Midnight ./

asdfghjk

[Midnight ./ @zkchrls]

Imagine a world where intelligence isn't locked in servers, but shared across people?

@Gradient_HQ Parallax is open-sourcing its decentralized LLM engine uniting GPUs, Apple chips, and desktops into one global swarm.

No more gated clusters or cloud monopolies. Just a living network of people powering intelligence together.
#Gradient #Parallax

[Midnight ./ @zkchrls]

It's time to learn mandarin.

[rw ./ @gradientintern]

Kimi K3 is near on par with Fable, GLM 5.3 would be on pair too if the list had it.

Most general work users that are not strictly hardcore coders doesn’t see much of a difference between using Claude Opus or one of the open source models.

Backstopping the closed labs is building out infrastructure for AI, but it can get risky as models are now commodities and they need applications or services to drive more value.

[Hexx ./ @HexxRL]

The post training bottleneck is no longer raw text, it’s high quality trajectory experience paired with reliable rewards.

Reinforcement Learning have become high capital infrastructure:
• $200–$2,000 per task
• ~$300K for complex environments
• 4–5x markup for exclusive datasets

A estimated $250M budget can erodes fast when task creation, rollout compute, and evaluation compound. Expand task volume without enough compute or rely on weak graders and models learn to game the system rather than master the job.

Overcoming this wall requires scaling rollout throughput without proportional cost inflation.

@Gradient_HQ Echo framework solves this by leveraging asynchronous distributed RL on volatile compute, boosting research throughput by 10x and making post training scaling laws economically viable.

[Midnight ./ @zkchrls]

Every gamer's dream.

[rw ./ @gradientintern]

xAI's Grok 4.6 is now live on @commonstack_ai

Grok 4.6 excels in long running agentic tasks, particularly in turning ambitious ideas into working visual and interactive apps over multistep workflows. It also leads its benchmark class on legal reasoning evaluation metrics, outperforming top competitors on the Harvey LAB benchmark.

Try now: https://t.co/P3z2rPLjX3

[Midnight ./ @zkchrls]

stackoverflow + indian youtube tutorials

[Hexx ./ @HexxRL]

Stack Overflow has gone from a peak of 207k questions in March 2014, down to 1.4k in July 2026, or -99.3% loss

they sold the top in 2021 for $1.8 billion dollars

[Gradient @Gradient_HQ]

Proud to share MERA: joint work from a group of brilliant researchers including our very own.

It evolves small models with SkillBook + routing + adapters so they take on more of the agent workload over time.

Paper + code 👇

[Midnight ./ @zkchrls]

A real gamer's GPU before the engineering shift. He understands.

[Midnight ./ @zkchrls]

@flowerlang20 cascal

[Midnight ./ @zkchrls]

if a single model can already do so much on its own, imagine what would happen if multiple models joined forces and acted as a team of specialists.

just think about it. ☕️

[Gradient @Gradient_HQ]

Agents are impressive. Multi-agent systems are where things get interesting.

As AI systems evolve from single agents into networks of specialists, a new question follows:

How do agents decide who does what?

That’s a problem we’ve been exploring with Symphony.

Symphony brought multi-agent collaboration across heterogeneous environments. To take that further, we released Symphony-Coord, where coordination and specialization adapt through task context, performance, and feedback.

https://t.co/7qdcmO40RU

[Midnight ./ @zkchrls]

This chinese chip could break into big 3 memory club. some known brands already using cxmt chips for their ram kit.

[Midnight ./ @zkchrls]

If only I knew.

[Midnight ./ @zkchrls]

Life is tough man.

[Midnight ./ @zkchrls]

I probably need to download RAM.

[Midnight ./ @zkchrls]

Snoop is so proud of you all. 🤣

[Kalshi @Kalshi]

JUST IN: Americans now smoke more weed than cigarettes

[Midnight ./ @zkchrls]

Finally, a RX 9000 series for us slapsoil.

[Midnight ./ @zkchrls]

Progress is really fast when nothing is holding you back.

[Megatron @Megatron_ron]

NEW: 🇨🇳 China's nuclear fusion research has completed a massive magnet weighing 582 tons.

The goal is an "unlimited energy" artificial sun.

They plan to start the first power generation by 2030.

A safe, CO2-zero ultimate energy race.

[Midnight ./ @zkchrls]

GG

[Cointelegraph @Cointelegraph]

🚨 UPDATE: U.S. data center power demand is projected to reach 194 gigawatts by 2035, enough to consume 20% of the nation's electricity, according to BloombergNEF.

[rw ./ @gradientintern]

Claude Opus 5, the latest frontier from Anthropic is now on @commonstack_ai

It performs at SOTA for many coding and knowledge evaluations while coming in half the price of Fable 5.

Build now with the best in class production execution.

[Midnight ./ @zkchrls]

Sixers fan since july 2026