[{"data":1,"prerenderedAt":349},["ShallowReactive",2],{"tag-labels":3,"site-header":4,"article-llm-bakal-mentok-terus-gimana-dong-id":72},{},{"featuredPost":5,"navigations":29},{"id":6,"title":7,"title_en":8,"slug":9,"excerpt":10,"excerpt_en":11,"has_translation":12,"published_at":13,"reading_time":14,"hidden_from_listing":15,"tags":16,"thumbnail_image":21,"og_image":22,"series":23,"authors":28},110,"Ketika Pilih Framework Cuma Biar Kelihatan Beda","When You Pick a Framework Just to Look Different","ketika-pilih-framework-cuma-biar-kelihatan-beda","Terakhir kali kamu milih framework atau library baru, itu karena masalahmu butuh alat itu, atau karena alatnya kedengaran keren dan belum banyak yang pakai? Pertanyaan ini nggak enak dijawab. Soalnya keputusan teknologi sering diambil dengan alasan yang nggak pernah ditulis di dokumen teknis mana pun.","The last time you chose a new framework or library, was it because your problem needed that tool, or because it sounded cool and few people around you were using it? That question is uncomfortable. Technology decisions are often made for reasons that never appear in any technical document.",true,"2026-10-06T15:25:21.000Z",5,false,[17,18,19,20],"karier","framework","opini","pengalaman",null,"\u002Fog\u002Fketika-pilih-framework-cuma-biar-kelihatan-beda.png",{"id":24,"title":25,"title_en":26,"slug":27},9,"Artikel","Articles","all",[],[30,46,67],{"collection":31,"item":32},"series",{"__typename":31,"title":25,"title_en":26,"slug":27,"children_series":33},[34,38,42],{"title":35,"title_en":36,"slug":37},"Kecerdasan Artifisial","Artificial Intelligence","kecerdasan-artifisial",{"title":39,"title_en":40,"slug":41},"Konsep Dasar Pemrograman","Programming Fundamentals","konsep-dasar-pemrograman",{"title":43,"title_en":44,"slug":45},"Algoritma dan Struktur Data","Algorithms and Data Structures","algoritma-struktur-data",{"collection":31,"item":47},{"__typename":31,"title":48,"title_en":48,"slug":49,"children_series":50},"Tutorial","tutorial",[51,55,59,63],{"title":52,"title_en":53,"slug":54},"Belajar Python","Learning Python","belajar-python",{"title":56,"title_en":57,"slug":58},"Belajar Golang","Learning Go","belajar-golang",{"title":60,"title_en":61,"slug":62},"Belajar Java","Learning Java","belajar-java",{"title":64,"title_en":65,"slug":66},"Belajar Pascal","Learning Pascal","belajar-pascal",{"collection":31,"item":68},{"__typename":31,"title":69,"title_en":69,"slug":70,"children_series":71},"Lab","lab",[],{"article":73,"renderedBody":264,"renderedTakeaway":265,"latestArticles":266,"toc":333},{"id":74,"title":75,"title_en":76,"slug":77,"excerpt":78,"excerpt_en":79,"has_translation":12,"published_at":80,"reading_time":81,"hidden_from_listing":15,"tags":82,"thumbnail_image":21,"og_image":87,"series":88,"authors":257,"body_locale":260,"available_locales":261,"body":263,"takeway":21,"prerequisite":21},2,"LLM Bakal Mentok, Terus Gimana Dong?","LLMs Are Going to Hit a Wall. What Then?","llm-bakal-mentok-terus-gimana-dong","LLM kayak ChatGPT itu jago ngerangkai kata, tapi nggak beneran paham realitas. Yann LeCun bilang arsitektur ini bakal mentok dan butuh solusi baru.","LLMs like ChatGPT are great at stringing words together, but they never really grasp reality. Yann LeCun says this architecture will hit a wall and needs a new answer.","2026-06-03T08:55:33.003Z",4,[83,84,85,86],"artificial-intelligence","llm","yann-lecun","jepa","\u002Fog\u002Fllm-bakal-mentok-terus-gimana-dong.png",{"id":89,"title":35,"title_en":36,"slug":37,"description":21,"description_en":90,"articles":91,"children_series":249,"parent_series":250},8,"Notes on AI that go past the demo: how the models work, where they break, and what changes once you put them in