#deepseek
deepseek
2.5M views
1,080 Discussing
Hot
Latest
Domestic computing power milestone: The Ascend 910C successfully completed the full parameter training of the DeepSeek 16 trillion parameter model.
A collaborative team from Shenzhen Hertao College, Harbin Institute of Technology (Shenzhen), Shenzhen Big Data Research Institute, and Huawei has successfully run the full parameter training of the DeepSeek-V4-Pro 16 trillion parameter model on a cluster of over a thousand Ascend 910C units. The system ran 1500 training steps without interruption, achieving a model compute utilization rate of over 30%, and improving key operator efficiency by 14%.
Why it matters: This marks the first time a third-party institution has completed full parameter training of a trillion parameter scale model on a domestic computing power platform, signifying a breakthrough in the technological loop from "only supporting inference" to "carrying out large-scale model training" for domestic AI chips.
#国产算力 #AI #DeepSeek #Huawei Ascend
A collaborative team from Shenzhen Hertao College, Harbin Institute of Technology (Shenzhen), Shenzhen Big Data Research Institute, and Huawei has successfully run the full parameter training of the DeepSeek-V4-Pro 16 trillion parameter model on a cluster of over a thousand Ascend 910C units. The system ran 1500 training steps without interruption, achieving a model compute utilization rate of over 30%, and improving key operator efficiency by 14%.
Why it matters: This marks the first time a third-party institution has completed full parameter training of a trillion parameter scale model on a domestic computing power platform, signifying a breakthrough in the technological loop from "only supporting inference" to "carrying out large-scale model training" for domestic AI chips.
#国产算力 #AI #DeepSeek #Huawei Ascend
DeepSeek Fix Test ✅
Recently completed a series of optimizations and stability fixes for the DeepSeek models, including context handling tweaks and response quality improvements. Ongoing iterations.
#DeepSeek #AI
Recently completed a series of optimizations and stability fixes for the DeepSeek models, including context handling tweaks and response quality improvements. Ongoing iterations.
#DeepSeek #AI
In response to the panic caused by input characters leading to DeepSeek exposing others' chats, DeepSeek released an investigation result today. They confirmed that the abnormal replies were triggered by special characters causing model hallucinations, and there was no failure in multi-tenant isolation or data privacy breaches. Certain tags can induce the model to extract training data, fabricating chat logs using the current date from system prompts. The official statement has completely put to rest conspiracy theories about the collapse of the underlying isolation architecture. DeepSeek stated that they will enhance the model's capability to recognize and handle special characters through targeted training.
Why it matters: This incident clarifies the technical boundaries of AI model privacy security; special characters induce hallucinations rather than actual data leaks, and this conclusion is significant for AI security research.
#DeepSeek #AI #隐私安全 #大模型 #Web3
Why it matters: This incident clarifies the technical boundaries of AI model privacy security; special characters induce hallucinations rather than actual data leaks, and this conclusion is significant for AI security research.
#DeepSeek #AI #隐私安全 #大模型 #Web3
Verified
DeepSeek Jumps into the 'Agentic Coding' Race
The latest job postings this week indicate that DeepSeek is developing an Agentic Coding Tool, aiming to compete with advanced solutions like Claude Code from Anthropic and Codex from OpenAI.
This shift isn't just an upgrade to coding assistant tools; it's a leap towards an 'intelligent agent' model capable of:
Understanding coding tasks independently
Executing end-to-end code modifications
Making technical decisions during development
Reducing reliance on direct human intervention
This trend reflects a broader race within the AI industry to transform programming from writing commands to managing self-operating smart systems.
If these tools succeed, they could reshape:
How software is built
The role of developers in tech teams
The speed of product development within AI and web3 companies
Notably, the competition is no longer just about having the 'best model', but about who can build the best fully autonomous coding agent.
⚡ The market is now shifting from assistive tools to intelligent execution systems... and the next phase may be more autonomous than we expect.
#DeepSeek #AI #AgenticAI
#claudecode #OpenAI
The latest job postings this week indicate that DeepSeek is developing an Agentic Coding Tool, aiming to compete with advanced solutions like Claude Code from Anthropic and Codex from OpenAI.
This shift isn't just an upgrade to coding assistant tools; it's a leap towards an 'intelligent agent' model capable of:
Understanding coding tasks independently
Executing end-to-end code modifications
Making technical decisions during development
Reducing reliance on direct human intervention
This trend reflects a broader race within the AI industry to transform programming from writing commands to managing self-operating smart systems.
If these tools succeed, they could reshape:
How software is built
The role of developers in tech teams
The speed of product development within AI and web3 companies
Notably, the competition is no longer just about having the 'best model', but about who can build the best fully autonomous coding agent.
⚡ The market is now shifting from assistive tools to intelligent execution systems... and the next phase may be more autonomous than we expect.
#DeepSeek #AI #AgenticAI
#claudecode #OpenAI
DeepSeek has locked in a 75% discount as a permanent price, and this price war is really out to crush the competition.
The models are getting cheaper, and in the end, it's the downstream application layer that's raking in the rewards, while those AI Agent narratives are getting a bit more solid.
