iiinova — where your digital self keeps evolving

Reposted by Russell Pierce

Molly Jeffery, PhD MPP @mollyjeffery.bsky.social · 30/08/2026

Can confirm. A student told me about a an LLM-based tool that is supposed to help with your lit review. I asked it a question where my work should have been prominent. None of my work came up. Lots of different men did.

29133

Russell Pierce @russellpierce.bsky.social · 31/08/2026

Relearning this morning why I don’t love the MCP spec. Actual usage almost always looks like an API or CLI. The spec itself is meant to support something a lot more like a structured conversation. But, the spec there is so flexible… it’s time consuming to write a workflow against it.

000

Russell Pierce @russellpierce.bsky.social · 14/08/2026

MCP was a dumb replacement for APIs, but it implied a real vision- stateful cross agent communication - and now with this revision… what is left that makes it worth the wrapping of an API?

000

Russell Pierce @russellpierce.bsky.social · 02/08/2026

@mattpocock.com how do you think of something like Gas Town relative to what you’re doing with wayfinder? Especially with regards to coordination via centralized task manager rather than git managed task management (e.g via beads or ticket)?

030

Russell Pierce @russellpierce.bsky.social · 01/08/2026

@databricksinc.bsky.social what’s up with @anthropic.com proprietary model serving costing $0.105 per DBU in Azure while it costs $0.07 per DBU everywhere else?

000

Russell Pierce @russellpierce.bsky.social · 05/06/2026

Why are the foundation model providers failing so hard at building a decent memory system? Is it just because the approach they’re using now costs more billable tokens? It’s a waste.

010

Russell Pierce @russellpierce.bsky.social · 23/03/2026

LLM providers are going to need to rethink their web result processing strategies. Using cheaper and less capable models on search results will almost certainly result in greater prompt injection risks for interpreting current events without higher authoritative intermediate subsystems.

100

Russell Pierce @russellpierce.bsky.social · 21/03/2026

A really fun statistically minded analysis by @statwonk.bsky.social. Worth the read!

100

Russell Pierce @russellpierce.bsky.social · 08/03/2026

I don’t know which I find more challenging a person being forced to act soulless in a support role because they work for a soulless company or when it is an AI being soulless on the direct behalf of the company. There is a thin hope of some procedure breaking human decency with actual human support.

020

Russell Pierce @russellpierce.bsky.social · 06/02/2026

I locked @openclaw-x.bsky.social in a sandbox and watched it skitter around, with Opus 4.5 and with gpt-oss:120b. The security doomers are spot on. But the threat model is similar in kind (but worse) that just giving a random person access to equivalent permissions needed to perform a given task.

110

Russell Pierce @russellpierce.bsky.social · 16/12/2025

The skill markdown file for a Claude skill must be static when packaged. But, nothing seems to prevent it from being a stub that then calls a method within the skill to populate dynamic content. My question is how many boilerplate tokens (aside from description) does a skill burn just by existing?

020

Russell Pierce @russellpierce.bsky.social · 16/12/2025

000

Russell Pierce @russellpierce.bsky.social · 12/12/2025

Sklearn's regression metrics function r2_score can yield negative values. No, it isn't yielding an adjusted r^2. Cursed.

110

Russell Pierce @russellpierce.bsky.social · 21/11/2025

Do other parents enjoy helping their kids out on studying for spelling tests? How do you do things so that it is more enjoyable?

000

Russell Pierce @russellpierce.bsky.social · 16/10/2025

It is f'ing awesome to realize that something you'd been shooting intuitively around for years but couldn't get a handle on has a fully realized version. It is a little bittersweet too. Moreso when you see that the realized version has been around for decades.

100

Russell Pierce @russellpierce.bsky.social · 10/10/2025

You can't make this stuff up. Gnome was dying on a new linux machine because... *checks notes* the screensaver wouldn't start because a bluetooth library wasn't available.

100

Russell Pierce @russellpierce.bsky.social · 03/10/2025

I've spent some of today using local models to squish down verbose webpages to the things I actually care about for doing additional work. Then I notice Qwen code does "Fetching content from {url} and processing with prompt: {prompt}"... makes me feel seen and validated.