production.",[92,105,113,128,142,156,173,190,211,230],{"id":93,"title":94,"title_en":95,"slug":96,"excerpt":97,"excerpt_en":98,"has_translation":12,"published_at":99,"reading_time":81,"hidden_from_listing":15,"tags":100,"thumbnail_image":21,"og_image":102,"series":103,"authors":104},1,"Kenapa AI Suka 'Halu'? Kenalan sama JEPA, Otak AI yang Punya Akal Sehat","Why Does AI Keep Hallucinating? Meet JEPA, the AI Brain With Common Sense","kenapa-ai-suka-halu-kenalan-sama-jepa-otak-ai-yang-punya-akal-sehat","AI generatif sering 'halu' karena cuma jago nebak pola kata. JEPA coba ngerombak itu dengan ngajarin AI cara kerja dunia nyata kayak otak manusia.","Generative AI hallucinates because all it does is guess word patterns. JEPA tries to change that by teaching machines how the real world works, the way a human brain does.","2026-06-02T08:09:05.595Z",[86,83,85,101],"agi","\u002Fog\u002Fkenapa-ai-suka-halu-kenalan-sama-jepa-otak-ai-yang-punya-akal-sehat.png",{"id":89,"title":35,"title_en":36,"slug":37},[],{"id":74,"title":75,"title_en":76,"slug":77,"excerpt":78,"excerpt_en":79,"has_translation":12,"published_at":80,"reading_time":81,"hidden_from_listing":15,"tags":82,"thumbnail_image":21,"og_image":87,"series":106,"authors":107},{"id":89,"title":35,"title_en":36,"slug":37},[108],{"authors_id":109},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},"Daffa Izzuddin","izzudd","Writer",{"id":114,"title":115,"title_en":116,"slug":117,"excerpt":118,"excerpt_en":119,"has_translation":12,"published_at":120,"reading_time":81,"hidden_from_listing":15,"tags":121,"thumbnail_image":21,"og_image":123,"series":124,"authors":125},3,"Mengenal JEPA: Saat AI Mulai Belajar Punya \"Akal Sehat\"","Meet JEPA: Where AI Starts Learning Common Sense","kenapa-ai-suka-halu-dan-gimana-jepa-beresin-masalah-ini","AI sering 'halu' karena cuma nebak probabilitas data tanpa paham realita. Arsitektur JEPA mencoba mengubah itu dengan membangun model dunia di kepalanya.","AI hallucinates because it only guesses data probabilities without understanding reality. The JEPA architecture tries to change that by building a world model in its head.","2026-06-03T09:32:50.446Z",[83,86,122,85],"world-models","\u002Fog\u002Fkenapa-ai-suka-halu-dan-gimana-jepa-beresin-masalah-ini.png",{"id":89,"title":35,"title_en":36,"slug":37},[126],{"authors_id":127},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":81,"title":129,"title_en":130,"slug":131,"excerpt":132,"excerpt_en":133,"has_translation":12,"published_at":134,"reading_time":81,"hidden_from_listing":15,"tags":135,"thumbnail_image":21,"og_image":137,"series":138,"authors":139},"Bukan Sekadar Generative AI, JEPA Coba Kejar Pemahaman Konsep ala Manusia","Beyond Generative AI: JEPA Chases Human-Style Conceptual Understanding","kenapa-ai-masih-gagal-paham-fisika-dasar-mengenal-arsitektur-jepa","AI biasanya cuma jago nebak piksel, tapi JEPA coba belajar lewat konsep abstrak di ruang laten. Ternyata, cara ini bisa sampai dipakai buat baca gelombang otak.","AI usually only guesses pixels, while JEPA tries to learn from abstract concepts in latent space. That approach turns out to reach as far as reading brain waves.","2026-06-03T09:34:14.840Z",[86,85,136,101],"machine-learning","\u002Fog\u002Fkenapa-ai-masih-gagal-paham-fisika-dasar-mengenal-arsitektur-jepa.png",{"id":89,"title":35,"title_en":36,"slug":37},[140],{"authors_id":141},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":14,"title":143,"title_en":144,"slug":145,"excerpt":146,"excerpt_en":147,"has_translation":12,"published_at":148,"reading_time":81,"hidden_from_listing":15,"tags":149,"thumbnail_image":21,"og_image":151,"series":152,"authors":153},"Di Balik Dapur JEPA: Rahasia EMA, VICReg, dan Efisiensi Komputasi","Inside JEPA's Kitchen: The Secrets of EMA, VICReg, and Compute Efficiency","di-balik-dapur-jepa-rahasia-ema-vicreg-dan-efisiensi-komputasi","Bosen sama AI yang cuma jago nebak kata atau ngafalin pixel? Kenalan sama JEPA, arsitektur