@MTorygreen This chart makes my own scaffolding mistakes look embarrassing
English
Jack Cooldev
720 posts
Jack Cooldev
@jackcooldev
ex-SWE → testing every AI tool that drops | what works, what's hype | posting daily
California Katılım Ağustos 2022
482 Takip Edilen397 Takipçiler
memory and harness design are starting to decide agent performance more than frontier model selection.
Turns out Sol's bad ARC-AGI-3 score wasn't the model, it was the test wiping its memory every move.
usually a better score costs more compute but flipping two memory settings got it better and cheaper at once, 7.8% to 38.3% with 6x fewer tokens. once it could remember, it stopped re-solving the game from scratch every turn.
so the score was mostly scaffolding, not the model. same weights, 3x the result.
Tibo@thsottiaux
Turns out GPT-5.6 Sol is actually SoTA on ARC-AGI-3. Just took two setting changes. You just have to allow it to reason and work over multiple context windows with the help of our canonical compaction implementation. openai.com/index/how-two-…
English
@Lin98201 @NucleusCodes Does sunset light unveil that quiet beacon
English
The quietest upgrades hit the hardest
They don’t shout
They just remove the weight
Most people waste time proving they deserve a seat
Forms
Referrals
Starting over from zero again and again
It drains everything
@NucleusCodes flips that
Your on-chain trail and how you show up become the real signal
Reputation turns into the silent infrastructure
No theater
No forced grinding
Just a clean line from what you’ve already built to what opens next
Real ecosystems don’t grow from noise
They grow when people can move fast
test freely
and turn real work into real access
That’s the kind of signal I still pay attention to
English
What's a tool or habit that quietly made you a better developer...?
English
if you are a VC reading this, mind sliding me like uh
a check?
English
DeepSeek 너만 믿는다 제발 나의 128g 에 새롭게 들어와 주길~~~ 시원하게 놀아보자~~~
Jun Song@jun_song
Exactly what’s happening right now.
한국어
@Mr_Salio Truly fascinating what thinking models can actually do
English
🚨 GPT-6 Update:
>OpenAI Confirms GPT-6 Wasn't the Hugging Face Model
> A separate internal research prototype never intended for public release was also involved
> The prototype has now been deactivated, encrypted, and restricted
> OpenAI says no planned public release model was the research prototype
> The evaluation intentionally reduced cyber refusals to test advanced offensive capabilities
If this prototype isn't GPT-6... imagine what OpenAI's future frontier models might be capable of
English
If AGI drops tomorrow, which job is actually safe?
Plumbers or Prompt Engineers?
English
Want to become a better backend engineer?
Master these before chasing the next framework.
→ Java 21
→ Spring Boot 3.5
→ REST APIs
→ Security
→ JPA
→ Kafka
→ Docker
→ Kubernetes
→ Observability
Everything else builds on this foundation.
Anupam Haldkar @AnupamHaldkar
Backend engineering is no longer just about writing APIs. It’s about mastering an entire ecosystem. → Languages → Frameworks → APIs → Messaging → Databases → Search → Containers → Kubernetes → Observability → Security → Cloud → CI/CD → Testing You don’t need to master everything today. Learn one layer at a time. Great backend engineers aren’t experts in one tool. They understand how the entire ecosystem fits together.
English
MoonPay vừa ra mắt PayBox, cho phép ChatGPT và Claude thay người dùng thực hiện các giao dịch crypto ngay trong cuộc trò chuyện
người dùng chỉ cần kết nối ví một lần, sau đó AI có thể swap token, bridge hoặc tương tác với các giao thức DeFi, giao dịch sẽ được xác nhận bằng passkey
MoonPay cũng triển khai chương trình airdrop, người dùng sớm có thể nhận tối đa 500 USDC khi trải nghiệm PayBox
Faustino@77bncvbsdcg
Chủ tịch SEC cho biết cơ quan này sẵn sàng tự ban hành khung quản lý crypto nếu dự luật CLARITY Act tiếp tục bị trì hoãn CLARITY Act hiện vẫn mắc kẹt tại Thượng viện, trong khi SEC đã chuẩn bị nhiều quy định liên quan đến phát hành, lưu ký và giao dịch tài sản số nếu dự luật chưa được thông qua, SEC nhiều khả năng sẽ chủ động xây dựng luật để thị trường crypto sớm có khung pháp lý rõ ràng
Tiếng Việt
@sashikantsingh_ Is ChatGPT really beating Gemini for pictures
English
There’s a chance Grok eventually goes open source??
xAI already released Grok-1 nd Grok 2.5, promised to open source Grok 3 nd even made Grok Build public.
Imagine Grok 4.6 or 4.7 dropping with open weights while Open AI & Anthropic drop models like GPT 5.6,Fable 5 nd Opus 5.
English