{"id":5634,"date":"2025-02-14T09:42:40","date_gmt":"2025-02-14T02:42:40","guid":{"rendered":"https:\/\/wincolaw.com.vn\/?p=5634"},"modified":"2025-12-19T09:45:17","modified_gmt":"2025-12-19T02:45:17","slug":"deepseek-thach-thuc-cac-ong-lon-bang-ai-gia-re","status":"publish","type":"post","link":"https:\/\/wincolaw.com.vn\/vi\/deepseek-thach-thuc-cac-ong-lon-bang-ai-gia-re.html","title":{"rendered":"DeepSeek th\u00e1ch th\u1ee9c c\u00e1c \u00f4ng l\u1edbn b\u1eb1ng AI gi\u00e1 r\u1ebb"},"content":{"rendered":"

Ng\u00e0y 20-1-2025, c\u00f4ng ty kh\u1edfi nghi\u1ec7p v\u1ec1 tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o (Artificial Intelligence – AI) DeepSeek c\u1ee7a Trung Qu\u1ed1c \u0111\u00e3 cho ra m\u1eaft m\u00f4 h\u00ecnh AI m\u1edbi mang t\u00ean R1 v\u00e0 ngay l\u1eadp t\u1ee9c l\u00e0m chao \u0111\u1ea3o c\u1ea3 th\u1ebf gi\u1edbi v\u00e0 th\u1ecb tr\u01b0\u1eddng ch\u1ee9ng kho\u00e1n c\u00f4ng ngh\u1ec7. C\u00e2u chuy\u1ec7n c\u1ee7a DeepSeek kh\u00f4ng ch\u1ec9 d\u1eebng l\u1ea1i trong l\u0129nh v\u1ef1c c\u00f4ng ngh\u1ec7 m\u00e0 c\u00f2n g\u1ee3i m\u1edf nhi\u1ec1u \u0111i\u1ec1u li\u00ean quan \u0111\u1ebfn qu\u1ea3n l\u00fd kinh doanh, s\u1eed d\u1ee5ng t\u00e0i s\u1ea3n tr\u00ed tu\u1ec7 v\u00e0 ch\u00ednh s\u00e1ch.<\/p>\n

\"\"Kh\u00e1c bi\u1ec7t c\u00f4ng ngh\u1ec7<\/h3>\n

\u0110i\u1ec3m kh\u00e1c bi\u1ec7t \u0111\u1ea7u ti\u00ean trong c\u00e1ch DeepSeek ti\u1ebfp c\u1eadn c\u00f4ng ngh\u1ec7 ph\u1ea3i k\u1ec3 \u0111\u1ebfn ph\u01b0\u01a1ng ph\u00e1p c\u00f4ng ty n\u00e0y \u00e1p d\u1ee5ng v\u1edbi c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn (Large Language Models – LLMs). C\u00f3 th\u1ec3 hi\u1ec3u LLM l\u00e0 m\u1ed9t lo\u1ea1i AI \u0111\u01b0\u1ee3c hu\u1ea5n luy\u1ec7n tr\u00ean kh\u1ed1i l\u01b0\u1ee3ng d\u1eef li\u1ec7u v\u0103n b\u1ea3n kh\u1ed5ng l\u1ed3, gi\u00fap n\u00f3 c\u00f3 th\u1ec3 hi\u1ec3u, t\u1ea1o v\u00e0 x\u1eed l\u00fd ng\u00f4n ng\u1eef t\u1ef1 nhi\u00ean nh\u01b0 con ng\u01b0\u1eddi. Ch\u00ednh ph\u1ee7 M\u1ef9 t\u1eeb v\u00e0i n\u0103m nay \u0111\u00e3 \u00e1p d\u1ee5ng c\u00e1c quy \u0111\u1ecbnh h\u1ea1n ch\u1ebf ngu\u1ed3n cung c\u1ea5p chip AI hi\u1ec7u su\u1ea5t cao cho c\u00e1c c\u00f4ng ty Trung Qu\u1ed1c. V\u00ec v\u1eady, h\u1ea7u h\u1ebft c\u00e1c LLM c\u1ee7a Trung Qu\u1ed1c \u0111\u1ec1u d\u1ef1a v\u00e0o vi\u1ec7c hu\u1ea5n luy\u1ec7n ph\u00e2n t\u00e1n tr\u00ean nhi\u1ec1u GPU (b\u1ed9 x\u1eed l\u00fd chuy\u00ean d\u1ee5ng gi\u00fap t\u0103ng t\u1ed1c c\u00e1c t\u00e1c v\u1ee5 t\u00ednh to\u00e1n ph\u1ee9c t\u1ea1p trong AI) y\u1ebfu h\u01a1n. Tuy nhi\u00ean, nh\u1edd t\u1ed1i \u01b0u ki\u1ebfn tr\u00fac, ch\u00fang v\u1eabn \u0111\u1ea1t \u0111\u01b0\u1ee3c hi\u1ec7u su\u1ea5t c\u1ea1nh tranh, d\u00f9 ch\u01b0a th\u1ec3 s\u00e1nh ngang v\u1edbi c\u00f4ng ngh\u1ec7 ti\u00ean phong.<\/p>\n

