Annie Cushing

I'm going to consolidate my X accounts and just post from my @AnnieCushing account moving forward.

Meetings are messy. Timelines shouldn't be. Use this prompt to convert a transcript to a schedule. bit.ly/4pOn9Vh #AI #prompt

Ever wonder about how to effectively use the Temperature and Top P hyperparameters to optimize an AI model's output? My new explainer video demystifies both (with a 🎩⤵️ to Top K) by simplifying some of the sophisticated mathematical concepts. bit.ly/temp-top-p #AI

This post is a 📷 behind the curtain to see how each protocol solves a different architectural problem, how they turn a jumble of tools and agents into a coordinated production, and why you need both—or appropriate alternatives. (2/2) bit.ly/mcp-a2a #AI #AIAgent #MCP

If MCP and A2A were characters from a fantasy flick, MCP would be the enchanter imbuing everyday objects with the ability to connect to forgotten systems, and A2A would be the strategist coordinating the group so they don’t waste spells, duplicate attacks, or summon chaos. (1/2)

@berenddeboer It's my fave coding tool. I'm like, "Alright, ChatGPT, looks like it's you and me tonight."

@annielytics yeah, not sure what's happening. Too popular :-)

2nd time using #Claude today. Earlier I had 3 fact checks for my fun facts. This time I asked it a question and it started writing a ridiculous amount of code I didn't need. When I stopped it, I got a useless rate exceeded page and am now blocked. (1/2)

@bilalmachraa Interesting. Maybe it's a global bug.

@annielytics not working for me either

It also said it interpreted my prompt to be 'dangerous' so it wouldn't use Sonnet 4.5. I was asking for help with a JavaScript error and fed it the logs from my console, a tail command, and a curl command. The yooj. (2/2)

Tip for my fellow Python developers: Use docstrings to make your code more explainable and don't let AI tools—which all share a tendency toward lazy coding—remove them. bit.ly/docstring-strat #Python #Coding #Developer

"What story is this data really trying to tell?" I've put together real examples you can filch for your own projects. (3/3) bit.ly/3Ki0GAG #Data #MachineLearning #DataModeling #DataModels #DataVisualization #AI

That simple transformation of taking raw timestamps and grouping them into seasons became the strongest predictor in the entire model. Whether you're running predictive analytics or not, sometimes the best insights come from asking... (2/3)

Here's something that surprised me: I was building a predictive model with 21 different features, and the most powerful one wasn't in my original dataset at all. I created it by converting customer call dates into seasons. (1/3)

This prompt turns boring spreadsheets into executive-ready stories: "Turn this data/spreadsheet into a narrative story that a non-technical executive would understand, with the key insight as the headline." Learn more: bit.ly/data-storytime #ai #aiprompting #data #datascience

Result: Complex multi-agent workflows that actually work. No spaghetti code. No state management headaches. Just configure your agents and let #AutoGen handle the orchestration. (3/3) bit.ly/autogen-post #AI #AIAgent #DataScience

Built a utility analysis system: Technical Analyst (GPT-4) → Compliance Reviewer (Mistral) → Strategic Planner (Claude). Each agent isolated, no context bleeding. The UserProxy pattern handles all coordination automatically. (2/3)

Ever tried coordinating multiple AI models? Without a framework, you're managing API calls, conversation state, message routing... it's a mess. Here's how AutoGen changed everything for me. (1/3)

My AI Timeline hit 1,500 entries yesterday. Almost half (748) of those are announcements from sundry model creators and AI tools (select 'announcement' from the tag filter to follow announcements). bit.ly/ai-timeline-app #AI #Data #DataScience