Mamba's memory problem - a must-read if you are really into LLMs optimization research.
0
1
4
2.2K
3
ICLR accepted nearly 5,000 papers this year. This must-read article highlights 14 that focus on real-world multi-agent production challenges and shares solutions.
0
2
4
694
2
I've been recently working on Royo, an autonomous mobile agent, which can operate my phone on my behalf using Gemma 3n, a local model by
@Google
built on top of Gallery by Google. Here is a thread of a few demos
1. Operates mobile apps and navigates https://t.co/IavBKRB1Q3
built on top of Gallery by Google. Here is a thread of a few demos
1. Operates mobile apps and navigates https://t.co/IavBKRB1Q3
3
1
12
3.4K
2
Are you in a dilemma about your next agentic project whether to choose an enterprise solution or an open-source framework? 🤔
We've compiled a detailed comparison👇🏻. We really hope it helps you. If you spot any corrections, pls let us know. https://t.co/Tz4uDwJPs7
We've compiled a detailed comparison👇🏻. We really hope it helps you. If you spot any corrections, pls let us know. https://t.co/Tz4uDwJPs7
1
5
16
4.1K
20
Don't miss today's newsletter analyzing LLMs research papers from Jan 17th to 23rd. It contains practical tips to use this week's research into your project!!👇🏻
https://t.co/CLCvoI9QkG https://t.co/9uaeVskiO8
https://t.co/CLCvoI9QkG https://t.co/9uaeVskiO8
0
1
12
2.2K
4
When released in 2017, Transformer couldn't process a single long article. Now we have million-token context windows. Here's how we got here in just 8 years.
1
5
12
2.7K
6
I’m building a Memory-First Context Compiler:
🧠 Neurons, not Logs: Memories are stored as atomic beliefs with confidence scores.
📉 Fixed Costs: Your prompt stays small (5k tokens) while your history grows to billions.
⚡ Zero Latency: No more waiting 10s for the LLM to read 1M
🧠 Neurons, not Logs: Memories are stored as atomic beliefs with confidence scores.
📉 Fixed Costs: Your prompt stays small (5k tokens) while your history grows to billions.
⚡ Zero Latency: No more waiting 10s for the LLM to read 1M
1
1
7
2.7K
1
Intelligence Needs Context
1
0
5
1.1K
1
what do you think??
0
0
5
345
1
Breaking the LLM Context Limit Barrier! Ever hit that pesky context window wall with your AI models? What if I told you we can engineer "infinite" context by building a intelligent system
No more forgetting mid-convo! This scales LLMs to handle book-length inputs or endless
No more forgetting mid-convo! This scales LLMs to handle book-length inputs or endless
0
1
4
1.4K
0
I started respecting AI researchers more after reading this article!! 👇👇👇
0
1
6
2.0K
5
Love this article!
If you are working on an agent, then don't miss this week's papers, like MAXS and ThinkAct covered here 👇🏻👇🏻👇🏻
If you are working on an agent, then don't miss this week's papers, like MAXS and ThinkAct covered here 👇🏻👇🏻👇🏻
1
2
10
2.3K
10
Sending our weekly newsletter tomorrow morning
👉 Get it in your inbox for free by subscribing (if you haven't already): https://t.co/XZbzCwpf5A https://t.co/Duy4kg7K6G
👉 Get it in your inbox for free by subscribing (if you haven't already): https://t.co/XZbzCwpf5A https://t.co/Duy4kg7K6G
0
0
4
195
0
Claude isn’t winning despite being expensive,
Not because of the Opus 4.5 hype,
Not because of branding.
It's winning because it’s just better!
Still: Anthropic >> OpenAI.
And others not even close.
Inside our org:
no one panicked,
No long debates,
No cost optimization https://t.co/IMy5XkiF0l
Not because of the Opus 4.5 hype,
Not because of branding.
It's winning because it’s just better!
Still: Anthropic >> OpenAI.
And others not even close.
Inside our org:
no one panicked,
No long debates,
No cost optimization https://t.co/IMy5XkiF0l
3
1
13
2.8K
1
Here's how "Claude code skills" can auto update your agents with customer interaction trace analysis ?
An autopilot for agents!
https://t.co/zRiDoMiuG7
An autopilot for agents!
https://t.co/zRiDoMiuG7
0
4
7
1.2K
3
Self-Adapting Language Models (SEAL)
From MIT (Massachusetts Institute of Technology), including CSAIL’s Improbable AI Lab.