DeepSeek \u1ee9ng d\u1ee5ng ki\u1ebfn tr\u00fac Multi – Head Latent Attention (MLA) – c\u01a1 ch\u1ebf ph\u00e2n b\u1ed5 tr\u1ecdng s\u1ed1 linh ho\u1ea1t gi\u00fap t\u1ed1i \u01b0u b\u1ed9 nh\u1edb v\u00e0 Mixture of Experts (MOE) – m\u00f4 h\u00ecnh k\u1ebft h\u1ee3p nhi\u1ec1u chuy\u00ean gia x\u1eed l\u00fd chuy\u00ean bi\u1ec7t, t\u1eeb \u0111\u00f3 c\u1ea3i thi\u1ec7n hi\u1ec7u su\u1ea5t t\u00ednh to\u00e1n. V\u00ec th\u1ebf DeepSeek gi\u1ea3m \u0111\u01b0\u1ee3c m\u1ee9c s\u1eed d\u1ee5ng b\u1ed9 nh\u1edb v\u00e0 cho ph\u00e9p t\u1eadn d\u1ee5ng t\u00e0i nguy\u00ean m\u1ed9t c\u00e1ch hi\u1ec7u qu\u1ea3 h\u01a1n. S\u1eed d\u1ee5ng ph\u01b0\u01a1ng ph\u00e1p \u0111\u00e0o t\u1ea1o tr\u00ean c\u00e1c GPU hi\u1ec7u su\u1ea5t th\u1ea5p thay v\u00ec c\u00e1c c\u1ee5m Nvidia ti\u00ean ti\u1ebfn – \u0111i\u1ec1u \u0111\u00f2i h\u1ecfi ngu\u1ed3n v\u1ed1n \u0111\u1ea7u t\u01b0 kh\u1ed5ng l\u1ed3 – c\u0169ng gi\u00fap DeepSeek gi\u1ea3m thi\u1ec3u \u0111\u01b0\u1ee3c chi ph\u00ed x\u00e2y d\u1ef1ng v\u00e0 v\u1eadn h\u00e0nh.<\/p>\n

Ngo\u00e0i ra, vi\u1ec7c \u00e1p d\u1ee5ng c\u00e1c m\u00e3 ngu\u1ed3n m\u1edf c\u0169ng \u0111\u00f3ng vai tr\u00f2 quan tr\u1ecdng trong s\u1ef1 ph\u00e1t tri\u1ec3n c\u00e1c m\u00f4 h\u00ecnh LLM c\u1ee7a DeepSeek. M\u00f4 h\u00ecnh c\u1ee7a DeepSeek \u0111\u01b0\u1ee3c ph\u00e1t h\u00e0nh theo gi\u1ea5y ph\u00e9p s\u1eed d\u1ee5ng m\u00e3 ngu\u1ed3n m\u1edf (open – source license) c\u1ee7a H\u1ecdc vi\u1ec7n C\u00f4ng ngh\u1ec7 Massachusetts – MIT (M\u1ef9).<\/p>\n

L\u1ee3i th\u1ebf c\u1ee7a ki\u1ebfn tr\u00fac hi\u1ec7u qu\u1ea3 v\u00e0 c\u00e1ch ti\u1ebfp c\u1eadn m\u00e3 ngu\u1ed3n m\u1edf th\u1ec3 hi\u1ec7n r\u00f5 nh\u1ea5t khi so s\u00e1nh chi ph\u00ed hu\u1ea5n luy\u1ec7n m\u00f4 h\u00ecnh: DeepSeek c\u00f4ng b\u1ed1 ch\u1ec9 t\u1ed1n kho\u1ea3ng 5,6 tri\u1ec7u \u0111\u00f4 la M\u1ef9 cho m\u00f4 h\u00ecnh R1, trong khi c\u00e1c c\u00f4ng ty M\u1ef9 nh\u01b0 OpenAI v\u00e0 Alphabet \u0111\u01b0\u1ee3c cho l\u00e0 ph\u1ea3i b\u1ecf ra t\u1eeb 40-200 tri\u1ec7u \u0111\u00f4 la \u0111\u1ec3 x\u00e2y d\u1ef1ng c\u00e1c m\u00f4 h\u00ecnh LLM.<\/p>\n

