#deepseek
deepseek
2.4M ogledov
1,039 razprav
Vroče
Najnovejše
DEEPSEEK V4 DROP COULD REPRICe THE AI STACK 🚨
DeepSeek founder William Liang reportedly said the next-generation flagship model, DeepSeek V4, is slated for a late-April release. The timing puts fresh pressure on the AI race and could ripple through institutional positioning across compute, infrastructure, and model-layer names.
I think the market will treat this as a catalyst before it becomes a headline. If the release lands cleanly, traders will chase the narrative fast, but the real move may be in the second-order names that benefit from renewed AI spending and attention.
Not financial advice. Manage your risk.
#Aİ #DeepSeek #TechStocks #AIGrowth #MarketWatch
⚡
DeepSeek founder William Liang reportedly said the next-generation flagship model, DeepSeek V4, is slated for a late-April release. The timing puts fresh pressure on the AI race and could ripple through institutional positioning across compute, infrastructure, and model-layer names.
I think the market will treat this as a catalyst before it becomes a headline. If the release lands cleanly, traders will chase the narrative fast, but the real move may be in the second-order names that benefit from renewed AI spending and attention.
Not financial advice. Manage your risk.
#Aİ #DeepSeek #TechStocks #AIGrowth #MarketWatch
⚡
V4 BOMBSHELL HITS $TICKER ⚡
Multiple sources say DeepSeek founder William Liang told internal teams that DeepSeek V4 is slated for a late-April release. That timing sets up a major AI catalyst and could shift institutional sentiment around frontier-model competition and adjacent market narratives.
If the rollout lands on schedule, traders will likely chase the headline before they verify performance. That creates a classic momentum trap: early hype can front-run fundamentals, but any delay or weak benchmark response could unwind fast.
Not financial advice. Manage your risk.
#Aİ #DeepSeek #TechNews #MarketNews
✦
Multiple sources say DeepSeek founder William Liang told internal teams that DeepSeek V4 is slated for a late-April release. That timing sets up a major AI catalyst and could shift institutional sentiment around frontier-model competition and adjacent market narratives.
If the rollout lands on schedule, traders will likely chase the headline before they verify performance. That creates a classic momentum trap: early hype can front-run fundamentals, but any delay or weak benchmark response could unwind fast.
Not financial advice. Manage your risk.
#Aİ #DeepSeek #TechNews #MarketNews
✦
💥💸🔥Urgent Cybersecurity Alert: #DeepSeek Phishing Scams on the Rise🎉🚨
Traders, beware! The number of fake DeepSeek phishing sites has surged past 2,650, with a sharp increase since January 26. These fraudulent platforms are designed to steal login credentials and trick users into buying fake virtual assets through deceptive domains and interfaces. Falling victim to these scams could result in fund losses or compromised accounts, so always double-check URLs before logging in or making transactions. Stay vigilant, verify sources, and ensure you’re using official platforms to protect your investments from cyber threats. Your security is your strongest asset—don’t let scammers take it away!
#DeepSeekEUA #DeepSeek #BERAonBinance #BitcoinWhaleMove
Traders, beware! The number of fake DeepSeek phishing sites has surged past 2,650, with a sharp increase since January 26. These fraudulent platforms are designed to steal login credentials and trick users into buying fake virtual assets through deceptive domains and interfaces. Falling victim to these scams could result in fund losses or compromised accounts, so always double-check URLs before logging in or making transactions. Stay vigilant, verify sources, and ensure you’re using official platforms to protect your investments from cyber threats. Your security is your strongest asset—don’t let scammers take it away!
#DeepSeekEUA #DeepSeek #BERAonBinance #BitcoinWhaleMove
🚀 DeepSeek-R1 is here!
