2. MEM1 This work introduces MEM1, an RL framework for training language agents that operate efficiently over long-horizon, multi-turn tasks by learning to consolidate memory and reasoning into a compact internal state.
LLMS
-
Zuckerberg Poaches OpenAI Researchers Ahead of Llama-5
By
–
Donc Mark Zuckerberg a volé énormément de bons chercheurs à Open AI les 2 dernières semaines. Si Llama-5 se plante après tout ce recrutement de talents… ça va être brutal.
-
How Many Training Pairs Used for Model Development
By
–
How many training pairs have you been using?
-
Opus Pricing at $75 per Million Output Tokens Analysis
By
–
Opus is $75 per 1M output tokens Claude Pro summarizes your input, and at times, this approach may be more cost-effective, but it isn't performant in some cases. Pluse they can subsidize We can take a look to see if there is an issue, but it's a costly model in the API
-
Microsoft loses WizardLM team to Tencent, creating AI gap
By
–
Imagine where Microsoft would be now if they hadn't destroyed the amazing WizardLM team, causing them to leave and go to Tencent… …where they're now releasing models far beyond anything Microsoft has created. https://
x.com/CanXu20/status
/CanXu20/status/1938981348922008002
… -
Abacus AI Offers 20k Credits for $10 with Unlimited Models
By
–
This happens only because we give you 20k credits for $10 10k credits cost $10 so in some sense the remaining credits are free! Plus you can use a bunch of models in a unlimited fashion In fact you can use RouteLLM in an unlimited fashion!
-
Making LLMs Work: Beyond Out-of-the-Box Limitations
By
–
Studies of LLMs keep looking at the (very real) failures of LLMs working out-of-the-box in complex use cases. I am always surprised that naked LLMs can get so far as generalist systems But if you want to really make an agent handle a complex workflow, you can often figure it out
-
Practical Solutions for AI Agent Reliability and Error Reduction
By
–
In practice, for many useful applications, many of the various obvious problems with AI agents (drift, hallucination, compounding errors) are more solvable than they are in theory Clever prompting, tool use, constrained topics,
LLM judges & organizational process close some gaps -
Space Curvature Technique Enhances AI Model Prompting
By
–
"space curvature" also has a nice effect in prompts pic.twitter.com/B5G0ssaarH
— fofr (@fofrAI) 28 juin 2025"space curvature" also has a nice effect in prompts
-
Two Paths to Mastering AI: LLM Understanding and Instruction Design
By
–
All the technical language around AI obscures the fact that there are two paths to being good with AI:
1) Deeply understanding LLMs
2) Deeply understanding how you give people instructions & information they can act on. LLMs aren’t people but they operate enough like it to work