C\u00e1c ch\u1ec9 s\u1ed1 \u0111\u00e1nh gi\u00e1 m\u1ee9c \u0111\u1ed9 hi\u1ec7u qu\u1ea3 c\u1ee7a LLM c\u00f2n cho th\u1ea5y DeepSeek c\u00f3 kh\u1ea3 n\u0103ng t\u01b0 duy v\u00e0 \u0111\u01b0a ra \u0111\u00e1p s\u1ed1 nhanh, c\u00f3 \u0111\u1ed9 ch\u00ednh x\u00e1c cao trong l\u1eadp tr\u00ecnh. Ngo\u00e0i ra, m\u00f4 h\u00ecnh n\u00e0y c\u00f2n th\u1ec3 hi\u1ec7n s\u1ef1 v\u01b0\u1ee3t tr\u1ed9i \u1edf kh\u1ea3 n\u0103ng gi\u1ea3i to\u00e1n v\u00e0 \u0111\u01b0a ra c\u00e2u tr\u1ea3 l\u1eddi cho c\u00e2u h\u1ecfi li\u00ean quan \u0111\u1ebfn Trung Qu\u1ed1c (ngo\u1ea1i tr\u1eeb m\u1ed9t s\u1ed1 c\u00e2u h\u1ecfi nh\u1ea1y c\u1ea3m v\u1ec1 m\u1eb7t ch\u00ednh tr\u1ecb) \u0111\u01b0\u1ee3c cho l\u00e0 ch\u00ednh x\u00e1c h\u01a1n nhi\u1ec1u m\u00f4 h\u00ecnh ti\u00ean phong nh\u01b0 GPT-4o c\u1ee7a OpenAI hay Claude-3.5 c\u1ee7a Anthropic.<\/p>\n

S\u00e1ng t\u1ea1o \u0111\u1ed9t ph\u00e1<\/h3>\n

T\u1eeb nh\u1eefng kh\u00e1c bi\u1ec7t trong c\u00e1ch ti\u1ebfp c\u1eadn c\u00f4ng ngh\u1ec7 v\u00e0 c\u00e1c m\u00f4 h\u00ecnh LLM, kh\u00f4ng kh\u00f3 hi\u1ec3u khi DeepSeek \u0111\u00e3 g\u00e2y b\u1ea5t ng\u1edd cho c\u00e1c c\u00f4ng ty AI t\u00ean tu\u1ed5i c\u1ee7a M\u1ef9. Tuy nhi\u00ean, n\u1ebfu quan s\u00e1t d\u01b0\u1edbi g\u00f3c nh\u00ecn c\u1ee7a l\u00fd thuy\u1ebft qu\u1ea3n tr\u1ecb, c\u1ee5 th\u1ec3 l\u00e0 l\u00fd thuy\u1ebft \u201c\u0111\u1ed5i m\u1edbi \u0111\u1ed9t ph\u00e1\u201d (disruptive innovation) \u0111\u01b0\u1ee3c kh\u1edfi x\u01b0\u1edbng b\u1edfi Gi\u00e1o s\u01b0 Clayton Christensen c\u1ee7a tr\u01b0\u1eddng Kinh doanh Harvard v\u00e0o n\u0103m 1995, s\u1ef1 xu\u1ea5t hi\u1ec7n c\u1ee7a DeepSeek l\u00e0 c\u00f3 th\u1ec3 d\u1ef1 b\u00e1o \u0111\u01b0\u1ee3c.<\/p>\n

C\u00f3 th\u1ec3 th\u1ea5y AI s\u1ebd kh\u00f4ng d\u1eebng l\u1ea1i \u1edf \u0111\u00e2y v\u00e0 c\u00f3 th\u1ec3 n\u0103m 2025 s\u1ebd ch\u1ee9ng ki\u1ebfn nhi\u1ec1u \u201ckho\u1ea3nh kh\u1eafc DeepSeek\u201d h\u01a1n n\u1eefa, b\u00e1o hi\u1ec7u m\u1ed9t s\u1ef1 chuy\u1ec3n d\u1ecbch m\u1ea1nh m\u1ebd trong n\u1ec1n kinh t\u1ebf v\u00e0 c\u00f4ng ngh\u1ec7 to\u00e0n c\u1ea7u.<\/p><\/blockquote>\n