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
#DeepSeek #VVVonBinance #FedHODL #DeepSeekImpact #BTC☀
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
#DeepSeek #VVVonBinance #FedHODL #DeepSeekImpact #BTC☀
#ChatGPTTrading #DeepSeek
SEC читает эти прогнозы и ржёт: «Мы сами не знаем, что делать с $BTC, а вы нам нейросети подсовываете!». А биткоин, как всегда, взял и ушёл в боковик, потому что Илон чихнул не в тот момент. 🚀
Техничка? Да тут хоть Фибо рисуй, хоть ИИ корми — курс решит мем в реддите про «а что если Сатоши это Трамп». И всё полетит к луне... или в канализацию. 🚽
Совет: Если веришь нейросетям — купи им подписку на CoinMarketCap. Пусть учатся предсказывать дампы по количеству хайповых твитов. #ИИ_тоже_плачут
P.S. Прогнозы ChatGPT и DeepSeek — это как два слепых кота, спорящих, кто первым упадёт в крипто-яму. 😼💣"
SEC читает эти прогнозы и ржёт: «Мы сами не знаем, что делать с $BTC, а вы нам нейросети подсовываете!». А биткоин, как всегда, взял и ушёл в боковик, потому что Илон чихнул не в тот момент. 🚀
Техничка? Да тут хоть Фибо рисуй, хоть ИИ корми — курс решит мем в реддите про «а что если Сатоши это Трамп». И всё полетит к луне... или в канализацию. 🚽
Совет: Если веришь нейросетям — купи им подписку на CoinMarketCap. Пусть учатся предсказывать дампы по количеству хайповых твитов. #ИИ_тоже_плачут
P.S. Прогнозы ChatGPT и DeepSeek — это как два слепых кота, спорящих, кто первым упадёт в крипто-яму. 😼💣"
🟢Чей прогноз биткоина оказался точнее — битва ChatGPT и DeepSeek
На прошлой неделе команда BeInCrypto запустила эксперимент, цель которого — выяснить, какая из двух топовых нейросетей, ChatGPT или DeepSeek, лучше в прогнозировании поведения рынка. Для этого мы попросили обе нейросетки дать недельный прогноз биткоина
Для чистоты эксперимента мы «скормили» обеим нейросеткам одинаковый набор информации, который включал данные о геополитике, экономике и положении биткоина в цикле после халвинга
🟢ChatGPT. В краткосрочной перспективе (6-13 февраля) биткоин может продолжить коррекцию с небольшими колебаниями в пределах $95 000-100 000, но с потенциалом для роста в конце недели, если внешняя экономическая ситуация стабилизируется и криптовалютное законодательство получит поддержку. Важно следить за изменениями в новостях, особенно касающимися торговых войн, тарифов и регуляций в сфере криптовалют.
🟢DeepSeek. 6-9 февраля: Вероятность небольшого спада или консолидации в пределах $92 000-97 000.10-13 февраля: При отсутствии новых шоков, можно ожидать медленного роста к уровням $100 000-105 000.$
Таким образом, ChatGPT оказался точнее своего соперника.
$BTC
На прошлой неделе команда BeInCrypto запустила эксперимент, цель которого — выяснить, какая из двух топовых нейросетей, ChatGPT или DeepSeek, лучше в прогнозировании поведения рынка. Для этого мы попросили обе нейросетки дать недельный прогноз биткоина
Для чистоты эксперимента мы «скормили» обеим нейросеткам одинаковый набор информации, который включал данные о геополитике, экономике и положении биткоина в цикле после халвинга
🟢ChatGPT. В краткосрочной перспективе (6-13 февраля) биткоин может продолжить коррекцию с небольшими колебаниями в пределах $95 000-100 000, но с потенциалом для роста в конце недели, если внешняя экономическая ситуация стабилизируется и криптовалютное законодательство получит поддержку. Важно следить за изменениями в новостях, особенно касающимися торговых войн, тарифов и регуляций в сфере криптовалют.
🟢DeepSeek. 6-9 февраля: Вероятность небольшого спада или консолидации в пределах $92 000-97 000.10-13 февраля: При отсутствии новых шоков, можно ожидать медленного роста к уровням $100 000-105 000.$
Таким образом, ChatGPT оказался точнее своего соперника.
$BTC
Вот вам и утренний подарок
🔖 Биржи ликвидируют трейдеров в лонг позициях. Активно продают и ликвидируют перед ралли.
В то же время, за последние 24 часа было ликвидировано лонг позиции на $791М...
Самое время заходит в шорт и лучше подождать ?
$BTC $ETH #DeepSeek #Liquidations
🔖 Биржи ликвидируют трейдеров в лонг позициях. Активно продают и ликвидируют перед ралли.