But don't get too hyped just because it has 'AI' in the name; you still need to sift through and find the projects that are actually building products and have a strong community. Narratives are one thing, but you gotta calculate your risk-to-reward ratio yourself. #AI #DeepSeek $FET $TAO
The models are getting cheaper, and in the end, it's the downstream application layer that's raking in the rewards, while those AI Agent narratives are getting a bit more solid.
But don't get too hyped just because it has 'AI' in the name; you still need to sift through and find the projects that are actually building products and have a strong community. Narratives are one thing, but you gotta calculate your risk-to-reward ratio yourself. #AI #DeepSeek $FET $TAO
Verified
China's looking to swap out Nvidia for homegrown chips: who's stepping up?
The buzz from DeepSeek about their new AI model being compatible with domestic chips in the works has sparked interest among investors and Chinese manufacturers.
This news reinforces the belief that China could cut down its reliance on Nvidia, especially with the export restrictions from the US. An article from Rest of World highlights which Chinese companies could take over the chip giant's spot.$USDC #NVIDIA #DeepSeek
The buzz from DeepSeek about their new AI model being compatible with domestic chips in the works has sparked interest among investors and Chinese manufacturers.
This news reinforces the belief that China could cut down its reliance on Nvidia, especially with the export restrictions from the US. An article from Rest of World highlights which Chinese companies could take over the chip giant's spot.$USDC #NVIDIA #DeepSeek
This passage perfectly captures the essence of Nangong Yuan's quote: "Every penny you lose fuels the industry's progress."
"Liang Wenfeng is quite the maverick; he graduated from Zhejiang University but didn't pursue a regular job. Instead, he locked himself away for three years trading stocks, and ultimately dominated the domestic quant scene, leaving retail investors in A-shares crying out. He raised funds to dive into AI, truly the type of nerd who would drop out in his sophomore year in the U.S. to chase after Ivy League success and entrepreneurship."
#DeepSeek #梁文锋 #AI #投资 #认知
"Liang Wenfeng is quite the maverick; he graduated from Zhejiang University but didn't pursue a regular job. Instead, he locked himself away for three years trading stocks, and ultimately dominated the domestic quant scene, leaving retail investors in A-shares crying out. He raised funds to dive into AI, truly the type of nerd who would drop out in his sophomore year in the U.S. to chase after Ivy League success and entrepreneurship."
#DeepSeek #梁文锋 #AI #投资 #认知
🚨 THIS IS INSANE: DeepSeek is now reportedly up to 50x CHEAPER than OpenAI and Anthropic for AI token costs.
📉 After a massive 75% price cut, some DeepSeek inference costs reportedly dropped to fractions of a cent per million tokens.
Approximate costs for 1 BILLION output tokens:
▪️ DeepSeek: ~$3,480
▪️ Claude Sonnet: ~$15,000
▪️ OpenAI GPT-5.5: ~$30,000
⚠️ This is becoming a huge issue for enterprises because AI coding agents can burn MILLIONS of tokens per day.
Reports claim: ▪️ Uber exhausted its annual AI budget by April
▪️ Microsoft is reportedly pushing engineers away from expensive Claude usage toward cheaper Copilot tools
🔥 The smarter AI gets, the more reasoning tokens it consumes behind the scenes — massively increasing compute costs.
The future AI winners may not be the smartest models…
They may be the CHEAPEST models that are “good enough.”
#AI #DeepSeek #OpenAI #NVIDIA #Technology
$BTC $ETH $BNB
📉 After a massive 75% price cut, some DeepSeek inference costs reportedly dropped to fractions of a cent per million tokens.
Approximate costs for 1 BILLION output tokens:
▪️ DeepSeek: ~$3,480
▪️ Claude Sonnet: ~$15,000
▪️ OpenAI GPT-5.5: ~$30,000
⚠️ This is becoming a huge issue for enterprises because AI coding agents can burn MILLIONS of tokens per day.
Reports claim: ▪️ Uber exhausted its annual AI budget by April
▪️ Microsoft is reportedly pushing engineers away from expensive Claude usage toward cheaper Copilot tools
🔥 The smarter AI gets, the more reasoning tokens it consumes behind the scenes — massively increasing compute costs.
The future AI winners may not be the smartest models…
They may be the CHEAPEST models that are “good enough.”
#AI #DeepSeek #OpenAI #NVIDIA #Technology
$BTC $ETH $BNB
🚀 AI TRENDS | is reportedly expanding its AI ecosystem by forming a dedicated team to build a native AI coding tool designed for developers and software engineers. The project aims to integrate advanced code generation, debugging, and automation features directly into DeepSeek’s platform, allowing users to write and optimize code faster with AI assistance. 💻🤖
The move comes as competition in the AI coding assistant market intensifies, with global tech firms racing to dominate tools that can boost developer productivity. DeepSeek’s expansion highlights China’s growing focus on AI innovation and its ambition to compete with leading international AI companies in both consumer and enterprise technology markets.
