Don't get left behind. Learn how to use AI in our day to day work. Enrol now (20% OFF): elearning.lk/course/online-sinhala-ai-fundamentals-and-productivity-program-class-in-sri-lanka-by-uditha-bandara-with-elearning.lk-b1
Deepseekla Knowledge distillation walata OpenAi models use kara kiyala directly mention karala na. eyala published karapu paper eke mention karala thiyenne Llama-3.1-8B, Llama-3.3-70B-Instruct and Qwen model series eka use kara kiyala. eka "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning" paper eke sharply mention karala thiyenwa benchmark ekka. Deepseek Reasoning model eka outstanding performance ekak denawa, ekata main reason eka thamai meyala LLMs use karana normal supervised learning approach eken beyond gihilla Reinforcement learning(RL) use karala thiyenwa training walata thawa e wagema CoT prompting model ekak widiyata focus karala thiyenawa specially advanced maths problem solve karanna lesi wenna.☺ superior performance eka denna main reason eka Knowledge distillation newe.
India walata system ekee Data back door ekak nathi unata acsees wenna. Athulee inna kenata pulwan back door ekata katath horen athlata enna acses eka denna💯💯
I feel that the whole interpretation is completely wrong. There’s no real innovation here, it’s just theft. It reminds me of The movie Italian Job, where one thief steals from another. Sam stole from the whole world, and now the Chinese have stolen from him. The tech world is celebrating because they’re getting everything for free, and China is being praised. Distillation isn’t a new concept in this industry. The U.S. has millions of talented people who could do the same thing, and they have the resources to do it. But the real problem is lawsuits. Right now, Sam is the biggest enemy in tech because of his monopoly and all the legal battles over stolen data. But in China, DeepSeek doesn’t have to worry about that. One thing to keep in mind: the enemy of your enemy isn’t always your friend.
"It doesn't matter if you're a Harvard professor or an underserved student in a developing nation, we all get access to the same answers. With AI that keeps getting better and better at answering all our questions, the marginal cost of research is rapidly approaching zero." - Aravind Srinivas
@@SanjayaElvitigala DeepSeek should totally train with raw data. Its founder Liang Wenfeng has a big name in China's finances since around 2018. he the brains behind high flyer a firm that used deep learning models for trading. back then, they built massive datasets and gpu clusters, so it makes sense that they’d use all that experience and tech for DeepSeek. After training, they’ve mostly relied on distillation fot refining knowledge from their previous models and maybe even from ChatGPT. But ChatGPT isn’t their only reference it’s just one of many sources.
Don't get left behind. Learn how to use AI in our day to day work. Enrol now (20% OFF): elearning.lk/course/online-sinhala-ai-fundamentals-and-productivity-program-class-in-sri-lanka-by-uditha-bandara-with-elearning.lk-b1
හෙළ වෙදකම Ai හරහා ඒක මාරම පොයින්ට් එකක් ❤
සුපිරි සංජය ඕවගේ තව දැනුම තියෙන අයත් එක්ක AI ගැන පොඩි පොඩ්කාස් ටිකක් කරන්න හරි වටිනවා දන්නේ නැති අයට දැනගන්න🎉
මේක මරු
උදිත බ්‍රෝගෙ පෙනුම, බෲනොගෙ මල්ලි කෙනෙක් වගේ😅
කිව්වා වගේම සිංහල නම් පට්ටම නිරවුල් deepseek වල😊
Eke thama Sinhalen mkuthma karaganna ba neda
@@sithumakash5756 chatgpt walata wada hodai.sinhala
DeepSeek ChatGPT වලින් distill කරා කියන එක සැකයක් විතරයි නේද? තාම තහවුරු කරලා නෑ නේද?
Issarahata enaa Quntam computers ekka Ai add unma loketa monawa weiida mee yana vidihata..?
Distillation එක පැහැදිලි කිරීම වැරදි.
Dan deepseek aith 2nd place ekata giya, chat gpt o3 mini model eka free use krna dela, it is better than deepseek R1.
Technically ගත්තොත් චීනේ හැමදාම කරපු වැඩේ ම තමයි නේද මේකෙත් වෙලා තියෙන්නේ? වෙන කොටසක් ලොකු සම්පත් සහා කාලයක් වැය කරලා ගොඩනඟපු මොඩල් එක base කරලා අළුත් එකක් develop කරලා..
ඕකටතමා කියන්නෙ market Competitors ලා ඕනි කියල 😂
මොනා උනත් අවසානෙ පැත්තක ඉන්න අපිට පට්ට ඉතින් 😅
Deepseekla Knowledge distillation walata OpenAi models use kara kiyala directly mention karala na. eyala published karapu paper eke mention karala thiyenne Llama-3.1-8B, Llama-3.3-70B-Instruct and Qwen model series eka use kara kiyala. eka "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning" paper eke sharply mention karala thiyenwa benchmark ekka.