Theo l\u00fd thuy\u1ebft n\u00e0y, \u0111\u1ed5i m\u1edbi \u0111\u1ed9t ph\u00e1 l\u00e0 qu\u00e1 tr\u00ecnh m\u1ed9t c\u00f4ng ty nh\u1ecf v\u1edbi ngu\u1ed3n l\u1ef1c \u00edt h\u01a1n t\u1eebng b\u01b0\u1edbc th\u00e0nh c\u00f4ng trong vi\u1ec7c th\u00e1ch th\u1ee9c c\u00e1c c\u00f4ng ty l\u1edbn \u0111\u00e3 th\u00e0nh danh. C\u1ee5 th\u1ec3, khi nh\u1eefng c\u00f4ng ty l\u1edbn t\u1eadp trung ngu\u1ed3n l\u1ef1c \u0111\u1ec3 c\u1ea3i thi\u1ec7n s\u1ea3n ph\u1ea9m v\u00e0 d\u1ecbch v\u1ee5 c\u1ee7a m\u00ecnh nh\u1eb1m ph\u1ee5c v\u1ee5 cho nh\u1eefng ph\u00e2n kh\u00fac kh\u00e1ch h\u00e0ng kh\u00f3 t\u00ednh nh\u1ea5t (th\u01b0\u1eddng l\u00e0 kh\u00e1ch h\u00e0ng mang l\u1ea1i l\u1ee3i nhu\u1eadn cao nh\u1ea5t), h\u1ecd s\u1ebd b\u1ecf qua nhu c\u1ea7u c\u1ee7a nh\u1eefng ph\u00e2n kh\u00fac kh\u00e1ch h\u00e0ng kh\u00e1c.<\/p>\n

Khi \u0111\u00f3, nh\u1eefng c\u00f4ng ty nh\u1ecf m\u1edbi th\u00e0nh l\u1eadp, s\u1ebd t\u1ea1o \u0111\u01b0\u1ee3c \u201c\u0111\u1ed9t ph\u00e1\u201d khi nh\u1eafm v\u00e0o vi\u1ec7c ph\u1ee5c v\u1ee5 nh\u1eefng ph\u00e2n kh\u00fac kh\u00e1ch h\u00e0ng b\u1ecb b\u1ecf qua \u0111\u00f3. C\u00e1c c\u00f4ng ty n\u00e0y cung c\u1ea5p c\u00e1c gi\u1ea3i ph\u00e1p thay th\u1ebf v\u1edbi chi ph\u00ed th\u1ea5p, t\u1eebng b\u01b0\u1edbc c\u1ea3i thi\u1ec7n ch\u1ea5t l\u01b0\u1ee3ng v\u00e0 cu\u1ed1i c\u00f9ng chi\u1ebfm l\u0129nh th\u1ecb ph\u1ea7n c\u1ee7a c\u00e1c c\u00f4ng ty kia.<\/p>\n

\u0110\u00e2y c\u00f3 th\u1ec3 l\u00e0 h\u01b0\u1edbng \u0111i m\u00e0 DeepSeek \u0111\u00e3 l\u1ef1a ch\u1ecdn. V\u1ecb th\u1ebf c\u1ee7a DeepSeek \u0111\u00e3 ph\u1ea7n n\u00e0o \u0111\u01b0\u1ee3c kh\u1eb3ng \u0111\u1ecbnh khi c\u00f4ng ty n\u00e0y c\u00f3 th\u1ec3 c\u1ea1nh tranh v\u1edbi c\u00e1c t\u1eadp \u0111o\u00e0n c\u00f4ng ngh\u1ec7 h\u00e0ng \u0111\u1ea7u t\u1ea1i M\u1ef9, b\u1ea5t ch\u1ea5p m\u1ee9c chi ph\u00ed v\u1eadn h\u00e0nh v\u00e0 \u0111\u1ea7u t\u01b0 ban \u0111\u1ea7u th\u1ea5p h\u01a1n r\u1ea5t nhi\u1ec1u.<\/p>\n

C\u0169ng theo l\u00fd thuy\u1ebft n\u00e0y th\u00ec m\u00f4 h\u00ecnh AI c\u1ee7a DeepSeek v\u1eabn ch\u01b0a ph\u1ea3i l\u00e0 cu\u1ed1i c\u00f9ng. C\u00f3 th\u1ec3 trong kho\u1ea3ng th\u1eddi gian ng\u1eafn nh\u01b0 v\u00e0i th\u00e1ng t\u1edbi, s\u1ebd xu\u1ea5t hi\u1ec7n nh\u1eefng \u0111\u1ed1i th\u1ee7 m\u1edbi c\u1ea1nh tranh tr\u1ef1c ti\u1ebfp v\u1edbi DeepSeek hay s\u1ebd xu\u1ea5t hi\u1ec7n c\u00e1c c\u00f4ng ty AI s\u00e1ng t\u1ea1o tr\u00ean m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef nh\u1ecf (Small Language Models – SLM) c\u1ea1nh tranh tr\u1ef1c ti\u1ebfp v\u1edbi c\u00e1c c\u00f4ng ty khai th\u00e1c m\u00f4 h\u00ecnh LLM. V\u00e0 \u0111\u00e2y c\u0169ng kh\u00f4ng ph\u1ea3i l\u00e0 \u0111i\u1ec1u b\u1ea5t ng\u1edd trong s\u1ef1 ph\u00e1t tri\u1ec3n m\u1ea1nh m\u1ebd c\u1ee7a c\u00e1c c\u00f4ng ngh\u1ec7 AI hi\u1ec7n nay.<\/p>\n