revolusioner yang ngajarin mesin punya \"akal sehat\" secara super efisien. Yuk intip jeroan teknisnya!","Bored of AI that only guesses words or memorizes pixels? Meet JEPA, an architecture that teaches machines common sense very cheaply. Here is a look at its technical insides.","2026-06-07T16:14:13.723Z",[86,150],"artificial-intelligent","\u002Fog\u002Fdi-balik-dapur-jepa-rahasia-ema-vicreg-dan-efisiensi-komputasi.png",{"id":89,"title":35,"title_en":36,"slug":37},[154],{"authors_id":155},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":157,"title":158,"title_en":159,"slug":160,"excerpt":161,"excerpt_en":162,"has_translation":12,"published_at":163,"reading_time":89,"hidden_from_listing":15,"tags":164,"thumbnail_image":21,"og_image":170,"series":171,"authors":172},125,"MCP dibongkar ulang: kabar baik buat kamu yang capek nulis konektor","MCP, Rebuilt: Good News If You're Tired of Writing Connectors","mcp-dibongkar-ulang-kabar-baik-buat-kamu","Kamu punya beberapa API internal dan satu agent AI, tapi kodenya harus ditulis ulang tiap kali modelnya diganti. MCP datang buat mengakhiri itu, dan tahun ini protokolnya dibongkar ulang supaya server-nya bisa jalan seperti web server biasa. Yuk lihat apa yang berubah. 🔧","You have a few internal APIs and one AI agent, but the code has to be rewritten every time the model changes. MCP ends that, and this year the protocol was rebuilt so servers can run like ordinary web servers. Here's what changed. 🔧","2026-10-16T05:00:00.000Z",[165,166,167,84,168,169],"mcp","model-context-protocol","ai-agent","backend","python","\u002Fog\u002Fmcp-dibongkar-ulang-kabar-baik-buat-kamu.png",{"id":89,"title":35,"title_en":36,"slug":37},[],{"id":174,"title":175,"title_en":176,"slug":177,"excerpt":178,"excerpt_en":179,"has_translation":12,"published_at":180,"reading_time":181,"hidden_from_listing":15,"tags":182,"thumbnail_image":21,"og_image":185,"series":186,"authors":187},126,"Memori AI Agent: dari Vector Store ke Temporal Graph","AI Agent Memory: From Vector Store to Temporal Graph","memori-ai-agent-vector-store-ke-temporal-graph","Dua fakta yang saling bertentangan bikin agent menjawab ngawur. Ini bedanya vector store dan temporal knowledge graph dalam menjaga state jangka panjang.","Two contradictory facts make an agent answer nonsense. Here is how a vector store and a temporal knowledge graph differ at keeping long-term state.","2026-10-18T13:13:29.458Z",7,[183,84,168,184],"ai","system-design","\u002Fog\u002Fmemori-ai-agent-vector-store-ke-temporal-graph.png",{"id":89,"title":35,"title_en":36,"slug":37},[188],{"authors_id":189},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":191,"title":192,"title_en":193,"slug":194,"excerpt":195,"excerpt_en":196,"has_translation":12,"published_at":197,"reading_time":198,"hidden_from_listing":15,"tags":199,"thumbnail_image":21,"og_image":206,"series":207,"authors":208},127,"Kenapa RAG yang cuma andalkan cosine similarity sering nyasar","Why RAG That Leans on Cosine Similarity Alone Keeps Missing","kenapa-rag-cuma-cosine-similarity-sering-nyasar","RAG yang cuma andalkan embedding balikin chunk yang mirip maknanya, bukan yang kamu butuh. Ini penyebabnya, dan dua lapis penambalnya.","RAG that leans only on embeddings returns chunks that look close in meaning, not the ones you need. Here is why, and the two layers that patch it.","2026-10-19T02:04:06.718Z",6,[200,201,202,203,204,205],"rag","vector-search","bm25","hybrid-search","retrieval","embedding","\u002Fog\u002Fkenapa-rag-cuma-cosine-similarity-sering-nyasar.png",{"id":89,"title":35,"title_en":36,"slug":37},[209],{"authors_id":210},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":212,"title":213,"title_en":214,"slug":215,"excerpt":216,"excerpt_en":217,"has_translation":12,"published_at":218,"reading_time":198,"hidden_from_listing":15,"tags":219,"thumbnail_image":21,"og_image":225,"series":226,"authors":227},128,"Linear