Deepseek Reasoning model eka outstanding performance ekak denawa, ekata main reason eka thamai meyala LLMs use karana normal supervised learning approach eken beyond gihilla Reinforcement learning(RL) use karala thiyenwa training walata thawa e wagema CoT prompting model ekak widiyata focus karala thiyenawa specially advanced maths problem solve karanna lesi wenna.☺ superior performance eka denna main reason eka Knowledge distillation newe.
There is evidence that they used ChatGPT to train this model. In the beginning, DeepSeek itself admitted that it was trained by OpenAI.
We will plan another AI talk about the research paper. We are not discussing RL and model result's in this.
India walata system ekee Data back door ekak nathi unata acsees wenna. Athulee inna kenata pulwan back door ekata katath horen athlata enna acses eka denna💯💯
ඔව් ඉතින් එහෙම බැලුවොත් ලෝකේ කිසිම system එකක් secure නෑ
❤❤❤
Me sarwa contat kara ganna bareda
I feel that the whole interpretation is completely wrong. There’s no real innovation here, it’s just theft. It reminds me of The movie Italian Job, where one thief steals from another. Sam stole from the whole world, and now the Chinese have stolen from him. The tech world is celebrating because they’re getting everything for free, and China is being praised.
Distillation isn’t a new concept in this industry. The U.S. has millions of talented people who could do the same thing, and they have the resources to do it. But the real problem is lawsuits. Right now, Sam is the biggest enemy in tech because of his monopoly and all the legal battles over stolen data. But in China, DeepSeek doesn’t have to worry about that.
One thing to keep in mind: the enemy of your enemy isn’t always your friend.
Searching is not as fast as ChatGPT...
government එකෙන් security audit එකක් කොහොමත් කරයි මං හිතන්නේ digital ID system එකට. ඕකට කෑ ගහන කට්ටියගේ expertලා ඉන්නවනම් එයාලවත් එකතු කරගෙනම audit එක කරන්න තියෙන්නේ. vulnerability එකක් තියෙනවනම් එයාලටත් හොයා ගන්න පුළුවන්නේ 🙂 මේකට අපේ වයසේ තේරුමක් තියෙන මිනිස්සුම publicity එක නිසා මේකට කෑ ගහන එක නම් ටිකක් අවුල්.
"It doesn't matter if you're a Harvard professor
or an underserved student in a developing nation,
we all get access to the same answers.
With AI that keeps getting better and better
at answering all our questions,
the marginal cost of research is rapidly approaching zero."
- Aravind Srinivas
In this case, the cost of intelligence is dropping, maybe even to zero. This will have an impact on everyone in the world.
dude is wrong about distilling chat gpt
මෙහෙම දෙයක් වෙන්න පුළුවන් නේද Open Source උනාට එක ඉන්දියවා ඔඉත Develop කරලා නේ දෙන්නේ , එතකොට ඒ හරහා Back door එකක් තියෙන්නේ බැරිද ?
බයිලා තමා මම් මේ දෑන් deepseek එකේ search එකක් දෑම්ම. server busy කියලා තමා කිව්වේ
අපි මේ කතා කරන දේ ඔයාට වැටහිලා නෑ. කියල දෙන්න හිතෙන්නේත් නෑ atitude එක නිසා ☺️
DeepSeek Busy වෙන්න හේතුව නම් එකට මේ වෙනකොට Malware Attack ගොඩක් එනවා . ගොඩක් වෙලාවට ChatGPT පැත්තෙන් වෙන්න පුළුවන්,
@ ammo sorry
True, I also experienced the same.
ලංකාවේ Digital ID කියන්නේ ගොඨාගේ කාබනික පොහොර වගේ . කල යුතු දෙයක් , ක්‍රමාණුකුලව .😂😅
fivrr වගේ online වැඩකරන අයට tax ගහන එක ගැනත් දැනුවත් කරන්න.
DeepSeek එකට Internet ඕනේ නෑ. මම දැක්කා ඩයල් එකක් Raspberry Pi එකේ දුවනවා Wifi Disable කරලා.
ua-cam.com/users/shortsZN6XS2d_izI
Deepseek එකේ ලෝගෝ එහෙම හදන්න බෑ
Ml gana one
ඔය චීනුන්ගේ දේවල් විශ්වාස කරන්න හොඳ නෑ ඕකුන් deepseek එකත් එක්ක මොනවා එවලද දන්නෑ...
china hapana
Most of the facts are wrong here.😅
like?
@@SanjayaElvitigala DeepSeek should totally train with raw data. Its founder Liang Wenfeng has a big name in China's finances since around 2018. he the brains behind high flyer a firm that used deep learning models for trading. back then, they built massive datasets and gpu clusters, so it makes sense that they’d use all that experience and tech for DeepSeek. After training, they’ve mostly relied on distillation fot refining knowledge from their previous models and maybe even from ChatGPT. But ChatGPT isn’t their only reference it’s just one of many sources.