I love how AI has rekindled discussion and interest in my bubble of "consciousness" and "the hard problem". I am largely persuaded by @danieldennett (RIP)'s Consciousness Explained, now known as Illusionism: consciousness… https://firefly.social/post/ff-4f842567d1534b52972af686a3b23436?s=bsky

Continue reading on Firefly.Social

I love how AI has rekindled discussion and interest in my bubble of "consciousness" and "the hard problem". I am largely persuaded by @danieldennett (RIP)'s Consciousness Explained, now known as Illusionism: consciousness is a real, emergent outcome of physical processes, while "qualia" or "the ineffable aspect of experiencing stuff" is fully explainable as functional physical processes. Perhaps those who balk might be interested in distinguishing "fully explainable" (given infinite resources) and "fully explained" (impossible in practice) with me? Now I'm especially intrigued by the connection to "qualia", "ineffable / lived experience", tacit knowledge, scientific modeling theory, and my moral stances around sentience. Sounds like from ChatGPT I should read Dennett's "Welcome to Strong Illusionism", Adam Jenkins's “What Mary Didn’t Know About Tacit Knowledge” (2003), and Michael Polanyi's "The Structure of Consciousness". Anybody wanna start a book club?

firefly.social

How do you strike the balance between improving your lifestyle, processes, tools, workshop, "shipping results", and a meaningful / eudaemonic / The Good Life™️ trajectory? I like to fold the eudamonia stuff into "shipping… https://firefly.social/post/ff-c7febcaf258c4a1297671b8f884aa079?s=bsky

Continue reading on Firefly.Social

How do you strike the balance between improving your lifestyle, processes, tools, workshop, "shipping results", and a meaningful / eudaemonic / The Good Life™️ trajectory? I like to fold the eudamonia stuff into "shipping results", perhaps a geeky/aspie approach. As a FLOSS grey beard, I've always been a yak-shaver: every tech project involves a lot of time thinking about / iterating on my underlying toolchains/infra. Coarsely, it is: self-sovereign hardware -> human-computer executive layer -> project tooling -> "tangible results" goal achievement. (Modern products basically lack the exec layer altogether. Soon to change w/ AI. :robot-winking-emoji: ) AI agents haven't changed the shape of my tech yak-stack at all! My leaning edge in AI is figuring out how to redistribute my effort on yak-stack frontier progress. Tangents: my "hardware" layer needs better bio integration, and all layers need better social integration. Any SF workout / diet buddies interested? Ping me!

firefly.social

A new Zcash-focused org just popped onto my radar: https://zcashlabs.org/blog/announcing-zcash-labs I'm especially keen to see how up-front funding for retroactive grants plays out. Love to see funding pathways increase… https://firefly.social/post/ff-096a680ed2124adfa0e94d29a6392c54?s=bsky

I'm going to start using a new microblogging convention by repurposing unicode glyphs 📍 to better express how my brain produces tokens: • "𛱼" means "there's a tangent I'd like to explore given time/interest, but I'm not… https://firefly.social/post/ff-09ea05aee7c2447a8d760f9e3eb08298?s=bsky

Just another day, running across just another example of how the status quo approaches to regulation which sounds like it's for a good purpose ends up just adding a billion paper cuts to the little guy, aggregating to… https://firefly.social/post/ff-c638859f821c4a68a047305a1a30e18c?s=bsky

LLM token efficiency is fascinating and feels like a primary edge skill in becoming a master clankherd. Who has the best resources for learning / engineering better token efficiency?

Suckered by Anthropic's biz model: hook me w/ a subscription that's "cheaper" for agent capabilities but only if I use their UI, which is just enough to get me entangled while shitty enough that I want to write my own, pushing me towards the pricier metered API. Checkmate. ♚

Out here amped up with a hive of clankers ready to conquer the world, yet I'm still just yak shaving… … Yak shaving on cyborg steroids. Skills skillin' skills.

Advertisement: "This battery bank has a 20,000 mAh capacity!" Me: "So, 20 Ah?" Advertiser: "No, it's a much bigger number! Look at all those zeroes!" Me, thinking: Coulombs would be an even bigger number...

My AI request-to-verify ratio is a lopsided bottleneck. Next experiment: #DearAgent: please review all past contexts, find the agent responses that I'm likely to have never reviewed, summarize+prioritize those results based on my priorities, clarifying my priorities as we go.