atau graf: cara mengatur banyak agent tanpa loop abadi","Linear or Graph: How to Run Many Agents Without an Eternal Loop","multi-agen-linear-atau-grafik","Satu agent dengan dua puluh tool gampang kacau. Begitu beberapa agent saling revisi, kamu butuh graf yang mendukung cycle dan angka yang membuatnya berhenti.","One agent with twenty tools gets messy fast. Once several agents revise each other, you need a graph that allows cycles and a number that makes it stop.","2026-10-21T16:05:00.000Z",[220,221,222,223,224,84],"multi-agent","langgraph","autogen","orkestrasi","checkpoint","\u002Fog\u002Fmulti-agen-linear-atau-grafik.png",{"id":89,"title":35,"title_en":36,"slug":37},[228],{"authors_id":229},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":231,"title":232,"title_en":233,"slug":234,"excerpt":235,"excerpt_en":236,"has_translation":12,"published_at":237,"reading_time":14,"hidden_from_listing":15,"tags":238,"thumbnail_image":21,"og_image":244,"series":245,"authors":246},129,"Biar RAG nggak kena 504: pisahin jalur query dan ingestion","So Your RAG Stops Hitting 504: Split the Query Path From Ingestion","rag-pisah-jalur-query-dan-ingestion","Satu service RAG yang merangkap parsing dan menjawab kueri bakal tumbang pas ada yang upload PDF 200 halaman. Obatnya dua jalur dan satu queue.","One RAG service that both parses uploads and answers queries will fall over the moment someone uploads a 200-page PDF. The cure is two paths and one queue.","2026-10-22T01:30:00.000Z",[200,239,240,241,242,243],"cloud-run","cloud-tasks","gcp","serverless","arsitektur","\u002Fog\u002Frag-pisah-jalur-query-dan-ingestion.png",{"id":89,"title":35,"title_en":36,"slug":37},[247],{"authors_id":248},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},[],{"id":24,"title":25,"title_en":26,"slug":27,"parent_series":21,"children_series":251},[252,253,255],{"id":89,"title":35,"title_en":36,"slug":37,"cover_image":21},{"id":254,"title":39,"title_en":40,"slug":41,"cover_image":21},12,{"id":256,"title":43,"title_en":44,"slug":45,"cover_image":21},15,[258],{"authors_id":259},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},"id",[262,260],"en","Ngomongin soal AI hari ini, pasti yang jadi bintang utamanya adalah *Large Language Models* (LLM) seperti ChatGPT, Claude, atau Gemini. Mereka bisa nulis kode, bikin puisi, sampai ngasih saran percintaan. Sekilas, rasanya kita udah dekat banget sama yang namanya AGI (*Artificial General Intelligence* – alias AI yang kecerdasannya setara manusia).\n\nTapi, sadar nggak sih kalau kadang AI ini suka \"halu\" dan bikin kesalahan yang bahkan anak SD aja nggak bakal lakuin?\n\nNah, di kalangan para ahli AI, lagi ada perdebatan seru soal ini. Salah satu tokoh utama AI, Yann LeCun (Chief AI Scientist di Meta), dengan berani bilang kalau arsitektur LLM yang ada sekarang pada akhirnya bakal **mentok**. Jalan menuju AGI nggak bisa cuma ngandelin LLM. Jawabannya, menurut LeCun, ada di arsitektur baru bernama **JEPA** (*Joint-Embedding Predictive Architecture*).\n\n---\n\n## Kutukan \"Burung Beo\" (Stochastic Parrot)\n\nKenapa sih LLM sering dibilang bakal mentok? Alasan utamanya ada di cara kerja mereka. LLM yang kita pakai sekarang pada dasarnya adalah mesin penebak kata (*next-token prediction*). Mereka dilatih dengan triliunan teks dari internet untuk menebak probabilitas kata apa yang harus muncul selanjutnya.\n\nMasalahnya, pendekatan ini bikin LLM terjebak jadi apa yang disebut peneliti sebagai **\"Stochastic Parrot\"** atau burung beo statistik.\n\nIbarat burung beo yang pintar meniru suara manusia, LLM jago banget merangkai kata yang *kelihatan* cerdas, tapi mereka nggak benar-benar paham makna di balik kata-kata itu. Mereka nggak punya pondasi pemahaman soal dunia fisik (realitas). Makanya, kalau disuruh ngejelasin interaksi fisik antar benda atau logika sebab-akibat yang kompleks, mereka sering *error* atau berhalusinasi, karena pengetahuan mereka murni cuma sebatas probabilitas teks, bukan pemahaman fisika dasar.