В то же время, за последние 24 часа было ликвидировано лонг позиции на $791М...
Самое время заходит в шорт и лучше подождать ?
$BTC $ETH #DeepSeek #Liquidations
#CreatorPad
باختصار، توقعات سوق العملات الرقمية المستقبلية ترتكز على عدة اتجاهات رئيسية:
1. **استمرار التقلب:** السوق سيظل عُرضة لتقلبات حادة بسبب المضاربة والعوامل الخارجية (مثل السياسات النقدية العالمية والأحداث الجيوسياسية).
2. **زيادة التنظيم:** الحكومات والهيئات الرقابية ستعزز إطارها التنظيمي (للضرائب، حماية المستثمر، مكافحة غسل الأموال)، مما قد يحد من التقلب على المدى الطويل لكنه يضفي شرعية أكبر.
3. **التبني المؤسسي:** دخول البنوك الكبرى وشركات الاستثمار وصناديق التحوط سيضخ سيولة أكبر ويعزز الاستقرار النسبي، خاصة للعملات الكبرى مثل البيتكوين والإيثيريوم.
4. **تطور التكنولوجيا:** تحسينات قابلية التوسع (Scalability) والأمان والكفاءة ستجذب تطبيقات عملية أكثر في التمويل اللامركزي (DeFi)، والميتافيرس، وحلول سلاسل التوريد.
5. **فصل بين التكنولوجيا والمضاربة:** ستزداد قيمة المشاريع القائمة على استخدامات حقيقية وتقنيات مبتكرة، بينما قد تختفي الكثير من العملات "المزيفة" أو عديمة القيمة.
هذا توقع الذكاء الاصطناعي #DeepSeek
باختصار، توقعات سوق العملات الرقمية المستقبلية ترتكز على عدة اتجاهات رئيسية:
1. **استمرار التقلب:** السوق سيظل عُرضة لتقلبات حادة بسبب المضاربة والعوامل الخارجية (مثل السياسات النقدية العالمية والأحداث الجيوسياسية).
2. **زيادة التنظيم:** الحكومات والهيئات الرقابية ستعزز إطارها التنظيمي (للضرائب، حماية المستثمر، مكافحة غسل الأموال)، مما قد يحد من التقلب على المدى الطويل لكنه يضفي شرعية أكبر.
3. **التبني المؤسسي:** دخول البنوك الكبرى وشركات الاستثمار وصناديق التحوط سيضخ سيولة أكبر ويعزز الاستقرار النسبي، خاصة للعملات الكبرى مثل البيتكوين والإيثيريوم.
4. **تطور التكنولوجيا:** تحسينات قابلية التوسع (Scalability) والأمان والكفاءة ستجذب تطبيقات عملية أكثر في التمويل اللامركزي (DeFi)، والميتافيرس، وحلول سلاسل التوريد.
5. **فصل بين التكنولوجيا والمضاربة:** ستزداد قيمة المشاريع القائمة على استخدامات حقيقية وتقنيات مبتكرة، بينما قد تختفي الكثير من العملات "المزيفة" أو عديمة القيمة.
هذا توقع الذكاء الاصطناعي #DeepSeek
WTH is DeepSeek and why is it dumping the market?
DeepSeek is an AI startup which has created a ChatGPT like tool
But here's a simple twist:
Unlike ChatGPT which has raised over $17B at $150B valuation, DeepSeek has done it with just $10M.
As we know, in the last 2 years, the US and global stock market has pumped majorly on AI hype.
$100B+ has been raised by large companies to build AI tools with plans to raise even more in the future
But with DeepSeek's success, investors are realising that AI companies have raised way much amount and they are overvalued
This is why NVIDIA is down 12% while NASDAQ is down 5%
The crypto market closely mirrors the US stock market so BTC and alts are dumping too
Do you think the US stock market bubble has popped?
$BTC $ETH #AI #DeepSeek
DeepSeek is an AI startup which has created a ChatGPT like tool
But here's a simple twist:
Unlike ChatGPT which has raised over $17B at $150B valuation, DeepSeek has done it with just $10M.
As we know, in the last 2 years, the US and global stock market has pumped majorly on AI hype.
