DEEPSEEK, AND FUTURE OF AI
SARASIJ MAJUMDER
DeepSeek came as a big tremor on the AI WORLD. Through
ingenious engineering, the Chinese startup has QUITE SIGNIFICANTLY reduced its
computing power needs, raising concerns about the future demand for Nvidia’s
high-end chips.
DeepSeek-R1 became rapidly popular with hundreds of
derivative models developed over just in a few days. This has negatively
impacted the market value of U.S.-based competitors with Nvidia registering a
record loss 27/01/25 -- Black Monday.
Tech investor Marc Andreessen has described this as “AI’s
Sputnik moment.” This is mainly due to two underlying reasons—the
cost-effectiveness of DeepSeek’s AI models and their ability to run efficiently
on less expensive hardware.
“DeepSeek has had some real innovations,” Nadella said during an investor call after Microsoft reported quarterly results on this Wednesday.
Breakthrough:
- DeepSeek’s
engineers, in their research paper, have revealed that they used
approximately 2,000 Nvidia H800 chips—less advanced than most AI chips—to
train the model.
- It
rolled out a preview of its reasoning model, R1, and v3, an advanced LLM
with a 700GB size and 685 billion parameters—outpacing any model
previously available for free download. DeepSeek’s v3 has 685 billion
parameters, meaning it can handle more complex tasks compared to Meta’s
Llama 3.
- “This
model’s success challenges the dominance of Big Tech … For India and other
emerging markets, this represents a unique opportunity to build
context-specific AI solutions, leveraging this democratised frontier to
solve local problems,” says Jaspreet Bindra, Co-founder and CEO, AI & Beyond.
Discussion:
“While DeepSeek is not exactly a breakthrough scientific
innovation, it is impressive from an engineering perspective,” said Xia Ben Hu, an
associate professor of computer science at Rice who researches machine learning
algorithms and systems relevant to applications in social informatics, health
informatics and information security. “Despite higher efficiencies, the demand
for compute won’t likely decrease.”
Hu has developed an open-source package called AutoKeras
that is among the most used automated deep learning systems on GitHub. Work on
deep collaborative filtering, anomaly detection and knowledge graphs from
Hu’s research group at
Rice has been included in the TensorFlow package, Apple production system and
Bing production system.
Anshumali
Shrivastava, an associate professor of computer science, electrical and
computer engineering and statistics at Rice, and a member of the Ken Kennedy
Institute, sees the arrival of DeepSeek as “entirely
expected.”
“More efficient AI alternatives are possible, and DeepSeek
has finally allowed that to sink in in a real sense,” said Shrivastava, whose work focuses on
large-scale machine learning, scalable deep learning, algorithms for big data
and graph mining and leveraging AI for cybersecurity. “AI optimization is
poised to become the next big focus.”
Shrivastava is the founder of Third AI, which develops
custom, private and cost-effective AI solutions, and he has been an Amazon
Visiting Academic, machine learning consultant at Blackstone and scientist at
FICO.
INDIAN PERSPECTIVE:-
The innovative DeepSeek AI and its rising popularity are
likely to make managements of US technology giants reconsider their AI capex
plans. This pivot from brute capex to cost-efficient AI platforms could benefit
Indian IT firms, some analysts said.
Services spending generally seems to follow the big-tech
capex cycle, just like during the cloud adoption phase, said MOFSL, which sees
similar developments with AI. It said Indian IT and outsourced engineering
often take a backseat during capex cycles but become integral to the value
chain once the focus shifts from innovation to cost optimization.
DISCLAIMER:- This is a purely a compiled, and edited ‘BLOG’—developed as
requested by a few students to familiarize
themselves on DEEP SEEK--for competitive examinations, and
circulated. Blogger has performed a 'REPORTER'S job. This BLOG will be updated.
Source:--
1.0 Business
Standard.
2.0 Rice
University.
3.0 MINT. & BLOOMBERG
4.0 You
can chat with DeepSeek-V3 on DeepSeek's official website: chat.deepseek.com
5.0 The
Group also provide Open AI-Compatible API at DeepSeek Platform:
platform.deepseek.com
6.0 Image:-
Google.
Comments
Post a Comment