@erry_aone@indopremier Sudah kak, coba chat CSnya aja via whatsapp dan sebutkan jenis devicenya secara detail (misal: iPhone 18 Pro Max)
Nanti diurus kok cepet
Once every ten days, a bright canary-yellow bullet train glides into Tokyo Station at full speed, yet not a single passenger is allowed to step on board.
Its windows are mostly covered from the inside, its timetable is never published for the public, and there are no ticketed seats in any of its seven cars.
In Japan, every child and every commuter knows this train by two English words: Doctor Yellow.
Since the Tokaido Shinkansen began running between Tokyo and Osaka in October 1964, Japanese bullet trains have carried more than seven billion passengers at speeds up to one hundred and eighty-five miles per hour with zero passenger fatalities from derailments or collisions.
Critics sometimes assume that kind of sixty-year safety record is a miracle of luck.
It is not luck; it is what happens when a nation builds a high-speed hospital train whose only patient is the steel track itself.
Why does Japan run a seven-car bullet train painted bright yellow with zero paying customers on board?
In the early days of high-speed rail, engineers realized that checking hundreds of miles of track on foot at night with hand gauges was too slow, and that tracks behave differently when a three-hundred-ton train slams over them at one hundred and sixty-eight miles per hour.
So Japanese railway engineers took a seven-car Class 923 Shinkansen trainset and packed every car from nose to tail with laser-optical sensors, gyroscopes, accelerometers, and high-speed cameras.
Running in the middle of the regular passenger schedule at a full one hundred and sixty-eight miles per hour, Doctor Yellow fires laser beams onto the left and right rails to measure track width, vertical dip, and curvature alignment down to one-tenth of a millimeter, about the thickness of a single human hair.
At the same time, two special observation domes built into the roof and laser pantographs check the overhead high-voltage copper wire for microscopic wear before a single spark can cause a power outage.
The instant Doctor Yellow detects a rail that has shifted by even a fraction of an inch, it beams the exact GPS coordinates to maintenance crews waiting along the line so they can fix the spot that very night after the last passenger train goes to sleep.
Because its schedule is kept secret so crowds do not pack the platforms, Japanese commuters treat spotting Doctor Yellow by chance as a genuine good-luck charm that brings happiness for the rest of the week.
Safety is never an accident when engineers check their work down to the width of a hair.
(Note: Every photograph-style illustration on my feed is created by AI.)
In a weird way, everybody is rooting for you. When you go to a job interview, the interviewer wants you to be amazing so they don’t have to look anymore. when you go on a date, the person wants to have an amazing date with you. If you make a piece of content, people want to see amazing pieces of content and they have no problem if it’s yours.
the instant you get into the mindset that people want you to do well, your entire world gets better. And you’ll be shocked to know this pattern keeps repeating in different areas of life.
this is f*cking insane. andrej karpathy: "the biggest regret of every engineer in my field right now is every AI conversation they never saved, and in 10 years the ones who started saving in 2024 will be running circles around the rest of us"
andrej karpathy, co-founder of openai, pulled 16M views on one idea: stop using AI to write code, use it to build a second brain that remembers every article, pdf, transcript and conversation you ever fed it
you point claude code at a folder, drop anything into it, and claude reads it, cross-links it with everything already there, and writes it into a living wiki. your knowledge finally compounds instead of leaking out of your head every friday
the stack:
→ brew install obsidian, mkdir ~/vault, cd ~/vault
→ run claude in that folder and type "init a knowledge vault with CLAUDE.md as the rules, /raw for sources, /wiki for your pages"
→ drop any file into /raw (youtube transcript, meeting notes, 400-page pdf, screenshot of a tweet) and type "ingest"
→ claude writes a wiki page, links it to 5 existing ones automatically, updates CLAUDE.md with any new pattern it noticed about how you think
→ never open a blank claude chat again. every question goes through the vault, every answer cites your own sources with dates and links
5 minutes to set up. compounding knowledge from that evening forward
the ugly part most will scroll past: in 10 years the engineers who started this in 2024 will remember every book, every podcast, every meeting with receipts. you will remember none of them. karpathy already picked which side he is on and you are still not on his
full step-by-step is in the article below
bookmark this before you waste another monday
Beberapa waktu lalu, ada perusahaan ajak ketemuan.
Mereka minta masukan karena lagi bangun AI untuk support operasional kantor, sebuah use case yang aku rasa dihadapi ribuan perusahaan lain saat ini.
Pas aku coba chat AI mereka, jawabannya muter2 dan ga nyambung.
Ternyata mereka kasih akses database ke AI agent yang serving chatnya, dan berharap AI modelnya bisa cukup pintar untuk decide sendiri SQL query apa yang dibutuhkan untuk jawab pertanyaan user, dan kemudian execute sendiri query tersebut.
Pendekatan kaya gitu riskan error. Untuk bangun AI yang efektif, langkah pertama adalah kita perlu tahu batasan kemampuan model AI yang kita gunakan.
Bahkan frontier model (c. 2026) sekalipun bakal sering error kalau pertanyaan dari user bisa diinterpretasikan open-ended jadi ribuan SQL query berbeda untuk raw data yang kita punya.
Di @PerkaraAI, saat sidang kami sedang berjalan, satu "aha moment" dimana jawaban dari AI yang kami rancang mulai kerasa manfaatnya, adalah ketika kami rancang semantic model sebagai jembatan antara AI models dengan data-data pada perkara hukum kami.
Semantic model ini ada banyak jenisnya, silahkan riset dan cari2 sendiri yang paling cocok seperti apa untuk use case yang dihadapi.