$100B+ has been raised by large companies to build AI tools with plans to raise even more in the future
But with DeepSeek's success, investors are realising that AI companies have raised way much amount and they are overvalued
This is why NVIDIA is down 12% while NASDAQ is down 5%
The crypto market closely mirrors the US stock market so BTC and alts are dumping too
Do you think the US stock market bubble has popped?
$BTC $ETH #AI #DeepSeek
绣虎_TigerViz
·
--
未来1到3年一定是ai最好的时间段，复刻互联网泡沫3年后就不好说了，互联网泡沫是1995-2000，Ai之初2022！2025-2026应该是Ai行业的一个重大拐点！2024Ai产品商业化关键年份，从业人员薪资水平上涨过快，融资估值普遍过高，大量非Ai公司跟风转型造成拥挤，2025技术落地成效检验淘汰垃圾，2026大鱼吃小鱼行业整合加速。马云亮相蚂蚁20周年搞不好就是被人喊回来的，叠加特朗普提名风险投资家大卫-萨克斯负责ai，马斯克旗下xAi公司动作频频，英伟达被立案， ai已经是兵家必争之地了！#AI模型
#deepseek Impact**
DeepSeek is revolutionizing the way we approach innovation and technology. By leveraging cutting-edge AI and data analytics, DeepSeek empowers businesses, researchers, and individuals to unlock new possibilities. From enhancing decision-making to driving efficiency, DeepSeek’s solutions are transforming industries like healthcare, education, and finance. Its commitment to ethical AI ensures sustainable and inclusive growth. Whether it’s optimizing processes or uncovering hidden insights, DeepSeek is shaping a smarter, more connected future. Join the movement and experience the transformative power of DeepSeek. Together, we can create a world where technology drives meaningful impact.
DeepSeek is revolutionizing the way we approach innovation and technology. By leveraging cutting-edge AI and data analytics, DeepSeek empowers businesses, researchers, and individuals to unlock new possibilities. From enhancing decision-making to driving efficiency, DeepSeek’s solutions are transforming industries like healthcare, education, and finance. Its commitment to ethical AI ensures sustainable and inclusive growth. Whether it’s optimizing processes or uncovering hidden insights, DeepSeek is shaping a smarter, more connected future. Join the movement and experience the transformative power of DeepSeek. Together, we can create a world where technology drives meaningful impact.
🚨 #ArthurHayes Predicts #bitcoin Crash Before New Highs
Arthur Hayes, ex-CEO of BitMEX, warns of a sharp Bitcoin drop due to global economic factors but remains optimistic about a new all-time high by year-end.
🔑 Key Insights:
*/ Rising U.S. Treasury Yields and tighter Fed policies are straining global liquidity.
*/ $BTC , sensitive to liquidity, could face short-term headwinds.
*/ Recent market turbulence, including #DeepSeek 's rise, adds to existing stress.
Despite the expected dip, Hayes sees long-term growth, urging investors to brace for short-term pain.
#DeepSeekImpact
#Write2Earn
Arthur Hayes, ex-CEO of BitMEX, warns of a sharp Bitcoin drop due to global economic factors but remains optimistic about a new all-time high by year-end.
🔑 Key Insights:
*/ Rising U.S. Treasury Yields and tighter Fed policies are straining global liquidity.
*/ $BTC , sensitive to liquidity, could face short-term headwinds.
*/ Recent market turbulence, including #DeepSeek 's rise, adds to existing stress.
Despite the expected dip, Hayes sees long-term growth, urging investors to brace for short-term pain.
