Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifying connectivity patterns. While yielding substantial performance gains, this diversification fundamentally compromises the identity mapping property intrinsic to the residual connection, which causes severe training instability and restricted scalability, and additionally incurs notable memory access overhead. To address these challenges, we propose Manifold-Constrained Hyper-Connections (mHC), a general framework that projects the residual connection space of HC onto a specific manifold to restore the identity mapping property, while incorporating rigorous infrastructure optimization to ensure efficiency. Empirical experiments demonstrate that mHC is effective for training at scale, offering tangible performance improvements and superior scalability. We anticipate that mHC, as a flexible and practical extension of HC, will contribute to a deeper understanding of topological architecture design and suggest promising directions for the evolution of foundational models. — Read More
Author Archives: Rick's Cafe AI
2025: The year in LLMs
This is the third in my annual series reviewing everything that happened in the LLM space over the past 12 months. For previous years see Stuff we figured out about AI in 2023 and Things we learned about LLMs in 2024.
It’s been a year filled with a lot of different trends. — Read More
Cybersecurity Changes I Expect in 2026
It becomes very clear that the primary security question for a company is how good their attackers’ ai is vs. their own.
— ISOs increasingly realize that there is no way to scale their human team to deal with how constant, continuous, and increasingly effective their attackers are becoming at attacking them
— It becomes a competition with how fast you can perform asset management, attack surface management, and vulnerability management on your company, but especially on your perimeter (which includes email and phishing/social engineering)
Read More
Planetary-Scale Deep Reasoning: Building Our Final Presidential Daily Brief Prompt & Comparing Gemini 3/2.5 Pro/Flash ASR/TOC
Over the last few days we have been exploring having Gemini 3 Pro “watch” an entire day of television news from a given channel from across the world and write a deeply reasoned and researched intelligence-style report that looks across all of that coverage and teases out the overarching themes, narratives, implications and future impacts of the day’s events. Yesterday we had Gemini 3 Pro interactively improve its own prompt to generate a final “ultimate” prompt to write a Presidential Daily Brief (PDB)-style intelligence report from a day’s broadcast transcripts. Today we’ll add a few final refinements and then demonstrate our new prompt on a single day of a Russian television news channel across Gemini 3 Pro, Gemini 3 Flash, Gemini 2.5 Pro and Gemini 2.5 Flash Thinking using both the full-day Chirp 1 ASR transcripts and a preprocessed story table of contents. No data was used to train or tune any model. — Read More
China Just Pulled Its Own Manhattan Project and No One Saw It Coming
Or: The West banned the machines. China hired the machinists. Sometimes plans just do not go how you planned them. Ironically, I had been writing this article for a month now, and all research pointed at China being way too far behind. Well…
December 2025. Reuters reveals that China completed an operational EUV lithography prototype in a high-security Shenzhen facility. Not through reverse engineering captured ASML machines. Not through some breakthrough in domestic optics manufacturing. Through something far simpler.
They recruited the humans who knew how to build them. — Read More
Welcome to the Machine, a guide to building infra software for AI agents
… I happen to have a bit of time these days, so I decided to write down a question I’ve been repeatedly thinking about lately.
The main reason is that I’ve been seeing one trend with increasing clarity: the primary users of infrastructure software are rapidly shifting from developers (humans) to AI agents.
… Because of this, I’ve started to rethink the problem from a more ontological perspective: when the core users of foundational software are no longer humans but AI, what essential characteristics should such software have? — Read More
AI Slop Report: The Global Rise of Low-Quality AI Videos
Kapwing’s new research shows that 21-33% of YouTube’s feed may consist of AI slop or brainrot videos. But which countries and channels are achieving the greatest reach — and how much money might they make? We analyzed social data to find out.
As the debate over the creative and ethical value of using AI to generate video rages on, users are getting interesting results out of the machine, and artist-led AI content is gaining respect in some areas. Top film schools now offer courses on the use and ethics of AI in film production, and the world’s best-known brands are utilizing AI in their creative process — albeit with mixed results.
Sadly, others are gaming the novelty of AI’s prompt-and-go content, using these engines to churn out vast quantities of AI “slop” — the “spam” of the video-first age. — Read More
What changes did AI actually bring to scientists this year?
In this wave of artificial intelligence, it’s easy to be swept up in grand narratives: computing power, models, the scale of parameters, disruption, and replacement. But the changes truly worth documenting often happen out of sight: how AI is used, how it enters daily life, and how it changes the way people work.
Last week, The Intellectual and Doubao jointly launched: “A Story Collection | How Were You ‘Amazed’ by AI This Year?”, not asking “how powerful is AI,” but a more specific question: When AI enters your work and life, what exactly does it change?
…Perhaps what is truly worth documenting is not what AI can do, but how researchers, after its intervention, re-understand their work, judgments, and responsibilities—what tasks can be automated, and what problems must still be decided by humans.
These scattered and specific experiences constitute the first batch of “field notes” of AI entering the scientific field. They may not be complete, but they are sufficiently honest. — Read More
How to Land a $500K AI PM Job at OpenAI (The 2026 Playbook)
… The talent shortage is brutal. Every company needs AI PMs. Few people have the skills.
OpenAI, Anthropic, Google DeepMind, and Meta all have open AI PM roles. They can’t fill them fast enough.
The hiring bar is high. You need product sense, technical depth, and hands-on AI experience. Most PMs have one or two. You need all three.
… The gap between supply and demand means comp packages keep climbing. Base salary plus equity plus signing bonuses. $500K is common. $700K+ for senior roles.
The AI PM job market dynamics show why this won’t change soon. — Read More
When AI Loses the Plot: How to Reset and Refocus Your Conversations
We’ve all been there. You’re deep in a conversation with your AI assistant, working through a complex problem, when suddenly it starts giving you responses that make no sense. The more you try to correct it, the worse it gets. Each new prompt seems to push the AI further from understanding what you actually need.
This frustrating phenomenon happens because AI models can lose track of context in lengthy conversations, especially when there have been multiple corrections or clarifications. The good news? There’s a simple yet powerful technique to get things back on track.
Full disclosure: I’ve been using a form of this forever, but I didn’t see it so succinctly explained and put together until I visited this Reddit thread from another user having the same problem. The idea and ensuing discussion are the basis for this post. Check out the full thread here. — Read More
Rick's Cafe AI 