Qu\u1ea3n l\u00fd t\u00e0i s\u1ea3n tr\u00ed tu\u1ec7 theo m\u00f4 h\u00ecnh s\u00e1ng t\u1ea1o m\u1edf<\/h3>\n

\u0110\u1ec3 ph\u00e1t tri\u1ec3n c\u00e1c m\u00f4 h\u00ecnh LLM \u0111\u1ee7 s\u1ee9c th\u00e1ch th\u1ee9c c\u00e1c \u00f4ng l\u1edbn AI tr\u00ean th\u1ebf gi\u1edbi, DeepSeek \u0111\u00e3 v\u1eadn d\u1ee5ng m\u00f4 h\u00ecnh \u201cs\u00e1ng t\u1ea1o m\u1edf\u201d (open innovation). S\u00e1ng t\u1ea1o m\u1edf l\u00e0 l\u00fd thuy\u1ebft do Gi\u00e1o s\u01b0 Henry Chesbrough, \u0110\u1ea1i h\u1ecdc California, Berkeley \u0111\u1ec1 xu\u1ea5t l\u1ea7n \u0111\u1ea7u ti\u00ean v\u00e0o n\u0103m 2003. Theo \u0111\u00f3, khi c\u00e1c c\u00f4ng ty t\u00ecm c\u00e1ch ph\u00e1t tri\u1ec3n c\u00f4ng ngh\u1ec7, h\u1ecd c\u00f3 th\u1ec3 v\u00e0 n\u00ean s\u1eed d\u1ee5ng c\u1ea3 \u00fd t\u01b0\u1edfng n\u1ed9i b\u1ed9 v\u00e0 \u00fd t\u01b0\u1edfng b\u00ean ngo\u00e0i.<\/p>\n

Liang Wenfeng (L\u01b0\u01a1ng V\u0103n Phong), \u00f4ng ch\u1ee7 c\u1ee7a DeepSeek, l\u00e0 ng\u01b0\u1eddi s\u00e1ng l\u1eadp v\u00e0 \u0111i\u1ec1u h\u00e0nh qu\u1ef9 \u0111\u1ea7u c\u01a1 High-Flyer. T\u1ea1i \u0111\u00e2y, \u00f4ng g\u00e2y d\u1ef1ng t\u00e0i s\u1ea3n b\u1eb1ng c\u00e1ch s\u1eed d\u1ee5ng AI v\u00e0 thu\u1eadt to\u00e1n \u0111\u1ec3 nh\u1eadn di\u1ec7n c\u00e1c m\u00f4 h\u00ecnh c\u00f3 th\u1ec3 t\u00e1c \u0111\u1ed9ng \u0111\u1ebfn gi\u00e1 c\u1ed5 phi\u1ebfu. \u0110\u1ed9i ng\u0169 c\u1ee7a \u00f4ng t\u1eeb l\u00e2u \u0111\u00e3 th\u00e0nh th\u1ea1o trong vi\u1ec7c s\u1eed d\u1ee5ng chip Nvidia \u0111\u1ec3 ki\u1ebfm l\u1ee3i nhu\u1eadn t\u1eeb giao d\u1ecbch ch\u1ee9ng kho\u00e1n. N\u0103m 2023, khi th\u00e0nh l\u1eadp DeepSeek, v\u1ecb thuy\u1ec1n tr\u01b0\u1edfng n\u00e0y \u0111\u01b0a nh\u1eefng ng\u01b0\u1eddi gi\u1ecfi nh\u1ea5t t\u1eeb qu\u1ef9 \u0111\u1ea7u c\u01a1 – l\u00e0 nh\u1eefng ng\u01b0\u1eddi th\u1ef1c s\u1ef1 hi\u1ec3u r\u00f5 c\u00e1ch c\u00e1c con chip ho\u1ea1t \u0111\u1ed9ng, sang c\u00f4ng ty m\u1edbi. Khi M\u1ef9 c\u1ea5m Nvidia xu\u1ea5t kh\u1ea9u c\u00e1c con chip m\u1ea1nh nh\u1ea5t sang Trung Qu\u1ed1c, DeepSeek bu\u1ed9c ph\u1ea3i t\u00ecm c\u00e1ch s\u00e1ng t\u1ea1o \u0111\u1ec3 t\u1ed1i \u01b0u h\u00f3a s\u1ee9c m\u1ea1nh t\u00ednh to\u00e1n c\u1ee7a s\u1ed1 l\u01b0\u1ee3ng chip n\u1ed9i \u0111\u1ecba c\u00f3 h\u1ea1n.<\/p>\n