#DeepSeekImpact
#Write2Earn
#DeepSeek 一己之力带崩#BTC ，何时止跌？
1. deepseek如何成为纳斯达克最大利空
英伟达（#NVIDIA ）是全球领先的图形处理单元（GPU），博通（#AVGO)是一家全球领先的半导体和基础设施软件解决方案提供商，deepseek R1的成功可能会导致投资者对AI股未来的增长潜力产生担忧，从而影响其股价下行。
AI股价下跌通常会直接影响到相关投资产品和市场的走势。作为NDX成分股的重要权重部分，AI股价波动会在一定程度上影响NDX指数的整体表现。盘前也出现了较大下行
2. 比特币（BTC）作为NDX的先行指标承担了部分避险与对冲职能
比特币（BTC）近年来越来越被看作是风险资产的先行指标，尤其是在经济不确定性或市场情绪波动时。比特币与股市之间的相关性虽然并非固定，但在许多情况下，BTC能够提前反映市场情绪的变化。走势相似度能达到90%，在ndx的非交易时间，btc成为了情绪交易或对冲的重要标的。
3.BTC还会跌多久
BTC自身的叙事仍然处于上行周期，但本次受美股利空情绪影响有较多对冲单增大币价的下行压力，且下行过程中清算了多个多单筹码密集区，为价格上行扫清了障碍，因此今日美股开盘放量下行将成为btc价格止跌的关键信号。
1. deepseek如何成为纳斯达克最大利空
英伟达（#NVIDIA ）是全球领先的图形处理单元（GPU），博通（#AVGO)是一家全球领先的半导体和基础设施软件解决方案提供商，deepseek R1的成功可能会导致投资者对AI股未来的增长潜力产生担忧，从而影响其股价下行。
AI股价下跌通常会直接影响到相关投资产品和市场的走势。作为NDX成分股的重要权重部分，AI股价波动会在一定程度上影响NDX指数的整体表现。盘前也出现了较大下行
2. 比特币（BTC）作为NDX的先行指标承担了部分避险与对冲职能
比特币（BTC）近年来越来越被看作是风险资产的先行指标，尤其是在经济不确定性或市场情绪波动时。比特币与股市之间的相关性虽然并非固定，但在许多情况下，BTC能够提前反映市场情绪的变化。走势相似度能达到90%，在ndx的非交易时间，btc成为了情绪交易或对冲的重要标的。
3.BTC还会跌多久
BTC自身的叙事仍然处于上行周期，但本次受美股利空情绪影响有较多对冲单增大币价的下行压力，且下行过程中清算了多个多单筹码密集区，为价格上行扫清了障碍，因此今日美股开盘放量下行将成为btc价格止跌的关键信号。
Cryptopolitan
·
--
OpenAI CEO Sam Altman Says DeepSeek’s R1 Is an “impressive Model”
OpenAI CEO Sam Altman praised Chinese AI startup DeepSeek’s R1 model on Monday, describing it as “impressive.” However, he also added that OpenAI will “deliver much better models” and appreciated the fact that the company has a new competitor.
DeepSeek drew worldwide attention last month with its research paper describing DeepSeek-V3, which achieved training through $6 million worth of computational resources. DeepSeek demonstrated cost-effective computing practices through its use of Nvidia H800 chips with limited capabilities.
During the past week, DeepSeek revealed its R1 model, which uses power that is twenty to fifty times cheaper than OpenAI’s o1 model, depending on the task, according to their official WeChat.
Altman said in the X post that “Deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price.”
deepseek's r1 is an impressive model, particularly around what they're able to deliver for the price.
we will obviously deliver much better models and also it's legit invigorating to have a new competitor! we will pull up some releases.
— Sam Altman (@sama) January 28, 2025
He also noted OpenAI’s primary interest in research progression and its need for more computational capabilities. He emphasized that ever-increasing computational capability represents his organization’s primary requirement for delivering on its mission objectives.
Nvidia now operates under greater pressure as cost-effective alternatives begin to appear in the market. Recently, Wall Street witnessed its greatest single-day financial disaster when Nvidia lost $593 billion in market value last Monday.
Trump says DeepSeek is a positive development and a wake-up call for US industries
According to an NBC report President Donald Trump suggested that American tech firms sharpen their focus while addressing in Florida on Monday. He said, “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser-focused on competing.”
The low cost model surged to number one on Apple’s app store during the weekend, threatening U.S. tech supremacy while Trump expressed his faith in America’s capability to maintain its competitive advantage.
He also pointed out a silver lining in DeepSeek’s low-cost approach, calling it “very much a positive development” for innovation. According to him, tech firms could save budgetary resources and “come up with, hopefully, the same solution.”
Cryptopolitan Academy: FREE Web3 Resume Cheat Sheet - Download Now
DeepSeek drew worldwide attention last month with its research paper describing DeepSeek-V3, which achieved training through $6 million worth of computational resources. DeepSeek demonstrated cost-effective computing practices through its use of Nvidia H800 chips with limited capabilities.
