The Deepseek is more like an AI supported search engine. My understanding its underneath MOE, the router likely using rule-base methodology to assign tokens to different Experts ( the much less computational equipped transformers). It will beat the other LLMs under the conditions that the inputs are highly aligned with the training datas, and for the other LLMs, still the whole model ( much more layers of the neural network) will be processed while per the deepseek as the token assignment performed by the router and as long as it is NOT based on the neural network but rule base, definitely the required number of steps will drastically reduced and then further reduced per the relatively less computational powerful “experts” -the transformers. Pros: faster and less resource required however, a big assumption the inputs must be relatively close/ correlated to the training data, it more like to faster and with intelligence to pick from the stored data to response. For other LLM, as they are build on powerful chips and multiplexing the input/output to develop the response and because of the computation power it use numerous / large amount of perimeter and the neural network to processing instead of the rule-base assignment that can miss important data that critically impacting the accuracy of the results, especially if the inputs are from the extremes ( 2 sides) of the population. Besides I guess the deepseek is much less useful for academic research ( eg for drug development, medical research that the other LLMs can base on the trained models and its computational power to emulate different scenarios for study).. i am hoping we can have chess playing between the Deepseek and Chatgpt 1, i guess the Chatgpt 1 will win. On the other hand, if we are not looking for a sophisticated engine for technology development and research but for chat engine, Deepseek is a good choice for mainlanders
In the context of their telephone call: 好自為之 Deepseek response : 在外交沟通中，尤其是中美之间，措辞往往非常谨慎，以传达信息而不加剧紧张局势。王毅使用“好自為之”可能反映了中国的立场，即美国在与中国的交往中应采取更加谨慎和负责任的态度. Chatgpt response : 喺呢個外交背景之下，王毅嘅講話係提醒盧比奧要喺處理雙邊關係方面，採取負責任同具建設性嘅態度。佢強調要堅守兩國領導人所設定嘅方向，並且鼓勵盧比奧為兩國利益同全球穩定作出正面嘅貢獻. Hello Mr. Shiu, Can you give your thoughts and analysis on Deepeek?
When you mentioned about Tikto as just reminds me about another interesting news today also . China has develop a New AI which known as “ Deepseek”. When the news was being announced earlier on today , I heard the US stock market may get fluctuated. I look forward to hearing the comments from Trump in due course . I also has a query . How this technology can be achieved by China if country like US has been imposing sanctions to China with all restriction of high technology chips to export from the last 2 years since ? Sounds to me like US is just tasting with their own medicine after all . What a surprise with the invention of “Deepseek” AI.
好自為之=you…watch out!
好自為之=等著瞧! (You wait and see....ha, ha...)
好自for your own good
為之behave accordingy
好似佢系川普，乜嘢事都知？ 全部都靠估，我就唔信吾加关税
你要計埋取消最惠國待遇，如是者，輸左除褲po上來，我賭😊
Trump 做嘢你估佢唔到，勿聴其言，請觀其行。上台前1日GAZA已停戰👍
美國指一個中國但也有一個中華民國，兩者沒有抵觸😅
You watch it!
I watching you😂😂😂
好似社團話睇路因住啊謝謝蕭生分享👍👍👍👍👍👍
支持
我會話: watch your steps 😁
各師兄各說各法🙏
nice😜
蕭生，講下Deepseek，美股跳水。
蕭山呀，Deepseek 點算呀？
The Deepseek is more like an AI supported search engine. My understanding its underneath MOE, the router likely using rule-base methodology to assign tokens to different Experts ( the much less computational equipped transformers). It will beat the other LLMs under the conditions that the inputs are highly aligned with the training datas, and for the other LLMs, still the whole model ( much more layers of the neural network) will be processed while per the deepseek as the token assignment performed by the router and as long as it is NOT based on the neural network but rule base, definitely the required number of steps will drastically reduced and then further reduced per the relatively less computational powerful “experts” -the transformers.
Pros: faster and less resource required however, a big assumption the inputs must be relatively close/ correlated to the training data, it more like to faster and with intelligence to pick from the stored data to response.
For other LLM, as they are build on powerful chips and multiplexing the input/output to develop the response and because of the computation power it use numerous / large amount of perimeter and the neural network to processing instead of the rule-base assignment that can miss important data that critically impacting the accuracy of the results, especially if the inputs are from the extremes ( 2 sides) of the population.
Besides I guess the deepseek is much less useful for academic research ( eg for drug development, medical research that the other LLMs can base on the trained models and its computational power to emulate different scenarios for study).. i am hoping we can have chess playing between the Deepseek and Chatgpt 1, i guess the Chatgpt 1 will win.
On the other hand, if we are not looking for a sophisticated engine for technology development and research but for chat engine, Deepseek is a good choice for mainlanders
You take care of yourself 😂😂😂
應該音譯海格撒特Hegseth
嘩，後面啲人打格仔，有進步👍
中共制裁既係參議員盧比奧, 唔係國務卿魯比奧
可以講下DeepSeek的睇法嗎？
可以講講關于DeepSeek的睇法嗎？
蕭生，可以睇睇tik tok 上面人反映算法變動問題，再報哩條新聞。
U better behave yourself.