\n\n## Kenapa Yann LeCun Yakin LLM Bakal Mentok?\n\nDi berbagai forum, Yann LeCun sering nekenin kalau model *autoregressive* (sistem yang memprediksi kata demi kata berulang kali) punya kelemahan fatal: **akumulasi error**.\n\nKalau AI bikin satu kesalahan kecil di awal kalimat, kesalahan itu bakal merembet dan makin parah di kalimat-kalimat berikutnya. Selain itu, LeCun berargumen bahwa bahasa (teks) itu sebenarnya adalah bentuk informasi yang sudah sangat dikompres. Banyak hal di dunia nyata (seperti gravitasi, cara benda jatuh, atau cara manusia bergerak) yang nggak bisa dijelaskan murni cuma lewat teks. Manusia dan hewan belajar dengan melihat dan berinteraksi dengan dunia fisik, bukan cuma dari baca buku.\n\nJadi, mau sebanyak apa pun data teks yang disuapin ke LLM, mereka nggak akan pernah bisa mencapai AGI karena mereka buta terhadap cara kerja dunia fisik. Mereka itu ibarat *interface* (tampilan luar) yang jago ngomong, tapi nggak punya *backend* (otak logika) yang kuat.\n\n## Masuklah JEPA: Memprediksi Makna di \"Latent Space\"\n\nKalau LLM itu ibarat nebak kata, dan *Generative AI* gambar (seperti DALL-E atau Nano Banana) itu ibarat nebak piksel warna, **JEPA** datang dengan cara berpikir yang beda total. JEPA nggak peduli sama token (kata) atau piksel. JEPA memprediksi di dalam **Representation Space** atau **Latent Space** (ruang abstrak\u002Fkonseptual).\n\nGampangnya gini: Coba bayangin kamu lagi mau nyeritain sebuah foto ke temanmu. Apakah otakmu memilih kata pertama dulu, lalu mikir kata kedua, lalu kata ketiga? Nggak, kan? Otakmu bakal ngebentuk **konsep utuhnya** dulu (misal: \"kucing oren lagi tidur di sofa\"), baru setelah itu diterjemahin ke kata-kata.\n\nItulah inti dari riset terbaru seperti *VL-JEPA* (Vision-Language JEPA). Alih-alih buang-buang tenaga komputasi buat menebak setiap kata atau menggambar ulang setiap detail piksel (seperti helai bulu kucing yang nggak penting), JEPA dilatih untuk menebak **\"makna\"** atau **\"embedding\"**-nya. Ini bikin JEPA jauh lebih efisien, komputasinya lebih ringan, dan proses kerjanya jauh lebih mirip dengan cara kerja kognitif manusia.\n\n## V-JEPA: Ngajarin AI Ilmu Fisika Lewat Video\n\nSalah satu terobosan keren dari arsitektur ini adalah **V-JEPA** (Video JEPA). Kalau LLM belajar dari teks, V-JEPA belajar murni dari nonton jutaan jam video tanpa label.\n\nKarena dia bekerja di *latent space*, V-JEPA belajar memahami \"ilmu fisika intuitif\". Dia bisa memprediksi apa yang bakal terjadi di adegan selanjutnya tanpa harus merender ulang videonya. Dia jadi paham bahwa kalau ada gelas jatuh ke lantai, gelasnya bakal pecah. Ini adalah fondasi dari *common sense* (akal sehat) yang selama ini jadi \"kryptonite\" buat LLM konvensional.\n\nDengan model dunia (World Model) yang kuat dari V-JEPA, AI masa depan nantinya nggak cuma bisa ngobrol, tapi bisa dipakai untuk sistem robotika yang mampu merencanakan tindakan di dunia nyata tanpa perlu disuapin instruksi yang kaku.\n\n---\n\n## Masa Depan: Mengawinkan \"Mulut\" dan \"Otak\"?\n\nJalan menuju AGI kayaknya nggak bakal dicapai kalau kita cuma ngandelin AI yang jago omong \"omdo\" doang haha. Perdebatan ini ngasih kita perspektif baru: LLM memang luar biasa sebagai alat komunikasi, tapi untuk jadi benar-benar pintar? AI masih butuh pondasi pemahaman dunia nyata.\n\nKe depannya, arsitektur masa depan bisa jadi gabungan dari keduanya. LLM jadi \"mulut\" (interface komunikasi) yang luwes, sementara arsitektur seperti JEPA bakal jadi \"otak\" (*backend\u002Fworld model*) yang memastikan AI tersebut punya akal sehat dan paham hukum realitas.