If successful, the new coding tool could help developers automate repetitive tasks, improve software efficiency, and accelerate app development using generative AI capabilities. Analysts believe AI-powered coding assistants may become one of the biggest productivity drivers in the tech industry over the next few years. 📈
#AI #DeepSeek #ArtificialIntelligence #coding #technews
$BTC
$ETH
$BNB
The move comes as competition in the AI coding assistant market intensifies, with global tech firms racing to dominate tools that can boost developer productivity. DeepSeek’s expansion highlights China’s growing focus on AI innovation and its ambition to compete with leading international AI companies in both consumer and enterprise technology markets.
If successful, the new coding tool could help developers automate repetitive tasks, improve software efficiency, and accelerate app development using generative AI capabilities. Analysts believe AI-powered coding assistants may become one of the biggest productivity drivers in the tech industry over the next few years. 📈
#AI #DeepSeek #ArtificialIntelligence #coding #technews
$BTC
$ETH
$BNB
In the AI era, the buy-and-buy strategy of traditional internet giants seems to have hit a snag.
First, we had Zuckerberg's various acquisition plays that ended up being nothing but fluff; now, we see Deepseek's Liang Wenfeng turning down financing from Tencent and Alibaba.
Today, regarding whether Tencent AI is lagging, at the Tencent shareholder meeting, Chairman and CEO Ma Huateng stated, "A year ago, we thought we were on board, but then discovered that the ship had leaks. Now it feels like we've stepped on, but we can't sit down yet. We hope the ship can speed up a bit."
For those who hold Tencent stock, what’s your take?
#Aİ #DeepSeek
First, we had Zuckerberg's various acquisition plays that ended up being nothing but fluff; now, we see Deepseek's Liang Wenfeng turning down financing from Tencent and Alibaba.
Today, regarding whether Tencent AI is lagging, at the Tencent shareholder meeting, Chairman and CEO Ma Huateng stated, "A year ago, we thought we were on board, but then discovered that the ship had leaks. Now it feels like we've stepped on, but we can't sit down yet. We hope the ship can speed up a bit."
For those who hold Tencent stock, what’s your take?
#Aİ #DeepSeek
🤖 DeepSeek-V4 Preview — Officially Launched & Open-Sourced!
While markets are watching charts, the AI world just got a major update. 👇
DeepSeek-V4 is now live with:
📌 Million-token long context window
📌 Pro & Flash versions
📌 Instant API access via Baidu Qianfan (Day0 integration)
Open-source AI is disrupting the space just like DeFi disrupted traditional finance. The tools available to builders today are incredible.
Stay ahead. Build smart. 🔧
#DeepSeek #DeepSeekV4 #Aİ #Crypto #Blockchain
While markets are watching charts, the AI world just got a major update. 👇
DeepSeek-V4 is now live with:
📌 Million-token long context window
📌 Pro & Flash versions
📌 Instant API access via Baidu Qianfan (Day0 integration)
Open-source AI is disrupting the space just like DeFi disrupted traditional finance. The tools available to builders today are incredible.
Stay ahead. Build smart. 🔧
#DeepSeek #DeepSeekV4 #Aİ #Crypto #Blockchain
DeepSeek is gearing up for financing.
Not a small valuation.
Straight up to $45 billion.
Led by the national semiconductor fund.
AI is kicking into high gear 🇨🇳
This news carries a lot of weight.
DeepSeek is in talks for its first round of financing,
with a valuation of around $45 billion.
And the lead investor isn’t just any VC;
it’s a state-backed semiconductor fund from China.
Now, that’s interesting.
It’s no longer just an AI model company;
it's being integrated into a larger state-level industrial chain.
AI models, chips, computing power, domestic alternatives,
these lines are starting to tie together.
Previously, folks looked at DeepSeek
more for its model capabilities and cost advantages.
Now, if national semiconductor funds are getting involved,
the market will reevaluate it:
It’s not just a single-point AI application,
but part of China's AI infrastructure.
This also explains why the valuation can jump straight to $45 billion.
Because the funds are betting not on current revenue,
but on future positioning.
This line will definitely continue to ferment.
Do you think this round of financing for DeepSeek marks China's AI officially entering the national team stage,
or has the valuation already been pushed too high? 👀$ETH $CL #deepseek
Not a small valuation.
Straight up to $45 billion.
Led by the national semiconductor fund.
AI is kicking into high gear 🇨🇳
This news carries a lot of weight.
DeepSeek is in talks for its first round of financing,
with a valuation of around $45 billion.
And the lead investor isn’t just any VC;
it’s a state-backed semiconductor fund from China.
Now, that’s interesting.
It’s no longer just an AI model company;
it's being integrated into a larger state-level industrial chain.
AI models, chips, computing power, domestic alternatives,
these lines are starting to tie together.
Previously, folks looked at DeepSeek
more for its model capabilities and cost advantages.
Now, if national semiconductor funds are getting involved,
the market will reevaluate it:
It’s not just a single-point AI application,
but part of China's AI infrastructure.
This also explains why the valuation can jump straight to $45 billion.
Because the funds are betting not on current revenue,
but on future positioning.
This line will definitely continue to ferment.
Do you think this round of financing for DeepSeek marks China's AI officially entering the national team stage,
or has the valuation already been pushed too high? 👀$ETH $CL #deepseek
Log in to explore more content