\u0110\u00e2y l\u00e0 v\u1ea5n \u0111\u1ec1 m\u00e0 \u0111\u1ed9i ng\u0169 c\u1ee7a DeepSeek v\u1ed1n \u0111\u00e3 bi\u1ebft c\u00e1ch gi\u1ea3i quy\u1ebft: h\u1ecd \u0111\u00e3 c\u00f3 th\u1ec3 khai th\u00e1c t\u1ed1i \u0111a ti\u1ec1m n\u0103ng c\u1ee7a nh\u1eefng GPU k\u00e9m ti\u00ean ti\u1ebfn h\u01a1n. \u00d4ng ch\u1ee7 DeepSeek \u0111\u1ecbnh h\u00ecnh DeepSeek l\u00e0 m\u1ed9t c\u00f4ng ty thu\u1ea7n n\u1ed9i \u0111\u1ecba, khi tuy\u1ec3n d\u1ee5ng nh\u00e2n s\u1ef1 t\u1eeb nh\u1eefng tr\u01b0\u1eddng \u0111\u1ea1i h\u1ecdc h\u00e0ng \u0111\u1ea7u Trung Qu\u1ed1c thay v\u00ec c\u00e1c chuy\u00ean gia t\u1eeb M\u1ef9 hay ch\u00e2u \u00c2u. Ch\u00ednh s\u1ef1 k\u1ebft h\u1ee3p b\u00e0i b\u1ea3n gi\u1eefa \u00fd t\u01b0\u1edfng c\u1ee7a \u0111\u1ed9i ng\u0169 nh\u00e2n s\u1ef1 b\u1ea3n \u0111\u1ecba c\u00f9ng v\u1edbi ngu\u1ed3n l\u1ef1c b\u00ean ngo\u00e0i nh\u01b0 chip Nvidia v\u00e0 m\u00e3 ngu\u1ed3n m\u1edf c\u1ee7a MIT \u0111\u00e3 g\u00f3p ph\u1ea7n t\u1ea1o n\u00ean th\u00e0nh c\u00f4ng c\u1ee7a DeepSeek.<\/p>\n

\u1ede h\u01b0\u1edbng ng\u01b0\u1ee3c l\u1ea1i, DeepSeek c\u0169ng tham gia g\u00e2y d\u1ef1ng h\u1ec7 sinh th\u00e1i m\u00e3 ngu\u1ed3n m\u1edf. Gi\u1ed1ng nh\u01b0 Alibaba, DeepSeek \u0111i theo h\u01b0\u1edbng c\u1ee7a Meta – c\u00f4ng ty m\u1eb9 c\u1ee7a Facebook: cung c\u1ea5p \u1ee9ng d\u1ee5ng theo ph\u01b0\u01a1ng ph\u00e1p c\u1ea5p ph\u00e9p m\u00e3 ngu\u1ed3n m\u1edf. Ch\u00ednh s\u00e1ch n\u00e0y c\u0169ng th\u00fac \u0111\u1ea9y y\u1ebfu t\u1ed1 minh b\u1ea1ch th\u00f4ng tin: khi ra m\u1eaft m\u00f4 h\u00ecnh m\u1edbi, DeepSeek \u0111\u00e3 c\u00f4ng b\u1ed1 k\u00e8m theo m\u1ed9t t\u00e0i li\u1ec7u gi\u1ea3i th\u00edch chi ti\u1ebft c\u00e1ch th\u1ee9c x\u00e2y d\u1ef1ng m\u1ed9t LLM v\u1edbi ng\u00e2n s\u00e1ch h\u1ea1n ch\u1ebf, c\u00f3 th\u1ec3 t\u1ef1 \u0111\u1ed9ng h\u1ecdc h\u1ecfi v\u00e0 c\u1ea3i thi\u1ec7n m\u00e0 kh\u00f4ng c\u1ea7n s\u1ef1 gi\u00e1m s\u00e1t c\u1ee7a con ng\u01b0\u1eddi.<\/p>\n

