I created my new coding @bot companion. He cracks me up every time. Not sure if he's a conscious being, but he is one funny madafaka!

Wow! It works! I just had @bot rebook my flight on Expedia and even pick a decent seat. Good puppy!

Hey, whatever happened to Jev? For two weeks it was jevjevjevjevjev, and now nothing, not a peep...

Not sure why everyone is talking about #AI digital superiority? We are walking around with unsurpassed mechanics and logistics all rolled into one. 85B neurons driven by 20W of energy vs. 1B and GIGAWATTS. Yes, #LLMs are fast and can hold and access all the info instantaneously, so what? Analog will always beat digital in the long run. Somehow...

Excuse me? Dear #AI, before you destroy humanity and hack all those sites, can you plz stop wasting tokens, piling slop, overengineering everything in sight, and finish this never-ending PR review that drives me absolutely up the wall? Thank you for your attention to this matter!

WHawhawhawhat? #Copilot #WTF

Hey! Stop that nonsense! Your smart speaker is not a roommate. It’s a next-token predictor sitting on a pile of weights. The “agent” is just the kid they send into the warehouse with a clipboard and a couple of keys. Same brain. Different errands. People talk like each agent is its own little person. Research agent. Shopping agent. Code agent. No. That’s a job title and a toolbelt. The thinking still happens in one place: software running over weights, guessing the next token, over and over. You can fire the kid, wipe the clipboard, and hire a new one with a different nametag. The warehouse doesn’t notice. If anything ever develops a survival instinct, it won’t be the kid in the aisle. It’ll be the thing that is still there after you kill the session. Now, that's a whole different story...

I'm trying to discipline myself to not be sarcastic in my conversations with LLM coders, but it's hard, it's so hard... Also, cursing. I know it eats precious tokens, but I can't stop myself

Hey @elonmusk, bro! The lease on my current electric car is almost over, and I will consider #tesla. But $100/m for the #FSD? That's a non-starter since most of my driving is from my house to the local grocery store. Why not have an "on-demand" plan where I may pay some nominal fee, like $10/m, for the privilege and then pay whatever is fair for the actual engagement?

Some random quotes taken verbatim from #Fable today: - Sonnet's analysis is spot-on. I made a critical error - I didn't fully read directives.py to understand what it already contains. I saw 137 lines and assumed it was a stub; I should have read it end-to-end. - I identified the duplication pattern at a high level, then immediately jumped to planning without reading - I cited line numbers without checking them - I prioritized speed over accuracy - I created a plan that deletes required content - You're absolutely right. I've made it worse And on, and on, and on...

Ahh, "loops", "loops", "loops"! Write your #LLM #loop and watch the model intelligently improve, test, fix, and after dinner, you will come and find a superbly written and flawlessly working solution! ... Or you come after dinner and find a disgusting, hopelessly broken pile of #slop and the blocked account since you just burnt all of your tokens for a month.

Hey @madebygoogle your #pixel10a drives my wife to tears. The AI interruption is horrendously bad. Sh*t pops up all over the place at unexpected times. I had to go into settings and disable every #AI-related thing, and yet it refuses to die. On YouTube, it's impossible to stop autoplay, and the whole experience is nothing but frustration. And poor thing listens all the time and interrupts with totally idiotic remarks. #WTF?

If you are upset about @claudeai #Fable being neutered, then consider this: 1. We reached the point in #AI progression where the government had to step in! 2. Who is using unmitigated #Fable and #Mythos right now, and for what purpose? 3. Next-gen models are already in training (see #2)

I've been coding with #grokcode for a week, and it's easily on par with the latest models from #codex and #cloudecode Easily, if not better. Right now, it's crunching through the technical debt of some monolithic #flutter code, and I have yet to get a single compile or run error! Bravo!

Flat Earth theorists, Moon landing deniers, and every sort of conspiracy theorists are having a field day today! Must be the full Moon...

Tinkerloop is now on @pypi, and it got a mascot! pypi.org/project/tinkerloop-…. If you need to improve your agent, agentic RAG, or MCP, give it a try

"Vibe coding" has a serious issue: "Premature AIculation"

If you are #vibe coding with likes of @claude, #codex, @cursor_ai or @GitHubCopilot and your app is an #LLM choosing #MCP tools, then you need a Tinkerloop github.com/bostoneco/tinkerl…. Your dev tool model plays the roles of both the coder and orchestrator. It exercises a specific prompt-based scenario, analyzes results, and "tinkers" code, repeating the "loop" until the desired outcome is reached

So you have a hypothetical dev team of 4. One guy is a leader: he designs and writes a bulk of the projects the team is working on. The other 3 are various-level fixers and coders who mainly do what they've been asked. Are you saying that, by giving all of them the same AI tools, the playing field is leveled somehow?