B
ultimately, operating costs are reduced by 75%
at the same time
adhere to independent research
it also brings the advantage of adapting to domestic hardware
eliminate the possibility of information leakage from the source
a "safety lock" for domestic information security
november 2022
stanford university center for large modeling
comprehensive evaluation of 30 major models worldwide
glm-130b developed by the team
as the only large model selected from asia
in terms of robustness, calibration error and unbiasedness
achieved outstanding results
“on an equal or smaller scale
we can reach the world's top level
even ahead in some indicators.”
zhang peng introduced
ai big models fly into ordinary people's homes
“our technology is going to solve some practical problems”
department of computer science, tsinghua university
under the influence of the concept of "penetrating the sky, connecting the earth, connecting the heart, and cultivating people"
zhang peng on "technology serves people"
have a deep understanding
beijing 2022 winter olympics attracts world attention
ice sports have entered thousands of households
various professional terms
the news anchor speaks at a speed of more than 200 words per minute
always makes sign language teachers "busy"
how to better satisfy
do hearing-impaired friends understand the requirements of the competition?
now
it is the “big model” that is used in social services
the critical moment
the digital human with sign language developed by the team
understanding audio semantics through “big models”
using semantic distillation model and sign language translation model
team converts speech into sign language
combined with muscle binding technology to achieve facial acquisition
at last
digital people appearing on the beijing tv stage
interpretation in fluent and natural sign language
for thousands of hearing-impaired friends
bring efficient and accurate real-time explanation
beijing 2022 winter olympics successfully concluded
the "translator" did not "go off work"
solid technology brings long-term effects
the digital human with sign language developed by the team
still active on tv and online platforms
scenic spots and public service hall
continue to provide barrier-free services for hearing-impaired friends
onward! towards general artificial intelligence
“let machines think like humans”
the world of technology is full of competition
advanced technology iterative upgrade
august 31, 2023
products developed based on chatglm
"zhipu qingyan" is online
provide users with real-time
general q&a, virtual conversation, code generation
creative writing and other services
cutting-edge science and technology
become an ai assistant available to everyone
january 16, 2024
the team released the new generation of large base model glm-4
compared with the previous generation
glm-4 has more impressive data
overall performance improved by 60%
reaching the level of gpt-4
the upgraded large model
able to understand and plan complex instructions based on user intent
autonomously call the web browser
code interpreter
wenshengtu cogview3 model, etc.
tools for complex tasks
up to 300 pages of text can be processed in one go
video animation generated by "qingying"
july 2024
"zhipu" launches generative video function "qingying"
just a paragraph or a picture
it can be generated automatically in 30 seconds
a high-quality video with a resolution of 1440×960
“zhipu qingyan” is open to all users
application of video call function in office scenarios
august 30, 2024
"zhipu qingyan" has been significantly upgraded again
the first domestic
video call function open to clients
users only need to open the video call interface
you can enjoy
it’s like a “face-to-face” communication experience with ai
the zhipu big model has realized
from language, text, pictures
then to the modal leap of video
model matrix for international advanced level of artificial intelligence
taking shape
“let machines think like humans”
explore the infinite possibilities of the future through technological advancement
artificial intelligence is liberating productivity.
it also leaves more room for technological innovation
from transforming the research results of tsinghua laboratory into
10 billion and 100 billion models released
then to the aicg model in the subdivision field
and product matrix construction
down to earth
committed to product research and development upgrades
zhang peng and his team take every step steadily and steadily
zhang peng
“in the future, big models will be able to help everyone do a lot of things”
in the dream of zhang peng and his team
after the big model
the ultimate goal of achieving general artificial intelligence
the exploration beyond the walls continues
where will the new path lead?
when will the vision of general artificial intelligence be realized?
tsinghua university researcher zhang peng and the zhipu team
keep moving
find the answer by exploring step by step