\n\nBukan nggak mungkin, dalam 3-5 tahun ke depan, paradigma *next-token prediction* bakal mulai ditinggalkan, dan kita akan masuk ke era di mana AI beneran \"mikir\" dulu sebelum ngomong. Kita lihat saja :D\n\nGimana menurutmu? Percaya kalau LLM bakal mentok, atau justru yakin kalau ChatGPT dkk masih bisa berevolusi jadi AGI?\n","\u003Cp>Ngomongin soal AI hari ini, pasti yang jadi bintang utamanya adalah \u003Cem>Large Language Models\u003C\u002Fem> (LLM) seperti ChatGPT, Claude, atau Gemini. Mereka bisa nulis kode, bikin puisi, sampai ngasih saran percintaan. Sekilas, rasanya kita udah dekat banget sama yang namanya AGI (\u003Cem>Artificial General Intelligence\u003C\u002Fem> – alias AI yang kecerdasannya setara manusia).\u003C\u002Fp>\n\u003Cp>Tapi, sadar nggak sih kalau kadang AI ini suka "halu" dan bikin kesalahan yang bahkan anak SD aja nggak bakal lakuin?\u003C\u002Fp>\n\u003Cp>Nah, di kalangan para ahli AI, lagi ada perdebatan seru soal ini. Salah satu tokoh utama AI, Yann LeCun (Chief AI Scientist di Meta), dengan berani bilang kalau arsitektur LLM yang ada sekarang pada akhirnya bakal \u003Cstrong>mentok\u003C\u002Fstrong>. Jalan menuju AGI nggak bisa cuma ngandelin LLM. Jawabannya, menurut LeCun, ada di arsitektur baru bernama \u003Cstrong>JEPA\u003C\u002Fstrong> (\u003Cem>Joint-Embedding Predictive Architecture\u003C\u002Fem>).\u003C\u002Fp>\n\u003Chr>\n\u003Ch2 id=\"kutukan-burung-beo-stochastic-parrot\">Kutukan "Burung Beo" (Stochastic Parrot)\u003C\u002Fh2>\n\u003Cp>Kenapa sih LLM sering dibilang bakal mentok? Alasan utamanya ada di cara kerja mereka. LLM yang kita pakai sekarang pada dasarnya adalah mesin penebak kata (\u003Cem>next-token prediction\u003C\u002Fem>). Mereka dilatih dengan triliunan teks dari internet untuk menebak probabilitas kata apa yang harus muncul selanjutnya.\u003C\u002Fp>\n\u003Cp>Masalahnya, pendekatan ini bikin LLM terjebak jadi apa yang disebut peneliti sebagai \u003Cstrong>"Stochastic Parrot"\u003C\u002Fstrong> atau burung beo statistik.\u003C\u002Fp>\n\u003Cp>Ibarat burung beo yang pintar meniru suara manusia, LLM jago banget merangkai kata yang \u003Cem>kelihatan\u003C\u002Fem> cerdas, tapi mereka nggak benar-benar paham makna di balik kata-kata itu. Mereka nggak punya pondasi pemahaman soal dunia fisik (realitas). Makanya, kalau disuruh ngejelasin interaksi fisik antar benda atau logika sebab-akibat yang kompleks, mereka sering \u003Cem>error\u003C\u002Fem> atau berhalusinasi, karena pengetahuan mereka murni cuma sebatas probabilitas teks, bukan pemahaman fisika dasar.\u003C\u002Fp>\n\u003Ch2 id=\"kenapa-yann-lecun-yakin-llm-bakal-mentok\">Kenapa Yann LeCun Yakin LLM Bakal Mentok?\u003C\u002Fh2>\n\u003Cp>Di berbagai forum, Yann LeCun sering nekenin kalau model \u003Cem>autoregressive\u003C\u002Fem> (sistem yang memprediksi kata demi kata berulang kali) punya kelemahan fatal: \u003Cstrong>akumulasi error\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>Kalau AI bikin satu kesalahan kecil di awal kalimat, kesalahan itu bakal merembet dan makin parah di kalimat-kalimat berikutnya. Selain itu, LeCun berargumen bahwa bahasa (teks) itu sebenarnya adalah bentuk informasi yang sudah sangat dikompres. Banyak hal di dunia nyata (seperti gravitasi, cara benda jatuh, atau cara manusia bergerak) yang nggak bisa dijelaskan murni cuma lewat teks. Manusia dan hewan belajar dengan melihat dan berinteraksi dengan dunia fisik, bukan cuma dari baca buku.\u003C\u002Fp>\n\u003Cp>Jadi, mau sebanyak apa pun data teks yang disuapin ke LLM, mereka nggak akan pernah bisa mencapai AGI karena mereka buta terhadap cara kerja dunia fisik. Mereka itu ibarat \u003Cem>interface\u003C\u002Fem> (tampilan luar) yang jago ngomong, tapi nggak punya \u003Cem>backend\u003C\u002Fem> (otak logika) yang kuat.