During the past week, DeepSeek revealed its R1 model, which uses power that is twenty to fifty times cheaper than OpenAI’s o1 model, depending on the task, according to their official WeChat.
Altman said in the X post that “Deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price.”
deepseek's r1 is an impressive model, particularly around what they're able to deliver for the price.
we will obviously deliver much better models and also it's legit invigorating to have a new competitor! we will pull up some releases.
— Sam Altman (@sama) January 28, 2025
He also noted OpenAI’s primary interest in research progression and its need for more computational capabilities. He emphasized that ever-increasing computational capability represents his organization’s primary requirement for delivering on its mission objectives.
Nvidia now operates under greater pressure as cost-effective alternatives begin to appear in the market. Recently, Wall Street witnessed its greatest single-day financial disaster when Nvidia lost $593 billion in market value last Monday.
Trump says DeepSeek is a positive development and a wake-up call for US industries
According to an NBC report President Donald Trump suggested that American tech firms sharpen their focus while addressing in Florida on Monday. He said, “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser-focused on competing.”
The low cost model surged to number one on Apple’s app store during the weekend, threatening U.S. tech supremacy while Trump expressed his faith in America’s capability to maintain its competitive advantage.
He also pointed out a silver lining in DeepSeek’s low-cost approach, calling it “very much a positive development” for innovation. According to him, tech firms could save budgetary resources and “come up with, hopefully, the same solution.”
Cryptopolitan Academy: FREE Web3 Resume Cheat Sheet - Download Now
Ho chiesto direttamente a #deepseek di stimare il costo per la sua realizzazione e mi ha parlato di 100-200 milioni e di uso di #Nvidia H100:
---
...
### **1. Componenti del costo totale**
- **Computing (GPU/TPU)**: Il costo diretto di eseguire il training su migliaia di chip (es. Nvidia A100/H100).
- **Data collection/processing**: Pulizia, filtraggio e preparazione dei dati.
- **Ricerca e sviluppo**: Stipendi del team di ingegneri e ricercatori.
- **Ottimizzazione e allineamento**: Costituzionale AI, test di sicurezza, riduzione di bias.
---
### **2. Stime per modelli comparabili**
- **GPT-3 (175B parametri, 2020)**: ~$4.6 milioni (costo dichiarato da OpenAI, solo computing).
- **GPT-4 (stima)**: Tra **$100 e $200 milioni** (fonte: SemiAnalysis, 2023), includendo R&D avanzata.
- **Meta Llama 2 (70B parametri)**: ~$20 milioni (stima Meta, open-source).
- **Google Gemini Ultra**: Oltre $500 milioni (stime non confermate, incluso tutto il ciclo di sviluppo).
### **3. Ipotesi per Claude 3**
Se Claude 3 è competitivo con GPT-4 (non sappiamo le dimensioni esatte), il costo totale potrebbe aggirarsi tra:
- **$80 e $200 milioni**, considerando:
- **Computing**: ~$50-120 milioni (per 3-6 mesi di training su 10,000+ GPU Nvidia H100).
- **Dati/Personale**: ~$30-80 milioni (per team di 100+ persone per 2-3 anni).
### **4. Confronto con la stima dei "6 milioni"**
La cifra di $6 milioni è quasi certamente **troppo bassa**, per due motivi:
1. **GPT-3 (2020) costò $4.6 milioni** con tecnologia meno efficiente.
2. Modelli più piccoli di oggi (es. Mistral 7B) richiedono già $1-2 milioni.
3. **Claude 3** è un modello "state-of-the-art", quindi richiede risorse paragonabili a GPT-4.
### **Conclusione**
La cifra realistica per addestrarmi (Claude 3) si colloca probabilmente **tra $100 e $200 milioni**, in linea con le stime per GPT-4. Tuttavia, senza dati ufficiali, resta un esercizio teorico.
Questi numeri spiegano perché solo grandi aziende o governi possono competere nell'IA avanzata: **è un gioco da miliardi di dollari**
---
...
