/4 Anthropic demonstrates subliminal learning where models inherit behavior from seemingly unrelated training data. Anthropic co-authors new research on subliminal learning, published in Nature. You train models on outputs from other models, assuming only visible content
@alphasignalai
-
Cursor and NVIDIA Build Multi-Agent System Optimizing CUDA Kernels 38%
By
–
/3 Cursor builds multi-agent system that optimizes CUDA kernels with 38% average speedup. Cursor, working with NVIDIA, develops a multi-agent system that writes and optimizes CUDA kernels automatically. In a three-week run across 235 real problems, the system improves
-
Google Proposes Memory Caching to Extend RNN Long-Context Handling
By
–
/2 Google gives RNNs growing memory so they can handle longer inputs without large compute increases. Google Research proposes Memory Caching (MC), a method that extends RNNs with memory that grows over time. Transformers solve long-context tasks using KV-cache but require
-
Meta AI Proposes Neural Computer Where AI Becomes the Hardware
By
–
The day AI replaces the computer wasn't supposed to come this fast.
— AlphaSignal AI (@AlphaSignalAI) 18 avril 2026
Right now, AI uses computers as tools.
This paper asks a different question: What if AI became the computer itself?
Neural Computer is a new paper from Meta AI.
It proposes a machine where computation,… pic.twitter.com/1AjqEC9LprThe day AI replaces the computer wasn't supposed to come this fast. Right now, AI uses computers as tools. This paper asks a different question: What if AI became the computer itself? Neural Computer is a new paper from Meta AI. It proposes a machine where computation,
-
Multica Open-Source Platform Turns Coding Agents Into AI Teammates
By
–
Your next 10 hires might not be human.
— AlphaSignal AI (@AlphaSignalAI) 17 avril 2026
Multica is a platform that turns coding agents into actual teammates.
It's the open-source version of Claude Managed Agents.
You assign issues to an agent the same way you'd assign to a colleague.
They pick up the work. Write code. Post… pic.twitter.com/mS2G3utKL8Your next 10 hires might not be human. Multica is a platform that turns coding agents into actual teammates. It's the open-source version of Claude Managed Agents. You assign issues to an agent the same way you'd assign to a colleague. They pick up the work. Write code. Post
-
In-Place Test-Time Training Lets LLMs Update Weights While Running
By
–
Someone just made LLM weights absorb new knowledge without retraining. Researchers built In-Place Test-Time Training. It lets a model update its own weights while running. Not just during training. The trick: repurpose a matrix already inside every MLP block as "fast
-
Generate Architecture Diagrams for GitHub Repos Using Claude Skill
By
–
This is so cool.
— AlphaSignal AI (@AlphaSignalAI) 17 avril 2026
Generate sleek architecture diagrams for any GitHub repo or local project in seconds.
It’s a single Markdown skill file that you can import and use right away in Claude. https://t.co/6zhDZPcGEf pic.twitter.com/OCmyoXjg5XThis is so cool. Generate sleek architecture diagrams for any GitHub repo or local project in seconds. It’s a single Markdown skill file that you can import and use right away in Claude.
-
Obliteratus Tool Removes Safety Refusals From Open-Weight LLMs
By
–
Someone built a tool that deletes censorship from any open-weight LLM with a single click.
— AlphaSignal AI (@AlphaSignalAI) 16 avril 2026
Most open-source models ship with built-in refusal.
They reject prompts even when the use case is legitimate.
Research, red-teaming, creative writing.
Obliteratus is a toolkit that… pic.twitter.com/KnNohJirS1Someone built a tool that deletes censorship from any open-weight LLM with a single click. Most open-source models ship with built-in refusal. They reject prompts even when the use case is legitimate. Research, red-teaming, creative writing. Obliteratus is a toolkit that
-
ASI-Evolve: Open-Source AI Framework That Designs Better AI Autonomously
By
–
Shanghai researchers built an AI that designs better AI than humans can. Most AI research still needs humans at every step. Read papers, form ideas, run experiments, repeat. ASI-Evolve is an open-source framework that does this loop entirely on its own. It works in four
-
Shannon Lite Open-Source AI Pentester Finds and Exploits Web App Vulnerabilities
By
–
Someone built an AI that hacks your web app before hackers do.
— AlphaSignal AI (@AlphaSignalAI) 15 avril 2026
Shannon Lite is an open-source AI pentester.
It reads your code, finds weak spots, and launches real attacks.
Most teams run a security test once a year. Code ships every week.
Months of untested code sit in… pic.twitter.com/DvIuKqYCFUSomeone built an AI that hacks your web app before hackers do. Shannon Lite is an open-source AI pentester. It reads your code, finds weak spots, and launches real attacks. Most teams run a security test once a year. Code ships every week. Months of untested code sit in