Next, a fireside chat with @pirroh from @Replit ! Code generation is a flagship use case for LLMs, and Replit agent is at the forefront, helping thousands of developers build full-featured apps. LangSmith has been vital in helping them there! – Using multiple models, very
CODE
-
GitHub MCP Server Connected to Terminal AI Agent for Testing
By
–
I connected the GitHub MCP server to my terminal AI agent, analyzed the code, and generated tests, right inside my terminal!@Warpdotdev just dropped MCP support, and you can now connect external data sources to your terminal's AI agent.
— Sumanth (@Sumanth_077) 14 mai 2025
It’s free and super easy to set up pic.twitter.com/CG2MN9Kxq3I connected the GitHub MCP server to my terminal AI agent, analyzed the code, and generated tests, right inside my terminal! @Warpdotdev just dropped MCP support, and you can now connect external data sources to your terminal's AI agent. It’s free and super easy to set up
-
Reasoning tokens alone cannot solve fundamental AI limitations
By
–
Adding more reasoning tokens is not going to fix this
-
Stable Audio Open: Weights, Code, and Research Paper Released
By
–
Weights, code, and paper @HuggingFace
: https://
huggingface.co/stabilityai/st
able-audio-open-small
… @Github
: https://
github.com/Stability-AI/s
table-audio-tools
… @arXiv
: https://
arxiv.org/abs/2505.08175 -
Bayesian Spam Filters Evolution: From Cutting Edge to Simple Implementation
By
–
Side note: even when implementing Bayesian filters (thanks PG!), which were once very cutting edge, never would have imagined that you could eventually write a fully featured spam classifier as:
-
MCP Course: Build Rich-Context AI Apps with Anthropic
By
–
New course: MCP: Build Rich-Context AI Apps with Anthropic. Learn to build AI apps that access tools, data, and prompts using the Model Context Protocol in this short course, created in partnership with Anthropic @AnthropicAI and taught by Elie Schoppik @eschoppik, its Head of… pic.twitter.com/NDZZzP8QYb
— Andrew Ng (@AndrewYNg) 14 mai 2025New course: MCP: Build Rich-Context AI Apps with Anthropic. Learn to build AI apps that access tools, data, and prompts using the Model Context Protocol in this short course, created in partnership with Anthropic @AnthropicAI and taught by Elie Schoppik @eschoppik
, its Head of -
Lua Security Model Limits Network Connections and HTTPS Support
By
–
Presumably because the Lua scripting security model doesn’t allow arbitrary network connections or HTTPS requests, for predictable reasons. It would have to have direct support, which is not a priority for the company.
-
Tools Should Implement Security Fixes, Not Users
By
–
its for the tools to implement, not the users e.g. supabase security advisor – one-click fix for security issues
-
Salesforce Launches xGen-Small Models with 4B and 9B Parameters
By
–
Salesforce released xGen-Small, a new family of small models It includes 4B and 9B param models, with the larger variant proving competitive on long-context understanding up to 128K tokens Also delivers strong performance on math and coding benchmarks
-
ai.txt: Domain-Specific Language for AI Internet Interactions
By
–
ai.txt: A Domain-Specific Language for Guiding AI Interactions with the Internet
Link: https://
arxiv.org/pdf/2505.07834