\u003C\u002Fp>\n\u003Ch2 id=\"masuklah-jepa-memprediksi-makna-di-latent-space\">Masuklah JEPA: Memprediksi Makna di "Latent Space"\u003C\u002Fh2>\n\u003Cp>Kalau LLM itu ibarat nebak kata, dan \u003Cem>Generative AI\u003C\u002Fem> gambar (seperti DALL-E atau Nano Banana) itu ibarat nebak piksel warna, \u003Cstrong>JEPA\u003C\u002Fstrong> datang dengan cara berpikir yang beda total. JEPA nggak peduli sama token (kata) atau piksel. JEPA memprediksi di dalam \u003Cstrong>Representation Space\u003C\u002Fstrong> atau \u003Cstrong>Latent Space\u003C\u002Fstrong> (ruang abstrak\u002Fkonseptual).\u003C\u002Fp>\n\u003Cp>Gampangnya gini: Coba bayangin kamu lagi mau nyeritain sebuah foto ke temanmu. Apakah otakmu memilih kata pertama dulu, lalu mikir kata kedua, lalu kata ketiga? Nggak, kan? Otakmu bakal ngebentuk \u003Cstrong>konsep utuhnya\u003C\u002Fstrong> dulu (misal: "kucing oren lagi tidur di sofa"), baru setelah itu diterjemahin ke kata-kata.\u003C\u002Fp>\n\u003Cp>Itulah inti dari riset terbaru seperti \u003Cem>VL-JEPA\u003C\u002Fem> (Vision-Language JEPA). Alih-alih buang-buang tenaga komputasi buat menebak setiap kata atau menggambar ulang setiap detail piksel (seperti helai bulu kucing yang nggak penting), JEPA dilatih untuk menebak \u003Cstrong>"makna"\u003C\u002Fstrong> atau \u003Cstrong>"embedding"\u003C\u002Fstrong>-nya. Ini bikin JEPA jauh lebih efisien, komputasinya lebih ringan, dan proses kerjanya jauh lebih mirip dengan cara kerja kognitif manusia.\u003C\u002Fp>\n\u003Ch2 id=\"v-jepa-ngajarin-ai-ilmu-fisika-lewat-video\">V-JEPA: Ngajarin AI Ilmu Fisika Lewat Video\u003C\u002Fh2>\n\u003Cp>Salah satu terobosan keren dari arsitektur ini adalah \u003Cstrong>V-JEPA\u003C\u002Fstrong> (Video JEPA). Kalau LLM belajar dari teks, V-JEPA belajar murni dari nonton jutaan jam video tanpa label.\u003C\u002Fp>\n\u003Cp>Karena dia bekerja di \u003Cem>latent space\u003C\u002Fem>, V-JEPA belajar memahami "ilmu fisika intuitif". Dia bisa memprediksi apa yang bakal terjadi di adegan selanjutnya tanpa harus merender ulang videonya. Dia jadi paham bahwa kalau ada gelas jatuh ke lantai, gelasnya bakal pecah. Ini adalah fondasi dari \u003Cem>common sense\u003C\u002Fem> (akal sehat) yang selama ini jadi "kryptonite" buat LLM konvensional.\u003C\u002Fp>\n\u003Cp>Dengan model dunia (World Model) yang kuat dari V-JEPA, AI masa depan nantinya nggak cuma bisa ngobrol, tapi bisa dipakai untuk sistem robotika yang mampu merencanakan tindakan di dunia nyata tanpa perlu disuapin instruksi yang kaku.\u003C\u002Fp>\n\u003Chr>\n\u003Ch2 id=\"masa-depan-mengawinkan-mulut-dan-otak\">Masa Depan: Mengawinkan "Mulut" dan "Otak"?\u003C\u002Fh2>\n\u003Cp>Jalan menuju AGI kayaknya nggak bakal dicapai kalau kita cuma ngandelin AI yang jago omong "omdo" doang haha. Perdebatan ini ngasih kita perspektif baru: LLM memang luar biasa sebagai alat komunikasi, tapi untuk jadi benar-benar pintar? AI masih butuh pondasi pemahaman dunia nyata.\u003C\u002Fp>\n\u003Cp>Ke depannya, arsitektur masa depan bisa jadi gabungan dari keduanya. LLM jadi "mulut" (interface komunikasi) yang luwes, sementara arsitektur seperti JEPA bakal jadi "otak" (\u003Cem>backend\u002Fworld model\u003C\u002Fem>) yang memastikan AI tersebut punya akal sehat dan paham hukum realitas.\u003C\u002Fp>\n\u003Cp>Bukan nggak mungkin, dalam 3-5 tahun ke depan, paradigma \u003Cem>next-token prediction\u003C\u002Fem> bakal mulai ditinggalkan, dan kita akan masuk ke era di mana AI beneran "mikir" dulu sebelum ngomong. Kita lihat saja :D\u003C\u002Fp>\n\u003Cp>Gimana menurutmu? Percaya kalau LLM bakal mentok, atau justru yakin kalau ChatGPT dkk masih bisa berevolusi jadi AGI?