Problem: powerful but static LLMs
The paper argues today’s LLMs do not natively adapt their weights when faced with new tasks, new knowledge, or new https://t.co/vnwebexFaV
From MIT (Massachusetts Institute of Technology), including CSAIL’s Improbable AI Lab.
Problem: powerful but static LLMs
The paper argues today’s LLMs do not natively adapt their weights when faced with new tasks, new knowledge, or new https://t.co/vnwebexFaV
2
1
6
2.4K
4
Google launched Universal Commerce Protocol (UCP) for Agentic Shopping
The Universal Commerce Protocol (UCP) is an open-source standard designed to power the next generation of agentic commerce.
By establishing a common language and functional primitives, UCP enables seamless https://t.co/AUtCksk6e7
The Universal Commerce Protocol (UCP) is an open-source standard designed to power the next generation of agentic commerce.
By establishing a common language and functional primitives, UCP enables seamless https://t.co/AUtCksk6e7
3
2
5
894
1
Current LLMs are geniuses with Alzheimer's.
They solve PhD-level problems... then instantly forget you exist.
I'm building the persistent memory base that could fix this → the missing layer for real AGI.
Change my mind: context windows are a dead end. Fight me. 🧠🤖 https://t.co/g8SukeUiip
They solve PhD-level problems... then instantly forget you exist.
I'm building the persistent memory base that could fix this → the missing layer for real AGI.
Change my mind: context windows are a dead end. Fight me. 🧠🤖 https://t.co/g8SukeUiip
2
1
5
949
0
Weekly newsletter: covering important research papers related to LLMs Research, published this week (Jan 2-9, 2026).
Get it delivered to you for free by subscribing to the newsletter: https://t.co/8pqQtWGdqA https://t.co/qMDWRh9dtK
Get it delivered to you for free by subscribing to the newsletter: https://t.co/8pqQtWGdqA https://t.co/qMDWRh9dtK
1
2
5
2.0K
1
If your LLM evals are multiple-choice, you’re measuring recall, not capability.
Real failures show up in:
•long-horizon tasks
•ambiguous specs
•partial success cases
Offline benchmarks lie.
Production logs tell the truth.
Real failures show up in:
•long-horizon tasks
•ambiguous specs
•partial success cases
Offline benchmarks lie.
Production logs tell the truth.
0
1
6
185
0
Founder @llmsresearch and https://t.co/dNtzHI7fNK All in on LLMs
616 Followers
5 Contributions
Research group analyzing Large language model (LLMs) papers and distributing a daily newsletter. Covers trends, categorization & summary of research papers
1.0K Followers
5 Contributions
Cross-Domain AI Architect | Building Mumbrane: Persistent LLM Memory that ends amnesia & slashes API costs by 30 to 60%
30 Followers
4 Contributions
GSOC '25 Mentor, C4GT & MSOC Mentor | GSoC '24 Contributor @ The Mifos Initiative | Research Intern @ UC Berkeley
159 Followers
1 Contributions
NLP Consultant & Researcher. AIxFunda newsletter for the latest LLM, Agents and RAG updates.
3.6K Followers
1 Contributions
12.8K
Total Members
+ 0
24h Growth
+ 4
7d Growth
Date Members Change
Feb 10, 2026 12.8K +0
Feb 9, 2026 12.8K +1
Feb 8, 2026 12.7K +2
Feb 7, 2026 12.7K +0
Feb 6, 2026 12.7K +0
Feb 5, 2026 12.7K +1
Feb 4, 2026 12.7K +0
Feb 3, 2026 12.7K -1
Feb 2, 2026 12.7K +1
Feb 1, 2026 12.7K +0
Jan 31, 2026 12.7K +1
Jan 30, 2026 12.7K -2
Jan 29, 2026 12.7K -3
Jan 28, 2026 12.8K —
No reviews yet
Be the first to share your experience!
Share Your Experience
Sign in with X to leave a review and help others discover great communities
Login with XLoading...
Come join the active community discussing the latest LLMs research papers and innovative applications of #LLMs
Community Rules
Be kind and respectful.
Keep posts on topic.
Explore and share.
Do not spam
Do not share promotional tweets or comments to spam the community. You will be immediate terminated from the community.
Don't just simply repost it
Add some thoughts on repost