If you were orchestrating ai agents precisely in a hierarchy rolling up to a single "Executive Assistant to The Human" while pushing the pareto frontier of cost vs effectiveness, do you make the EA the costly/smart model, or can cheaper/dumber delegate to smart effectively?

I can finally put to words why "tech" is bittersweet for me from ChatGPT analysis of "general tech" convos. This is all "conversational", not agentic/coding/building flow. 15% learning, 32.5% building, 2.5% doing (instructional) - 50% YAY! Then 50% bullshit snarls & snags. 🧵

One way I find microblogging to be socially connecting is via sharing playful speculations or questions. Now I notice an AI-age tension: I can often ask AI for further info on my speculations or questions, then I may not post. Lost connection opportunities!

"Why read words when you can just kill the stuff the words tell you stuff about?"

I think about this a lot.

Me: Hears people worrying about AI taking all our jobs. Glances at backlog whose growth rate is headed for orbit due to AI-fuelded productivity. Scratches head…

A clear AI win is to describe what I want, then ask "What are the best technologies available to meet my desires?" I am (we are?) in a phase that uses AI to glue together those pieces. What proportion of future tech will be commoditized (many user) vs customized per-user?

For my work project this morning, I had one my intermittent whimsical thought: "I'm going to learn this the old fashion way by reading and doing!" Good ol' `nix` land, never fails to deliver a good old fashioned slog...

With the discovery of the Zcash Orchard vulnerability, @ShieldedLabs followed our principle of honoring user agency in disclosure: straight impact facts, avoid speculative assurances without providing explicit rationale, no "panic/mrkt" mgmt. Price fell: aggregate choice. 🧵

Props to @cipherscan_app! Public analytics around Zcash metadata privacy is an invaluable resource to educate/protect users. Zcash's leading private on-chain payments protocol has inevitable metadata leaks wallets can improve; here's a view into that: https://cipherscan.app/privacy/wallets

New Zcash node just dropped! Can't wait to check it out: https://zakura.com/announcements/introducing-zakura/ X announcement: https://x.com/zkDragon/status/2077418522909761944

Ok, Zcashers, who else is prepping champaign bottles along with an emergency hardhat/toolbelt combo for the height 3,417,100 #zcashd end-of-support farewell? https://zcash.github.io/zcash/

I'm excited to see @sovright_ ramping up (formerly Bootstrap board) and with a focus on more permissionless mining, an age old and important gap in Zcash. The wallet recovery tool could be invaluable, too! (Check out the X handle for video explainers.) https://sovright.com/#thesis

Sovright — The choice is yours

Sovright builds the infrastructure that preserves our right to choose: who sees your data, who controls your money, and how you exist online. It began with Zcash. It doesn't end there.

sovright.com

Google Calendar, now with Eventually Inconsistent Technology. Too many UX flaws to describe here. Any better calendar recommendations?

Are the AI orchestration champs just bullshitting me or am I missing some significant AI skill/knowledge? I run into time-wasting slop for most multi-step, single context / serial agentic requests. How can concurrent orchestration possibly work? Anyone wanna help me level up?

Plz help a skeptic understand how important AI-assisted math proofs are. Is (Dis)Proof discovery of a known conjecture qualitatively different from searching a space constrained by inference steps? If yes, what's new? If not, I'm not too impressed. …

When programming, I eschew Hungarian notation when it's not necessary, and I define "not necessary" as a boolean quality of "the type system and tooling". But when it _is_ necessary for readability, I then wish English and Noun-Then-Adjectives ordering. ;-)

I've used a lot of apps & FLOSS for planning, journaling, calendaring, task tracking, etc... Here's how all of that stuff can't compete with a paper journal: - strong privacy - local first (no net needed) - pen life ⋙ battery life - use in natural light at wide angles …🧵

I'm fantasizing about ZKPs for refactoring long-term append-only universal/distributed data structures; "decentralized permissionless db migrations". Perhaps "succinct blockchains" are an instance of this; I'm curious about broader applicability.

When I hear people concern themselves with the parochial issue of whether a participant is "human or AI" or requesting disclosures or disclaimers of AI usage...

a cartoon character is holding a newspaper that says " old man yells at cloud "

ALT: a cartoon character is holding a newspaper that says " old man yells at cloud "

media.tenor.com