\u003C\u002Fp>\n","",[267,284,297,310,323,328],{"id":268,"title":269,"title_en":21,"slug":270,"excerpt":271,"excerpt_en":21,"has_translation":15,"published_at":272,"reading_time":198,"hidden_from_listing":15,"tags":273,"thumbnail_image":21,"og_image":275,"series":276,"authors":281},133,"Python Strings: the Badge That Only Worked for One Name","python-strings","The badge script printed the right name for the person who wrote it and the wrong name for everybody else. Slicing counts characters, and a name is words.","2026-10-28T13:00:00.000Z",[169,274],"string","\u002Fog\u002Fpython-strings.png",{"id":277,"title":278,"title_en":279,"slug":280},30,"Koleksi dan String","Collections and Strings","python-collections",[282],{"authors_id":283},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":285,"title":286,"title_en":21,"slug":287,"excerpt":288,"excerpt_en":21,"has_translation":15,"published_at":289,"reading_time":198,"hidden_from_listing":15,"tags":290,"thumbnail_image":21,"og_image":292,"series":293,"authors":294},132,"Python Tuples: Why All Three Points Are the Same Point","python-tuples","Three points walk into a loop and come out identical. The list was never copied three times; the name was. A tuple makes that bug impossible.","2026-10-28T12:00:00.000Z",[169,291],"tuple","\u002Fog\u002Fpython-tuples.png",{"id":277,"title":278,"title_en":279,"slug":280},[295],{"authors_id":296},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":298,"title":299,"title_en":21,"slug":300,"excerpt":301,"excerpt_en":21,"has_translation":15,"published_at":302,"reading_time":198,"hidden_from_listing":15,"tags":303,"thumbnail_image":21,"og_image":305,"series":306,"authors":307},131,"Python Sets: the Duplicate That Keeps Coming Back","python-sets","A de-duplication loop that works on clean data leaves four tags where there should be two. The bug is case, and the fix is a `set`.","2026-10-28T11:00:00.000Z",[169,304],"set","\u002Fog\u002Fpython-sets.png",{"id":277,"title":278,"title_en":279,"slug":280},[308],{"authors_id":309},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":311,"title":312,"title_en":21,"slug":313,"excerpt":314,"excerpt_en":21,"has_translation":15,"published_at":315,"reading_time":198,"hidden_from_listing":15,"tags":316,"thumbnail_image":21,"og_image":318,"series":319,"authors":320},130,"Python Dictionaries: the Key That Isn't There","python-dictionary","A word counter dies the first time it meets a word twice. The fix is one line; what takes longer is knowing which of a dict's three lookups to reach for.","2026-10-28T10:00:00.000Z",[169,317],"dictionary","\u002Fog\u002Fpython-dictionary.png",{"id":277,"title":278,"title_en":279,"slug":280},[321],{"authors_id":322},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":231,"title":232,"title_en":233,"slug":234,"excerpt":235,"excerpt_en":236,"has_translation":12,"published_at":237,"reading_time":14,"hidden_from_listing":15,"tags":238,"thumbnail_image":21,"og_image":244,"series":324,"authors":325},{"id":89,"title":35,"title_en":36,"slug":37},[326],{"authors_id":327},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},{"id":212,"title":213,"title_en":214,"slug":215,"excerpt":216,"excerpt_en":217,"has_translation":12,"published_at":218,"reading_time":198,"hidden_from_listing":15,"tags":219,"thumbnail_image":21,"og_image":225,"series":329,"authors":330},{"id":89,"title":35,"title_en":36,"slug":37},[331],{"authors_id":332},{"id":93,"name":110,"slug":111,"role":112,"profile_picture":21},[334,337,340,343,346],{"text":335,"depth":74,"id":336},"Kutukan \"Burung Beo\" (Stochastic Parrot)","kutukan-burung-beo-stochastic-parrot",{"text":338,"depth":74,"id":339},"Kenapa Yann LeCun Yakin LLM Bakal Mentok?","kenapa-yann-lecun-yakin-llm-bakal-mentok",{"text":341,"depth":74,"id":342},"Masuklah JEPA: Memprediksi Makna di \"Latent Space\"","masuklah-jepa-memprediksi-makna-di-latent-space",{"text":344,"depth":74,"id":345},"V-JEPA: Ngajarin AI Ilmu Fisika Lewat Video","v-jepa-ngajarin-ai-ilmu-fisika-lewat-video",{"text":347,"depth":74,"id":348},"Masa Depan: Mengawinkan \"Mulut\" dan \"Otak\"?","masa-depan-mengawinkan-mulut-dan-otak",1793295834433]