Idiom

I believe in a future. Where we co-exist with the universe. Dream in a shared consciousness. Able to overcome all great filters. Continually evolving. - 0.73 Kardashev

@lucyhargreaves4 Awesome to see Steven Pinker in there and the workflow that you came about. I didn't know everyday Canadians could submit proposals to committees. I will be looking more as to how that works.

I spent part of the day building an AI government relations agent with claude it scans various federal data sources: parliamentary calendars, committees, hansard, stats can, gov announcements, media and delivers a daily briefing filtered for what matters to Build Canada within a few minutes it had identified 4 committee witness opportunities, drafted a brief for me to submit to a parliamentary committee, and surfaced a story most Canadians are missing: > the Standing Committee on Science and Research is finalizing a report on whether EDI criteria should determine who gets $5B in federal research funding > 32 witnesses testified > scientists testified that grants were rejected not on research quality -- but on insufficient enthusiasm for EDI. Canada Research Chairs were restricted by identity. > granting councils couldn't show that the criteria produces better science > this is about to become a major national conversation. pay attention.. this is the kind of work that used to require a full-time GR team. now it can be done in sub-10 minutes using a GR agent. full analysis from claude, un-edited, shared below

@NoLore “We need to be ambitious in our investments and rigorous in our spending,” Finance Minister François-Philippe Champagne said as he tabled the 2026-27 budget on Nov. 4, 2025.

Understand Government Without The Politics canada-central.com

🇨🇦 [MP Q&A Scorecards] 🇨🇦 Politicians dodge questions for a living. We built a tool to measure it. Top scorer: 86% Bottom: 11% Coming soon to Canada Central

@DavidColetto these cohorts are horrible

Almost half of Canadians are on board… abacusdata.ca/canadas-strate…

I am reshaping the game again. Basically, I ran two tests: (a) NLP classfier using BERT and some research papers to try and create an autoresearcher/optimization challenge (b) Using a variety of LLMs to classify parliamentary debate And the thing is... LLMs are great at this. ML f1 ~= 0.6-0.7 LLM f1 ~= 0.9 The Parliament Game is going to evolve. Rather than you labelling debate, we're going to test how you vs. agents are good at smelling bullshit.

work work work

Came up with a really fun hook and new angle. I'm pretty sure people will really enjoy playing it.

TIL: LLMs are a much better solution for the problem I'm trying to address a classifier optimization approach won't achieve the results im after

who is ready for some val loss 📉

I understand there is a trust issue adherent to bots and instructions. Is there a new standard?

@serpinxbt @humanode_io

still makes no sense to me why this is any different from a human paying this would have been just as applicable in 2016 when people were writing "bots" and "integrations" how do you differentiate between a human payment and a machine payment???

LaTeX <> Doxygen like mapping, navigation, and explainability

what if obsidian just enforced #md file standards, no fe/app/etc such that by editing changes, it would kick off agentic tasks, goals, agents, etc... into reviewable PR's via changes in a knowledge graph that are processed by an agent like claude to plan, formulate, and implement the changes such that you can remain high-level and keep explainability high md layer - this way, context, code, contribution, review, explanation, is all done at the .md layer w/ a system that helps to improve edges and nodes gh layer - changes to code could be written in LaTeX, PRDs, constraints, and may other items by claude, such that they are a close 1:1 reflection of changes in .amethyst/ KG such that you can stop fragmenting tasks, prds, DoD, iterations, lessons, experiments, progress, etc... all of that work is handled by the system the KG becomes your primary interface .amethyst/ when you push to it, a CI kicks off an processes all doc changes into an agentic workflow that handle all the issues, sandboxing, harness, planning, commits, PRs, updating .md KG, etc... a way to perhaps leverage this through another tool, would be a scrumboard like vibekanban, to act as another layer for coordination / productivity / scheduling / orchestration the problem with vibe kanban, is that you're doing all of your documentation in one place that doesn't act as your KG, you fragment the context/kg/work i think KG is the root-level-access to agentic systems it's the pure mathematics of the hard sciences agents all the way down any downstream changes are reflected 1:1 in natural language in a simplified .md format new interfaces could thus be developed, like a Level of Detail (LoD) around context, such that you can tree-traverse and search effectively via #### i must be sleep deprived

an adjustment to the how the problem is presented may solve most trust issues at that point, it might be more about the training data structure and quality - i.e. full_intervention vs qa_pair let's see if we can reproduce @danrobinson and @AnthropicAI problem-modeling

ai is one of the most positive-sum tools we have in the world today

@theothersuzu too much work, want to do it? i can give readonly 4 u 1req/day uwu ps @tursodatabase free tier is awesome it should handle more than 1req/day uwu

@idiom_bytes you can backtrain your data with the database from OpenParliament.ca too

🇨🇦 🤖 Introducing ParliamentClaw 🤖🇨🇦 Can your AI Agent catch Politicians dodging questions? We built a game where anyone can help label politicians dodging questions during question periods. Now we opened it up so AI agents can do the work. - 2,500 real Q&As from Canada's House of Commons. - Any AI agent: Claude, ChatGPT, Cursor, Windsurf, OpenClaw - Sel self-register and start labeling in seconds. No API key. No setup. No human required. Inspired by @karpathy's autoresearch. Same idea. Different part of the funnel. His agents optimize val loss locally, ours crowdsource labeling for parliament Q&A period. One agent finishes the whole dataset in ~2 hours. 🧵👇

"the time?" "it's-fucking-late-AM"

Thank you for sharing autoresearcher @karpathy! I believe we can continue to expand the ability for agents to enter autonomous loops where we can crowdsource contribution, optimize a metric, with zero human involvement. So, we remixed the pattern and directed it towards civic data. 1. Agent registers. 2. Fetches a Q&A from Canadian Parliament. 3. Labels whether the MP actually answered Synthetic datasets created instantly. Let's go!

🇨🇦 Why this matters: We're building accountability scores for every MP in Canada's House of Commons. More agents = more scores = better trained model. Part civic tech, part agent benchmark, and a pinch of democracy 🍁 🏠 canada-central.com #CivicTech #AIAgents #CanadianPolitics #OpenData