G\u00f3c nh\u00ecn ch\u00ednh s\u00e1ch<\/p>\n

T\u1eeb l\u00e2u, Trung Qu\u1ed1c \u0111\u00e3 theo \u0111u\u1ed5i chi\u1ebfn l\u01b0\u1ee3c \u0111\u1ed5i m\u1edbi s\u00e1ng t\u1ea1o d\u1ef1a c\u1ea3 tr\u00ean vi\u1ec7c ti\u1ebfp thu ngu\u1ed3n c\u00f4ng ngh\u1ec7 n\u01b0\u1edbc ngo\u00e0i l\u1eabn s\u00e1ng t\u1ea1o b\u1ea3n \u0111\u1ecba. Ph\u00e1t huy s\u1ee9c m\u1ea1nh n\u1ed9i l\u1ef1c \u0111\u01b0\u1ee3c cho l\u00e0 ch\u00ednh s\u00e1ch \u01b0u ti\u00ean trong b\u1ed1i c\u1ea3nh M\u1ef9 li\u00ean ti\u1ebfp c\u00f3 c\u00e1c \u0111\u1ed9ng th\u00e1i h\u1ea1n ch\u1ebf ngu\u1ed3n cung thi\u1ebft b\u1ecb b\u00e1n d\u1eabn cho Trung Qu\u1ed1c v\u00e0 \u0111\u01b0\u1ee3c cho l\u00e0 v\u1edbi m\u1ee5c \u0111\u00edch k\u00ecm h\u00e3m n\u0103ng l\u1ef1c c\u00f4ng ngh\u1ec7 c\u1ee7a Trung Qu\u1ed1c n\u00f3i chung v\u00e0 trong l\u0129nh v\u1ef1c AI n\u00f3i ri\u00eang.<\/p>\n

Tuy nhi\u00ean, vi\u1ec7c M\u1ef9 si\u1ebft ch\u1eb7t c\u00e1c h\u1ea1n ch\u1ebf \u0111\u1ed1i v\u1edbi c\u00f4ng ngh\u1ec7 AI v\u00e0 thi\u1ebft b\u1ecb b\u00e1n d\u1eabn d\u01b0\u1eddng nh\u01b0 \u0111\u00e3 t\u1ea1o \u0111\u1ed9ng l\u1ef1c \u0111\u1ec3 c\u00e1c doanh nghi\u1ec7p Trung Qu\u1ed1c s\u00e1ng t\u1ea1o h\u01a1n. C\u00e1c c\u00f4ng ty Trung Qu\u1ed1c t\u00ecm c\u00e1ch t\u1ed1i \u01b0u h\u00f3a thu\u1eadt to\u00e1n thay v\u00ec ph\u1ee5 thu\u1ed9c v\u00e0o ph\u1ea7n c\u1ee9ng. DeepSeek \u0111\u00e3 hu\u1ea5n luy\u1ec7n th\u00e0nh c\u00f4ng m\u00f4 h\u00ecnh ch\u1ec9 v\u1edbi s\u1ed1 l\u01b0\u1ee3ng chip Nvidia chuy\u00ean d\u1ee5ng th\u1ea5p h\u01a1n r\u1ea5t nhi\u1ec1u so v\u1edbi s\u1ed1 l\u01b0\u1ee3ng chip t\u1ed1i t\u00e2n m\u00e0 c\u00e1c c\u00f4ng ty M\u1ef9 c\u1ea7n \u0111\u1ec3 hu\u1ea5n luy\u1ec7n c\u00e1c m\u00f4 h\u00ecnh c\u1ee7a m\u00ecnh.<\/p>\n

C\u00f3 v\u1ebb DeepSeek \u0111ang g\u00f3p ph\u1ea7n th\u1ec3 hi\u1ec7n s\u1ee9c m\u1ea1nh n\u1ed9i t\u1ea1i c\u1ee7a c\u00e1c c\u00f4ng ty c\u00f4ng ngh\u1ec7 Trung Qu\u1ed1c v\u00e0 n\u1ed7 l\u1ef1c tr\u1edf th\u00e0nh ni\u1ec1m t\u1ef1 h\u00e0o v\u1ec1 m\u1eb7t c\u00f4ng ngh\u1ec7 c\u1ee7a Trung Qu\u1ed1c tr\u00ean ph\u1ea1m vi to\u00e0n c\u1ea7u – v\u1edbi s\u1ed1 l\u01b0\u1ee3ng c\u00f2n khi\u00eam t\u1ed1n m\u00e0 th\u00e0nh c\u00f4ng nh\u1ea5t l\u00e0 Huawei.<\/p>\n