000

Russell Pierce @russellpierce.bsky.social · 02/10/2025

It's funny how Anthropic silently boosted usage available on Pro plans this week in the middle of the week AND changed the weekly reset to Wednesday from Monday. I presume it is because they released a new feature that chews through tokens like mad and they want people to fall in love with it.

000

Russell Pierce @russellpierce.bsky.social · 01/10/2025

Developers Log Supplemental. #AI (so you can ignore it if you like). Background food for thought... if you're billed by token... what motivation does an LLM provider have to reduce your token usage? To serve more with their current fixed capacity. Will that last?

111

Russell Pierce @russellpierce.bsky.social · 01/10/2025

Unstructured or poorly structured data? LLMs can be helpful for this! The cleaning code can be audited to help validate correctness (don't trust re-authored datasets). 13 minutes and $3.75 list price to clean that dataset. $0.43 per 1k rows. Not bad! Thanks for sharing @jeremy-data.bsky.social

020

Russell Pierce @russellpierce.bsky.social · 01/10/2025

I'm sorry, how the heck do they release The Mandalorian & Grogu movie on May 22nd instead of May 4th?

110

Russell Pierce @russellpierce.bsky.social · 24/09/2025

TL;DR the Zellij - terminal multiplexer is great (@zellij-dev.bsky.social). ZOMG... I had played around with screen and tmux way back in the day... but like my desire to play with emacs it was limited by my mental bandwidth/priority to pick up the key combinations.

130

Russell Pierce @russellpierce.bsky.social · 18/09/2025

I'm sorry, the ultimate complaint against Kimmel is *checks notes* he went on FCC regulated broadcasting and told lies? Um, look it's an evolving environment such that what's actually going on may not be very clear... and even if it was don't look at the AM dial y'all.

040

Russell Pierce @russellpierce.bsky.social · 05/09/2025

@anthropic.com you know what would be pretty nice? The ability for the user to set a cap on Claude code usage so that they have usage left over for other tasks. Of course, for driving upgrades that's probably wrong. But as a public benefit corporation I hope you might consider it.

010

Russell Pierce @russellpierce.bsky.social · 14/08/2025

Has anybody been able to make Gemini Flash 2.5 make two tool calls in a single response? I found it took me extra prompt development to get the calendar tool response to include much beyond the calendar entries themselves.

share.google

Control your Android mobile device & apps with Utilities - Gemini Apps Help

When you use Gemini as your mobile assistant on  your Android device, you can ask it to do certain actions for you with the Utilities app. For example, you can ask it to manage your alarms and ti

100

Russell Pierce @russellpierce.bsky.social · 15/07/2025

If you're interested in the AI space, but aren't familiar with the history and discourse on AI prior to 2010, the "AI winter", and/or don't always understand the perspective of @yann-lecun.bsky.social, then the rest of this post might be for you.

110

Russell Pierce @russellpierce.bsky.social · 05/07/2025

It's mildly interesting to consider that LLMs are perhaps a system for which the Sapir-Whorf hypothesis is most true (i.e., language one speaks influences the way one thinks about reality). What to know: the 'voice' you ask these models to respond in will tend to also affect the content itself.

041

Russell Pierce @russellpierce.bsky.social · 22/05/2025

Server Sent Events (SSE) for the Model Context Protocol (MCP) seems to me to be a pretty big over complication for the purposes of providing a streaming response. The need to maintain an open pingable host seems to bedevil a simple serverless context holder. Why is it this way?

120

Russell Pierce @russellpierce.bsky.social · 16/05/2025

The R folks didn't really give the Python folks enough grief about poor threading support in Python. The era is almost over though... labs.quansight.org/blog/free-th... Python leaves behind the GIL and @posit.co enters to offer paths with a consistent approach? It'll be fun to see what happens next.

labs.quansight.org

The first year of free-threaded Python

A recap of the first year of work on enabling support for the free-threaded build of CPython in community packages.

