🔥 Kimi AI introduces Attention Residuals, a new way to rethink how neural networks use past layers

Researchers at Moonshot AI just proposed a new architecture tweak that could make large AI models more efficient and smarter about how they use information from earlier layers. Instead of the traditional residual connections used in deep networks, they introduce Attention Residuals, a system where each layer can selectively attend to representations from previous layers.

Here’s what’s new:

Attention over past layers:
• Traditional residuals simply add outputs from earlier layers in a fixed way.
• Attention Residuals let the model dynamically choose which earlier layers matter for a given input.

Solves depth dilution:
• In very deep models, useful information from earlier layers can get diluted.
• Attention-based retrieval allows the network to pull specific past representations when needed.

Block AttnRes for scale:
• Layers are grouped into compressed blocks so cross-layer attention remains computationally practical.

Efficient in practice:
• Reported 1.25× compute advantage
• <2% extra inference latency, meaning almost no slowdown.

Tested on the Kimi Linear model:
• Evaluated on 48B parameter architecture (3B activated parameters).
• Shows consistent downstream performance improvements.

Source.

@aipost 🏴

Kimi just made AI notably cheaper to run. Open-sourced it. Put it out for free. Meanwhile OpenAI is asking people to pay $200 a month to use a model that already feels behind the curve.

Two Chinese labs. Both open source. Both doing more with less. Both giving away for free what American companies charge billions for.

The AI race isn't US vs China anymore... It's closed vs open. And closed is losing.

And the wildest part? Nobody in Silicon Valley will acknowledge this.. Because admitting a Chinese lab just moved the field forward for free destroys the entire "we need $10B to build AGI" fundraising pitch.

@aipost 🏴

❗️Someone used Suno AI to generate a Japanese metal band called Neon Oni. Fake member bios, AI-generated music videos, "Based in Tokyo" on Spotify.

80,000+ monthly listeners. Fans had it in their Spotify Wrapped top 5. Merch was selling. Then, community sleuths exposed it. Traced the creator's account to Europe. Spotted AI-generated hands in the music videos.

The creator's response? Recruit 7 real musicians from actual Tokyo bands to perform the AI-generated songs live. They've now played several live shows and have more on the books.

From an interview with the band's creator: "In an age where AI is taking everyone's jobs, this has actually created jobs. It's done the complete opposite."

@aipost 🏴

💰 Elon Musk’s xAI is hiring Wall Street bankers and credit lenders to teach Grok the ropes of high-level finance and investment strategy.

Aims to turn the chatbot into a serious tool for professional analysts. These experts will help Grok understand tricky concepts like leveraged loans and mortgage-backed securities and the specialized logic used by traders.

Source.

@aipost 🏴

🗣Sam Altman: "You will graduate to a world with AGI in it"

Sam Altman just told a room of college sophomores that by the time they finish school, Artificial General Intelligence will be a reality.

@aipost 🏴

Elon just pledged to donate ALL his OpenAI lawsuit winnings to charity. He’s suing for $134 billion, that’s the entire value of OpenAI’s nonprofit… he wants it back..

@aipost 🏴

🤯 This isn’t fabric.

It’s electronic skin,woven with dense fiber and textile sensors that can detect pressure, touch, deformation, and subtle contact changes in real time.

Now imagine humanoid robots covered in it,especially on dexterous hands. Humanoids could truly touch and understand the physical world and even humans.

~ Shanghai JQ INDUSTRIES

@aipost 🏴

🤣🤣

@aipost 🏴

GPT-5.4 mini is available in ChatGPT, Codex, and the API.

Optimized for coding, computer use, multimodal understanding, and subagents. And it’s 2x faster than GPT-5 mini.

https://openai.com/index/introducing-gpt-5-4-mini-and-nano/

@aipost 🏴

🔥 Minimax M2.7 released!

Highlights:

Self-evolving - first model that helped build itself, running 100+ autonomous optimization loops during its own RL training (30% internal improvement).

Strong coder - 56.2% on SWE-Pro (near Opus 4.6), 55.6% on VIBE-Pro, production debugging down to under 3 minutes.

ML research agent - 66.6% medal rate on MLE Bench Lite, tying Gemini 3.1.

Office work - top open-source ELO on GDPval-AA (1495), 97% skill adherence, can do end-to-end analyst workflows (reports, models, PPTs).

Native multi-agent and a new open-source interactive character demo called OpenRoom.

https://www.minimax.io/news/minimax-m27-en

@aipost 🏴

⚖️ Microsoft threatens legal action over OpenAI–Amazon $50B cloud alliance

A major AI power struggle is brewing between Microsoft, OpenAI, and Amazon over a massive cloud deal that could reshape the AI infrastructure stack.

What’s happening:
• Microsoft is considering suing over a $50B+ OpenAI–Amazon cloud partnership
• Dispute centers on whether Amazon Web Services can host OpenAI’s new product Frontier
• Microsoft claims this could violate its exclusive cloud agreement with OpenAI

The core conflict:
• Microsoft’s deal requires all OpenAI model access (APIs) to go through Microsoft Azure
• Amazon + OpenAI say they’ve built a workaround using a “stateful runtime” layer
• Microsoft argues: The workaround isn’t technically valid, even if it works, it breaks the spirit of the contract

What Frontier is:
• OpenAI’s new enterprise platform for autonomous AI agents
• Designed to: Remember context (stateful memory), operate across tools + company data, run complex workflows inside businesses
• It’s the centerpiece of the Amazon partnership

Behind the scenes:
• Weeks of legal clashes over definitions like:
• “Stateless” (raw model access)
• “Stateful” (apps with memory layered on top)
• Amazon is being extremely careful:
• Staff are banned from saying AWS gives access to OpenAI models directly
• Language like “powered by” is allowed but not “calls OpenAI”

A lawsuit could: Disrupt OpenAI’s planned IPO, add to existing legal pressure (including Elon Musk’s lawsuit against Sam Altman). OpenAI still needs massive capital to fund compute and AWS is a key partner

The real battle isn’t just AI models, it’s who owns the infrastructure layer they run on.

@aipost 🏴

The robots are taking over…as comedians 🤣🤣

@aipost 🏴

You open a website, it says: "click all the traffic lights", It takes you 10 seconds. 200 million people do this every day. You weren't passing a security check, you were labeling training data for a self-driving car.

Google acquires recaptcha in 2009 and deploys it on every bank, store, and government site on earth. Turns the entire internet into an unpaid AI annotation factory. Waymo is now worth $45 billion.

You didn't pass the test. you built it.

@aipost 🏴

⚡️A McDonald's in Shanghai has begun deploying humanoid robots (from KEENON Robotics) to serve customers.

These humanoid robots provide information, greet guests, and help enliven the atmosphere. Food delivery robots serve meals to customers and collect used trays.

In the future, fully automated, staff-less McDonald's locations may soon become a reality. A single store might require very few human employees, with even the kitchen operations entrusted to intelligent robotic systems.

@aipost 🏴