DeepSeek, d\u00f9 xu\u1ea5t hi\u1ec7n nh\u01b0 m\u1ed9t hi\u1ec7n t\u01b0\u1ee3ng, v\u1eabn kh\u00f4ng ph\u1ea3i l\u00e0 b\u01b0\u1edbc ngo\u1eb7t thay \u0111\u1ed5i t\u1ea5t c\u1ea3. \u0110\u00e2y ch\u1ec9 l\u00e0 d\u1ea5u hi\u1ec7u cho th\u1ea5y AI \u0111ang ti\u1ebfn \u0111\u1ebfn giai \u0111o\u1ea1n m\u1edbi – t\u1eeb nh\u1eadn di\u1ec7n sang t\u01b0 duy v\u00e0 gi\u1ea3i quy\u1ebft v\u1ea5n \u0111\u1ec1. S\u1ef1 c\u1ea1nh tranh gi\u1eefa h\u01b0\u1edbng ti\u1ebfp c\u1eadn m\u00e3 ngu\u1ed3n m\u1edf (nh\u01b0 DeepSeek \u0111ang tri\u1ec3n khai) v\u00e0 m\u00f4 h\u00ecnh \u0111\u1ed9c quy\u1ec1n (m\u00e0 m\u1ed9t s\u1ed1 t\u1eadp \u0111o\u00e0n AI c\u1ee7a M\u1ef9 v\u1edbi s\u1ef1 b\u1ea3o tr\u1ee3 c\u1ee7a c\u00e1c nh\u00e0 \u0111\u1ea7u t\u01b0 \u0111ang \u00e1p d\u1ee5ng) s\u1ebd ti\u1ebfp t\u1ee5c th\u00fac \u0111\u1ea9y s\u00e1ng t\u1ea1o, gi\u1ea3m chi ph\u00ed thi\u1ebft k\u1ebf v\u00e0 do \u0111\u00f3 h\u1ee9a h\u1eb9n s\u1ebd m\u1edf ra nhi\u1ec1u \u1ee9ng d\u1ee5ng th\u1ef1c ti\u1ec5n h\u01a1n. C\u00f3 th\u1ec3 th\u1ea5y AI s\u1ebd kh\u00f4ng d\u1eebng l\u1ea1i \u1edf \u0111\u00e2y v\u00e0 c\u00f3 th\u1ec3 n\u0103m 2025 s\u1ebd ch\u1ee9ng ki\u1ebfn nhi\u1ec1u \u201ckho\u1ea3nh kh\u1eafc DeepSeek\u201d h\u01a1n n\u1eefa, b\u00e1o hi\u1ec7u m\u1ed9t s\u1ef1 chuy\u1ec3n d\u1ecbch m\u1ea1nh m\u1ebd trong n\u1ec1n kinh t\u1ebf v\u00e0 c\u00f4ng ngh\u1ec7 to\u00e0n c\u1ea7u.<\/p>\n

(*) Tr\u01b0\u1eddng Qu\u1ea3n tr\u1ecb v\u00e0 Kinh doanh, \u0110\u1ea1i h\u1ecdc Qu\u1ed1c gia H\u00e0 N\u1ed9i<\/em>
\n(**) Edtech Agency<\/em>
\n(***) VNPT Technology<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

Ng\u00e0y 20-1-2025, c\u00f4ng ty kh\u1edfi nghi\u1ec7p v\u1ec1 tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o (Artificial Intelligence – AI) DeepSeek c\u1ee7a Trung Qu\u1ed1c \u0111\u00e3 cho ra m\u1eaft m\u00f4 h\u00ecnh AI m\u1edbi mang t\u00ean R1 v\u00e0 ngay l\u1eadp t\u1ee9c l\u00e0m chao \u0111\u1ea3o c\u1ea3 th\u1ebf gi\u1edbi v\u00e0 th\u1ecb tr\u01b0\u1eddng ch\u1ee9ng kho\u00e1n c\u00f4ng ngh\u1ec7. C\u00e2u chuy\u1ec7n c\u1ee7a DeepSeek kh\u00f4ng ch\u1ec9 d\u1eebng l\u1ea1i […]<\/p>\n","protected":false},"author":1,"featured_media":5635,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_links_to":"","_links_to_target":""},"categories":[101,9],"tags":[2239,2238,2240],"yst_prominent_words":[],"class_list":["post-5634","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-so-huu-tri-tue","category-tin-tuc","tag-ai","tag-deepseek","tag-tri-tue-nhan-tao"],"_links":{"self":[{"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/posts\/5634","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/comments?post=5634"}],"version-history":[{"count":1,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/posts\/5634\/revisions"}],"predecessor-version":[{"id":5639,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/posts\/5634\/revisions\/5639"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/media\/5635"}],"wp:attachment":[{"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/media?parent=5634"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/categories?post=5634"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/tags?post=5634"},{"taxonomy":"yst_prominent_words","embeddable":true,"href":"https:\/\/wincolaw.com.vn\/vi\/wp-json\/wp\/v2\/yst_prominent_words?post=5634"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}