⚡️ Google unveils Nested Learning: a brain-inspired AI model training

Hey

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers.

Every day, we identify and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.

Here’s today’s roundup:

Summary

Read time: 4 min 39 sec

Top News

Google launches Nested Learning, a new brain-inspired method to fix AI’s catastrophic forgetting

5,629 Likes

For years, deep learning has relied on a simple assumption: a model is one big optimization problem that learns from data all at once. But that simplicity has a price, catastrophic forgetting.

Train a model on new data, and it often forgets what it already knew. Google Research wants to fix that with a new paradigm called Nested Learning, unveiled with a proof-of-concept architecture named Hope.

The problem: memory that doesn’t last
Machine learning models struggle to retain knowledge over time. Even large language models, with billions of parameters, lose context when fine-tuned or retrained. This is the same issue that prevents continual learning, the ability to learn new skills without erasing old ones.

The insight: models are more like brains than we thought
Nested Learning reimagines a model not as one giant learner but as a collection of smaller learners nested inside each other.

Each one runs its own optimization loop and updates at its own speed, just like different brain regions processing information at distinct rhythms.

The breakthrough: self-modifying and context-aware learning
Google’s Hope model uses Nested Learning to structure memory across multiple timescales. Instead of treating architecture and training as separate, it unifies them into one system of layered optimization problems.

This means the model can decide which parts to update quickly and which to keep stable.

Key components

Deep Optimizers: Treats standard algorithms like Adam as memory systems that compress gradient updates.
Continuum Memory System (CMS): Extends Transformer memory into layers that update at different frequencies.
Self-Modifying Architecture: The Hope model learns to adjust its own learning rules over time.

Results
Hope outperforms Titans, Samba, and Transformers on language modeling and reasoning tasks. It achieves lower perplexity, higher accuracy, and handles long-context reasoning better, solving “Needle-in-a-Haystack” tasks with more stability and precision.

How you can use it
You can simulate nested optimization today in PyTorch or JAX by assigning different update rates to model components or optimizers. This lets you fine-tune large models while preserving old knowledge, a practical first step toward continual learning.

LEARN MORE

Free Webinar: How to Evaluate External Data Quality with Confidence

Sponsored

Small sample checks don’t reveal the truth. Every provider claims “best data” but without the right evaluation tools, you can’t know what you’re actually getting.

Join Coresignal’s free webinar on December 2 with Data Analyst Egidijus Griska to learn how to evaluate B2B data quality using concrete metrics, benchmarks, and a repeatable, neutral framework.

See how you can:

Identify freshness issues, duplicates, and deleted records early
Benchmark data quality using objective metrics not claims
Apply a 6 step evaluation framework across any dataset
Avoid costly mistakes before signing vendor contracts

The session is on December 2 at 11 AM EST for 45 minutes plus live Q&A.

SAVE YOUR SPOT

partner with us

Signals

Moonshot AI unveils K2 Thinking, an open-source low cost model rivalling GPT-5 and Claude 4.5 on reasoning and coding 9,585 Likes

Google introduces File Search Tool, a fully managed RAG system built directly into Gemini API 5,382 Likes

OpenAI debuts GPT-5-Codex-Mini, offering 4× more usage and raises ChatGPT rate limits by 50% 2,836 Likes

Google expands availability of Opal, its no-code AI app builder, to over 160 countries 2,028 Likes

xAI announces 24-hour Grok hackathon with early access to unreleased Grok models and X APIs 3,371 Likes

Speed Up Model Dataset Creation: Use Encord + GPT- 4o

Process and label robotics data automatically, validate edits, and export structured datasets for training. See the full workflow live in a focused 20-minute session.

Secure your spot now ↗

Tutorials

Automate Gemini workflows using n8n and Cloud Run setup 1,493 Likes

You’ll build an automated pipeline that runs Gemini inside n8n and Google Cloud Run. The tutorial helps you handle input data, manage environment variables, and automate reasoning tasks across serverless environments efficiently.

OpenAI walks through Codex setup for automated GitHub reviews 1,248 Likes

OpenAI shows how to set up Codex to automatically review GitHub pull requests and run reviews in the Codex CLI. They explain how Codex models detect bugs, test fixes, and improve verification across engineering workflows.

GitHub details how to code, debug, and ship faster with Copilot 871 Likes

This guide shows how GitHub Copilot now supports multi-step workflows across the editor, terminal, and GitHub. It explains how you set it up, use Agent Mode, generate tests and pull requests, review code and automate tasks to streamline your development.

Top Read

OpenAI Outlines How to Handle the Next Phase of AI Progress

OpenAI published an update on current AI trajectories and the policies needed to manage them. The post combines internal data on scaling with recommendations for governments and research institutions to prepare for the next wave of capability jumps.

Key takeaways from the post:

OpenAI estimates that AI “intelligence per dollar” is improving roughly 40× per year, faster than past hardware scaling trends.
The team expects smaller breakthroughs in 2026 followed by a larger step-change in 2028 or later, as research cycles mature.
The post calls for international coordination on model safety testing, reporting, and evaluation, similar to aviation or cybersecurity standards.
It recommends public funding for red-team exercises, AI safety research, and stress-testing to improve system resilience.
OpenAI also supports monitoring “compute concentration” to track who trains large models and at what scale.

The report frames AI progress as predictable in trend but uncertain in outcome, arguing that preparedness, not prediction, should guide policy and research investment.

KNOW MORE

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 250,000+ AI developers?

WORK WITH US

How was today’s email?

Awesome Decent Not Great

Unsubcribe

AlphaSignal
214 Barton Springs Rd, Austin, TX, USA