they (or more likely more efficient replacements) will play a role. but spending so much money, time, and energy on them has quite likely delayed us. we need new ideas, not slightly better LLMs.
LLMS
-
Multi-Agent Communication Costs: Beyond Token Overhead
By
–
Multi-agent systems are powerful but expensive. However, the cost isn't in the reasoning itself. It's in the communication. Agents exchange full text messages, consuming tokens for every coordination step. When agents need to collaborate on complex problems, this overhead adds
-
LLM Over-Focus Causing AI Field Stagnation
By
–
you could argue the analogy but the general point is correct; an over focus on LLMs has allowed the field to stagnate.
-
AI21Labs Partners With Together to Open Enterprise AI Access
By
–
We’re taking a major step toward making AI more open, reliable & enterprise-ready.@AI21Labs is partnering with @togethercompute to give teams seamless access to open-source & frontier models through the Maestro orchestration system.
— AI21 Labs (@AI21Labs) 26 novembre 2025
👉 Read more: https://t.co/hyx4YNCum9
📽️… pic.twitter.com/xWip625yQqWe’re taking a major step toward making AI more open, reliable & enterprise-ready. @AI21Labs is partnering with @togethercompute to give teams seamless access to open-source & frontier models through the Maestro orchestration system. Read more: https://
ai21.com/blog/ai21-toge
ther-ai-partnership/?utm_source=org-twitter
… -
Opus 4.5, Gemini 3.0, and GPT 5.1 Thinking Combined for Magic
By
–
Opus 4.5 + Gemini 3.0 + GPT 5.1 Thinking combine together to make MAGIC
-
Self-Search Reinforcement Learning Improves Language Model Information Retrieval
By
–
Researchers introduced Self-Search Reinforcement Learning (SSRL), a method that teaches language models to simulate web searches to better retrieve information from their own parameters. SSRL fine-tuning improved accuracy on multiple question-answering benchmarks and even boosted performance when paired with real web search tools. Read our summary of the paper in The Batch: hubs.la/Q03VV2d-0
→ View original post on X — @marcusborba, 2025-11-26 00:00 UTC
-
OpenAI to Launch GPT Store v2 for Apps
By
–
BREAKING : OpenAI prepares to enable developers to publish their Apps and Workflows on ChatGPT via the OpenAI Platform. A new Apps directory (GPT Store v2?) will be added to ChatGPT as well. Other additions:
– "Hermes" feature will let builders publish (and share?) workflows -
LangChain 1.1 Adds Programmatic Model Capabilities Access
By
–
LangChain 1.1 You can now programmatically access model capabilities and supported features through model.profile. This powers some new features in middleware. One example: SummarizationMiddleware can now dynamically trigger based on the model's available context.
-
LLM as Judge: Managing AI-Generated Content Noise
By
–
One AI use case that is only getting more popular: LLM as a judge. Everyone still talks about AI generating more content, but not enough people are talking about: 1) the horrific deluge of noise we're going to have to deal with in the next 12 months when everyone nanobananas a
-
Grok 4.1 Now Available for Pro and Max Users
By
–
And now, Grok 4.1 for all Pro and Max users! I have been really impressed by the speed and cost-efficiency of 4.1 and we’re beginning to consume a lot of Grok 4.1 tokens for internal use too. Look forward to 4.20 when it’s ready! https://t.co/0oPb8CcS3V
— Aravind Srinivas (@AravSrinivas) 25 novembre 2025And now, Grok 4.1 for all Pro and Max users! I have been really impressed by the speed and cost-efficiency of 4.1 and we’re beginning to consume a lot of Grok 4.1 tokens for internal use too. Look forward to 4.20 when it’s ready!