❤❤❤❤😊
蕭生，這事已經發生了兩三天，你現在才討論，太慢了，在別的網站已經談了很多關於這事件．
😅
Rubio 唔識中文，王偽唔識英文，咁佢哋點通電話？😂
任何语言嘅粗口都喺世界通行嘅。。。😂😂
😊
蕭生，美國有150萬軍隊😅
同樣是選擇不賣TikTok ，現在退出市場和一月十九日前退出有什麼差別？我能想到的差別就是，如果在十九日前，TikTok 還可以安然退出。現在如果選擇不賣，已經不可能安然退出。現在它明顯被當成關稅的leverage，不賣的話還會被關稅報復。所以我怎麼看都是對TikTok來說更差的十字路口。
In the context of their telephone call: 好自為之
Deepseek response :
在外交沟通中，尤其是中美之间，措辞往往非常谨慎，以传达信息而不加剧紧张局势。王毅使用“好自為之”可能反映了中国的立场，即美国在与中国的交往中应采取更加谨慎和负责任的态度.
Chatgpt response :
喺呢個外交背景之下，王毅嘅講話係提醒盧比奧要喺處理雙邊關係方面，採取負責任同具建設性嘅態度。佢強調要堅守兩國領導人所設定嘅方向，並且鼓勵盧比奧為兩國利益同全球穩定作出正面嘅貢獻.
Hello Mr. Shiu, Can you give your thoughts and analysis on Deepeek?
Support
Look before you leap
Look before you trump ( jump )
係盧比奧
我理解的，好自为知 "Reflect on yourself."
我非常希望知道假若Rubio 不好自為之的話
會唔會第三次制裁佢呢？
🤭
ch no hear..
😂😂😂
片尾歌叫咩名？
‘You are going to face the music.
我認為不會不加關稅，可能不會除去最惠國待遇。不過無論論如何，中國對美國在戰略上處在上風，尤其侵侵的外交失誤上，我個人認為三年內美國將完全失去對中國的優勢，台灣十分危險！
美國根本沒有對付中國的意志，漸漸邁向失敗之路。侵侵更加速這種狀態，唯一的好消息是中國經濟即使有外銷也不是一時三刻可以回復，給美國數年的時間，這是最後的希望！
自嗨吧！
好自為之翻譯即係 Hao Zi Wei Zhi 囉。
😂😂😂😂
我反對無能力既人當兵
我反對無能力嘅人升職
我支持以能力入職入學
🤗🤗🤗🤗🤗🤗🤗🤗🤗🤗
奇就奇在卤水鸡比唔cup佢線囉😅
好易翻譯啫 ： “ 我增鬼死你呀 ！ I hate you ! "
楚漢時代的項羽聞名於世當今的政治家只有王毅能予以相比
其實2月份果 10% 關稅吾加能令到中國同意出售 TikTok 系值得嘅，因為參眾2院如果通過取消中國最優惠遇國，系果時在笫一段階加中國 30% 關稅都可以，中國應該吾會糸最低關稅國家的
咁習夾平小學雞,佢都可以又經濟,又科技,又國防,樣樣都係自己親自部署,親自指揮哪！美國都係有樣學樣啫,中共唔係已經向世界指明方向嗎？
因為trump要中國幫佢重建美國既工業體系，所以短期唔會逼得中國咁盡
氹2百斤賣完TikTok , 過幾個月再加7佢關稅＇放蛇咬呀？
好自为之 better watch your back 😂
好自為之，即係等於說我放長雙眼睇你點撞板。
When you mentioned about Tikto as just reminds me about another interesting news today also . China has develop a New AI which known as “ Deepseek”. When the news was being announced earlier on today , I heard the US stock market may get fluctuated. I look forward to hearing the comments from Trump in due course .
I also has a query . How this technology can be achieved by China if country like US has been imposing sanctions to China with all restriction of high technology chips to export from the last 2 years since ? Sounds to me like US is just tasting with their own medicine after all . What a surprise with the invention of “Deepseek” AI.
總之侵侵講咩做咩都一定要讚好 同意 通過
否則就是左膠 回塘 冇critical thinking
我小學老師都係咁譯好自為之的英文, 佢真勁.
今日美股大跌！重災區是AI股，英偉達及台積電，蕭生你點睇
Trump想tiktok賣俾佢老友, 佢就可以問佢收回佣1000億.
唔因 Tiktok 單野而加關稅同唔加中國關稅係兩回事,所以就算 Tiktok 賣左俾美國都好,侵侵一樣可以以第二種理由加中國關稅
😂😂😂😂😂😂😂😂😂😂😂😂😂😂
個人覺得特朗普想拎到tiktok, 係想借tiktok 令本來中國用的手法由他用, 即是由他主導民情, 加上由共和黨之前提出總統任期問題, 他可能好想變做普丁
No play/Tik Tok..//// my fr no play only young trump... said only young..watch>> .
Buy US stocks now?
如果以佢嘅語景嘅話, 最接近嘅應該係Mind your own business.
Donald Trump 以私交和私人利益來用人, 唔理經驗資歷, 唔止呢個電視人 Pete Hegseth, 仲有 Robert Kennedy Jr. 就只因為他投誠, 這班人對任職單位毫無經驗, 這是獨裁者的行為
同意
普通话嘅反贼通常都系中南海听床师，广东话嘅反贼就变身常委听床师咩？
王毅又点会唔识英文啊！网上都可以容易搜到巨讲英文。
唉！對呀Trump攪嘅亂局，真係無話可說。我哋放長雙眼睇啦！
好自為之=You should watch out for your surroundings.
有冇咁把炮
Donald Trump 為私人及圈內人利益, 不加中國關稅, 侵粉一定又讚好
可以講下DeepSeek的睇法嗎？