Dalam perkara kami, semantic model yang kami pilih untuk organize knowledge dalam sebuah perkara, adalah multi-layered knowledge graph, dimana beberapa layer di dalamnya menggunakan ontologi yang sudah dirancang khusus dengan semantik investigasi perkara hukum.
Knowledge graph ini yang bisa dilihat pada kutipan post di bawah, dan menjadi semacam "otak pengetahuan" dari sebuah perkara. Sebuah perkara bisa memuat ~1 juta entitas dan keterhubungan antar entitas.
Lalu, contoh konkretnya gimana?
Misalnya, user ingin @PerkaraAI generate daftar seluruh pertemuan yang seorang saksi tertentu klaim karyawannya pernah lakukan dengan "Pejabat Z".
AI agent yang berusaha jawab pertanyaan di atas bisa saja generate SQL query berikut:
SELECT DISTINCT k.nama
FROM karyawan k
JOIN dokumen d ON d.isi LIKE '%' || k.nama || '%'
WHERE d.isi LIKE '%Pejabat Z%'
AND d.isi LIKE '%meeting%';
Terlihat baik2 saja, tapi sebenarnya bisa generate false positive ketika ada unstructured data “Tolong sampaikan ke Pak Z, Budi tidak bisa ikut meeting” dalam dokumen. Akar permasalahannya karena query tersebut ngga bisa paham semantik "tidak bisa ikut meeting".
Kontras dengan reorganizing seluruh pengetahuan dalam sebuah perkara jadi semantic models. AI agents di @PerkaraAI ngga perlu lagi langsung hit raw data, tapi bisa pakai high-level semantic graph query seperti:
“Siapa saja <orang <tipe karyawan>> yang <bekerja di> <perusahaan X> dan <mengikuti> <event <tipe meeting> <pada tanggal Y> <bersama <orang <tipe pejabat>> Z>>?”
Semantic relationships seperti <mengikuti> sekarang bisa digunakan oleh AI models untuk menentukan predikat keterhubungan yang cocok ketika mencari data yang bisa menjawab pertanyaan pengguna.
Hasil pencariannya kemudian bisa digunakan untuk berbagai keperluan. Di perkara kami, termasuk untuk analisa perkara, bedah saksi, dan RAG untuk jawaban chat.
Terus teknis bikin semantic model gimana? Ini banyak sekali possibilitiesnya, silahkan ditanyakan ke coding agent masing2 pendekatan apa yang cocok untuk situasi yang dihadapi.
Di @PerkaraAI kami simply bikin event-driven ETL pipeline untuk reorganize raw data yang ada, tapi dalam beberapa kasus cukup pakai materialized views di PostgreSQL.
Terakhir, jangan lupa, selain cari arsitektur semantic model yang dirasa cocok untuk use casenya, yang ngga kalah pentingnya adalah rancang ontologi yang cocok seperti apa.
e.g., untuk perusahaan yang proses bisnisnya ekspor impor, tentu perlu ontologi yang berbeda dengan proses bisnis rumah sakit, atau persidangan sebuah perkara hukum.
Semoga sharing di atas membantu, dan mohon masukannya jika ada feedback atau ada cerita arsitektur lain. 🫰🏻 Di era coding agents seperti sekarang, inspirasi2 arsitektur yang datangnya dari pengalaman dan real use case sangat dibutuhkan.
Mulai besok, aku mau berbagi sedikit bocoran rahasia dapur @PerkaraAI 🕵🏻♂️
Gimana cara bikin AI yang lebih akurat & jarang halusinasi?
Apalagi untuk dokumen2 ribuan halaman, padahal ~150 halaman saja AI seperti ChatGPT/Claude/Gemini sudah halu.
Cluenya di video fitur di bawah 👇🏻
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models
“I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that.
“With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor.
“You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something.
“That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’
“I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
this is pure f*cking GOLD
how to launch your first ai agent (end-to-end walkthrough)
if i’d had this last year, my first agent would’ve been in production by tomorrow instead of spending two weeks lost in the underbrush
do it right, and the whole game gets rewritten:
En lugar de ver Netflix este domingo, dedica 1 hora a esto.
Un CURSO COMPLETO de Claude que te enseña a automatizar lo que te roba 3 horas al día.
El lunes lo agradecerás.
Ternyata bisa ngerti veritasium itu sebuah privilege.
Gue nonton veritasium sebagai teman makan siang.
Namun bagi sebagian orang itu kayak test TOEFL/IELTS dan kuliah 4 sks.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: https://t.co/OGyPb7yaYt
Anthropic engineer:
"99% of people use Claude Code like Google, and only 1% are running swarms of self-learning Claude agents
I'm running 100+ agents in a loop. I have Chief agent, PM agents - they manage the whole team"
in a 30-minute workshop, an Anthropic engineer revealed how to get max value from Claude Code at min. cost
this is worth more than another $500 vibe-coding course
watch today, then read how to build self-improving agentic systems with Fable in the article below
>sam altman: "we tried this because there were rumor on the internet"
>rumor: math professor close to solving navier-stokes
>the professor: using codex for a year
>openai: has all his logs
>checked every codex session tagged to N-S
>found the most promising one
>spun up 10,000 agents to finish the proof where he couldn't
>they got caught
>told the guy if you say anything you'll destroy your own career
>he ruins his career over this
>openai: "we did not see his work"
>also openai: "we cannot rule out that de-identified data from their usage helped improve our models"
they stole it lmao