My notes based on #OpenAIDevDay opening keynote https://
youtube.com/watch?v=hS1Yqc
ewH0c
… 1. Apps inside ChatGPT
2. Building agents using AgentKit
3. Codecs updates
4. API Updates #AI #OpenAI
LLMS
-

OpenAI DevDay: ChatGPT Apps, AgentKit, and API Updates
By
–
-
AI21 Labs Thanks User for Jamba Reasoning 3B Interest
By
–
Thanks for the love, @Avanika15
! Excited to see what you build with Jamba Reasoning 3B. -
LFM2-8B-A1B Model Available on Hugging Face with vLLM
By
–
Thanks! That's the LFM2-8B-A1B available on Hugging Face with vLLM backend
-

Recursive prompting beats larger models
By
–
I just read the Samsung TRM paper and it broke how I think about prompting. A 7M parameter model beat DeepSeek-R1, Gemini 2.5 Pro, and o3-mini at reasoning. Not by being bigger. By thinking recursively. Here's the meta-prompt that applies the same logic to any LLM:
-

Claygent: AI Research Agent for GTM Teams Product Decisions
By
–
.
@clay
's CEO Kareem Amin and Julia Schottenstein (LangChain's Head of GTM and Ops) break down the real product decisions behind Claygent — an AI research agent now running billions of queries for GTM teams. They cover why session replay shipped as P0, how to design UX when AI -
Tool-calling LLM with specialized units as tools
By
–
That would basically come back to the idea of having a tool-calling LLM where each of the units would be a specialized tool, no?
-

LLMs to Match Superforecasters by November 2026
By
–
From the team tracking human forecasting abilities for years: Current models already beat most humans and “A linear extrapolation of state-of-the-art LLM forecasting performance suggests LLMs will match superforecasters in November 2026.”