### **1. Componenti del costo totale**
- **Computing (GPU/TPU)**: Il costo diretto di eseguire il training su migliaia di chip (es. Nvidia A100/H100).
- **Data collection/processing**: Pulizia, filtraggio e preparazione dei dati.
- **Ricerca e sviluppo**: Stipendi del team di ingegneri e ricercatori.
- **Ottimizzazione e allineamento**: Costituzionale AI, test di sicurezza, riduzione di bias.
---
### **2. Stime per modelli comparabili**
- **GPT-3 (175B parametri, 2020)**: ~$4.6 milioni (costo dichiarato da OpenAI, solo computing).
- **GPT-4 (stima)**: Tra **$100 e $200 milioni** (fonte: SemiAnalysis, 2023), includendo R&D avanzata.
- **Meta Llama 2 (70B parametri)**: ~$20 milioni (stima Meta, open-source).
- **Google Gemini Ultra**: Oltre $500 milioni (stime non confermate, incluso tutto il ciclo di sviluppo).
### **3. Ipotesi per Claude 3**
Se Claude 3 è competitivo con GPT-4 (non sappiamo le dimensioni esatte), il costo totale potrebbe aggirarsi tra:
- **$80 e $200 milioni**, considerando:
- **Computing**: ~$50-120 milioni (per 3-6 mesi di training su 10,000+ GPU Nvidia H100).
- **Dati/Personale**: ~$30-80 milioni (per team di 100+ persone per 2-3 anni).
### **4. Confronto con la stima dei "6 milioni"**
La cifra di $6 milioni è quasi certamente **troppo bassa**, per due motivi:
1. **GPT-3 (2020) costò $4.6 milioni** con tecnologia meno efficiente.
2. Modelli più piccoli di oggi (es. Mistral 7B) richiedono già $1-2 milioni.
3. **Claude 3** è un modello "state-of-the-art", quindi richiede risorse paragonabili a GPT-4.
### **Conclusione**
La cifra realistica per addestrarmi (Claude 3) si colloca probabilmente **tra $100 e $200 milioni**, in linea con le stime per GPT-4. Tuttavia, senza dati ufficiali, resta un esercizio teorico.
Questi numeri spiegano perché solo grandi aziende o governi possono competere nell'IA avanzata: **è un gioco da miliardi di dollari**
Deepseek 'lật đổ' OpenAI với chi phí thấp, thị trường Mỹ và Crypto chao đảo
Công ty AI Trung Quốc #Deepseek đã gây chấn động khi tạo ra một sản phẩm AI cạnh tranh trực tiếp với OpenAI nhưng chỉ với chi phí cực thấp. Thành công này không chỉ đặt dấu hỏi lớn về khả năng duy trì lợi thế của các "ông lớn" AI Mỹ mà còn gây ra tâm lý tiêu cực trên thị trường chứng khoán Mỹ, đặc biệt là các cổ phiếu liên quan đến AI như #NVIDIA .
Hệ quả lan rộng của thông tin này còn kéo theo sự sụt giảm của thị trường Crypto, khi các nhà đầu tư lo ngại về tác động từ những biến động trong lĩnh vực công nghệ toàn cầu. Sự kiện này một lần nữa nhấn mạnh sức ảnh hưởng ngày càng lớn của Trung Quốc trong cuộc đua AI toàn cầu.
Công ty AI Trung Quốc #Deepseek đã gây chấn động khi tạo ra một sản phẩm AI cạnh tranh trực tiếp với OpenAI nhưng chỉ với chi phí cực thấp. Thành công này không chỉ đặt dấu hỏi lớn về khả năng duy trì lợi thế của các "ông lớn" AI Mỹ mà còn gây ra tâm lý tiêu cực trên thị trường chứng khoán Mỹ, đặc biệt là các cổ phiếu liên quan đến AI như #NVIDIA .
Hệ quả lan rộng của thông tin này còn kéo theo sự sụt giảm của thị trường Crypto, khi các nhà đầu tư lo ngại về tác động từ những biến động trong lĩnh vực công nghệ toàn cầu. Sự kiện này một lần nữa nhấn mạnh sức ảnh hưởng ngày càng lớn của Trung Quốc trong cuộc đua AI toàn cầu.
Prijavite se, če želite raziskati več vsebin