Armağan | AI Research

EEE student | AI Research & Deep Learning
Medical AI • Computer Vision • ML • Papers & Engineering Notes

[Armağan | AI Research @ArmaganAkcan]

1/7

Türkçe yapay zeka içerikleri çoğu zaman iki uca sıkışıyor:

yüzeysel “yeni model çıktı” haberleri
veya fazla akademik, okunması zor metinler.

Ben bu hesapta araştırma ve mühendislik sürecini göstermek istiyorum.

[Armağan | AI Research @ArmaganAkcan]

@jonegptbc Interested

[SCR @SCR01111]

All you need for System Design Interviews 📌

https://t.co/4wJ3TEhkad

[Noisy @noisyb0y1]

ANTHROPIC LEAKED A FILE THEY SPENT 4 YEARS BUILDING - 10 WORKERS BUILD A COMPANY IN 24 HOURS AND COST $16.07

OpenAI and Microsoft already replaced their engineers with this file - and save $300,000 a month.

you write the root at the start and press release at the end, and all 10 workers read the same file.

Opus 5 splits the task and judges the result, Sonnet 5 builds across 3 parallel branches, Haiku 4.5 writes the docs and scans the market.

by the time you press the button the diff has passed 2 gates - one asks does it run, the other asks should it ship at all.

the 8 build hours are 67% of the bill, while the night shift and the auditor cost 53 cents between them on the batch queue.

the auditor rewrites one line in the root - and tomorrow the ring turns on a better file.

save this and paste it into Claude Code - a day builds the machine, the market decides the number ↓

[Paul Graham @paulg]

Someone asked what I'd do if I were 17. I'd learn how to build LLMs from scratch, and then train ones as powerful as I could with whatever hardware I could get access to.

[Matt Dancho (Business Science) @mdancho84]

How To Build An AI Agent

(bookmark this for later)

[Sumanth @Sumanth_077]

Agentic-first RL training framework from NVIDIA!

Molt is a PyTorch-native RL training framework built for agentic research. Three components make up the entire stack: Ray for placement and async queues, vLLM for rollout, and NVIDIA AutoModel with FSDP2 for training. That's it.

The agentic-first design means reward is any Python you write. A mathematical grader, an LLM-as-judge, a multi-turn tool evaluator, a VLM environment. You define what "good" means inside an 'Env' or 'ChatAgent' class. No pretrained reward model required.

Two agent interfaces cover different workflows. The 'Env' interface is Gymnasium-style - the framework owns the LLM loop and your `step()` returns a reward. The 'ChatAgent' interface gives you the loop via the OpenAI or Anthropic SDK, so any external harness that already speaks those APIs works without modification.

The stack is intentionally minimal. One trainable actor, optional KL reference workers, fully async rollout and training overlap. Around 8.6K lines of RL code compared to verl's 62K. Small enough to read end-to-end and understand every gradient.

The same script that trains an 8B model scales to 1T-class MoE. DeepSeek-V3-class actors run at `ep_size 256` with TP, EP, and CP parallelism plus Adam CPU offload. No rewrite between scales.

Supported algorithms: REINFORCE, REINFORCE baseline, RLOO, GRPO, DR-GRPO, GAE (PPO), and on-policy distillation.

Key capabilities:

• Reward is any Python function - graders, LLM-as-judge, multi-turn tools, VLM environments
• Fully async rollout, training, and weight sync overlap
• Scales from 8B to 1T-class MoE with the same script
• Two agent interfaces: Env (Gymnasium-style) and ChatAgent (OpenAI/Anthropic SDK)
• REINFORCE, RLOO, GRPO, DR-GRPO, GAE, on-policy distillation

100% open source

I've shared the link in the replies!

[Mehul Mohan @mehulmpt]

if you're in tech and feel like you're falling behind:

◆ basically delete instagram/tiktok right now
◆ come to X very infrequently (15-20 min/day to catch up AI)
◆ actively put effort to learn something new every day

stakes have never been higher to go back and be a student

[david yan @dzyan01]

happy to announce that we've discovered a third axis of pre-training for the physics of GPT moment scaling laws!

[M Editor @jonegptbc]

I’m hiring dev for my startup monthly 2000$ per month, no degree required

[Oktay Arslan @oktayarslan]

2026'da Üniversiteye Yeni Yerleşen Mühendislik Öğrencilerine Tavsiyeler:

1. İlk yıl kendinize ciddi bir çalışma laboratuvarı kurun.

Evinizde ayrı bir oda ayırın.

Örneğin:
- MacBook Pro M5 Max - güçlü taşınabilir bilgisayar
- NVIDIA RTX 5090 tabanlı masaüstü bilgisayar - yapay zeka ve bilgisayarlı görü çalışmaları
- Eğitim amaçlı osiloskop: Tektronix veya Keysight
- Sinyal üreteci: Keysight
- Lehim istasyonu: Hakko
- 3B yazıcı: Prusa veya Bambu Lab
- Mikrodenetleyici: STM32, ESP32, Arduino
- Geliştirme bilgisayarı: Raspberry Pi
- Robotik: NVIDIA Jetson + çeşitli algılayıcılar + motorlar

İlk yıl için 10.000 - 30.000 dolar seviyesinde bir laboratuvar kurmak gayet makul.

Amaç oyuncak toplamak değil.

Dört yıl boyunca sürekli bir şeyler üretmek.

[Machine Learning (ML) Papers @Memoirs]

Does Mapping Non-Maximal Probabilities to GMM Components Matter for S-JEPA Encoder Representations?

Wenxuan He, Yunpeng Li, Shan Liang
https://t.co/at0RC2lopR [𝚌𝚜.𝙻𝙶 𝚌𝚜.𝚂𝙳]

[Fran Pradas | Ingeniero de IA @franpradasAI]

El rol mejor pagado en 2026 es INGENIERO DE IA.

Si no tuviera formación, y quisiera conseguir trabajo en esto

Haría esto en 6 meses:

[16VC @16vchq]

16VC Ambassador Program

Building founder communities, locally.

Interested in becoming a 16VC Ambassador?

Comment below — we’ll reach out.

[Armağan | AI Research @ArmaganAkcan]

@defyneric Interesred

[eric @defyneric]

＿＿＿＿＿＿
| We’re hiring |
|＿＿＿＿＿＿|
\ (•̀ᴗ•́) /
\ /
| |
| |
/ \

Open roles at Flex:

→ Software Engineers
→ Product Designers
→ Growth
→ B2B Partnerships
→ Customer Success
→ Compliance & Risk
→ Credit & Underwriting
→ Operations
→ Content & Social
→ Brand & Creative
→ Sales

& more.

Think you can help build the future of business banking?

Comment the role you want, what you’d build or improve at Flex, and your best work.

Comments preferred, but DMs are open too

[Armağan | AI Research @ArmaganAkcan]

@RemotejobWtessy I'm interested

[REMOTE JOB WITH TESSY @RemotejobWtessy]

If you don’t mind working overnight hours, this might be worth checking out. The schedule may not suit everyone, which could mean fewer applicants competing for the role.

I found an overnight remote job paying up to $30/hour USD. 👇

[Armağan | AI Research @ArmaganAkcan]

@Maaztwts in

[Maaz (agentic arc) @Maaztwts]

hackathon soon

$1000+ cash prize

reply “in” if you want to join

[Lead Her Logic @LeadHerLogic]

Can you stay awake from 1 AM - 4 AM every weekdays?

I have another remote role for you that pays $100/hr!!!