040

Russell Pierce @russellpierce.bsky.social · 03/05/2025

One person's syntactic sugar is another's strychnine... and sometimes both at the same time for the same person.

020

Russell Pierce @russellpierce.bsky.social · 30/04/2025

Come on... it is kind of hilarious that LaTeX has this progression for text sizes... \normalsize \large \Large \LARGE \huge \Huge

060

Russell Pierce @russellpierce.bsky.social · 09/04/2025

I find that for the first time in my life I'm less than enthusiastic about open sourcing/publishing something. LLMs will scrape it and make my having shared it less valuable (to anybody other than LLM companies and their customers). Anybody else addressed similar feelings of reservation?

173

Russell Pierce @russellpierce.bsky.social · 08/03/2025

Does the loss of almanacs and encyclopedias mean that we can't remake the classic Carmen San Diego games? Are they teaching a skill that isn't as relevant anymore?

110

Russell Pierce @russellpierce.bsky.social · 03/03/2025

#dallas how degraded do we think tornado detection and warning are right now? Are things going to flow forward on autopilot right now, or do we need to exercise extra care?

000

Reposted by Russell Pierce

Justin Pot @justinpot.com · 21/02/2025

When you see a headline you want to click, stop. Breathe. Ask yourself: why do I want to click this? What does that say about me? Do I like what it says about me? And, most importantly, am I likely to learn anything?

justinpot.com

Reject anxiety bait – Justin Pot

Reject anxiety bait

084

Reposted by Russell Pierce

Alejandra Caraballo @esqueer.net · 16/02/2025

When the Nazis took power and gutted the Weimar era civil service in government, they also claimed the need to increase efficiency and to ensure loyalty to the nazi party. They did it through the law for the restoration of the professional civil service. en.wikipedia.org/wiki/Law_for...

en.wikipedia.org

Law for the Restoration of the Professional Civil Service - Wikipedia

232394862

Russell Pierce @russellpierce.bsky.social · 15/02/2025

I kind of miss the small blast radius of live journal. I feel like targeting on this and similar platforms is just so... insufficient. I want to follow public people... but I want to be able to share with curated subsets. This was something Google + or whatever it was called got right.

010

Russell Pierce @russellpierce.bsky.social · 15/01/2025

My whole life I heard "mechs down at nav gamma" as "mechs down as Nasgana"... which I just took to be a fantasy place name. #mechwarrior

120

Russell Pierce @russellpierce.bsky.social · 05/01/2025

My evolution from "I hate install disk copy protection, what if I lose or damage the disk" to "I love install cartridge/disk DRM because then I can buy (and get) old games at reasonable prices" strikes me as weird.

000

Russell Pierce @russellpierce.bsky.social · 01/01/2025

Over/under in Virgin Galactic being real about developing and releasing a Mach 3 jet? My gut reaction is that it is just thrashing vapor from a company in a bad spot.

000

Russell Pierce @russellpierce.bsky.social · 24/12/2024

When you want to know something specific to your company how often do you go to documentation vs search Slack? Is that for the best or not? Is being searchable and low cost to generate resulting in less high quality documentation?

400

Russell Pierce @russellpierce.bsky.social · 21/12/2024

The best part of working in tech is that there is always something new (to you). Today, I stumbled into overrides in the Chrome Dev Tools (developer.chrome.com/blog/new-in-...). Pretend that API gave me the answer I wanted... what does that look like?

developer.chrome.com

What's New in DevTools (Chrome 117) | Blog | Chrome for Developers

Override XHR/fetch requests and hide extension requests from the Network panel, see changes in fetch priority in the Performance panel, experience multiple UI improvements, check out new colors and ex...

030

Russell Pierce @russellpierce.bsky.social · 14/12/2024

Skeleton Crew is (so far) the best thing out of Disney's Star Wars since Mandalorian. A long time ago in a galaxy far far away it is the 80s and kids ride bikes with limited parental supervision resulting in adventure. Sure, it's for kids. But, good Goonies 'for kids' rather than Jar Jar

310

Russell Pierce @russellpierce.bsky.social · 10/12/2024

How is python packaging this ... erratic? I could swear I'm staring at a problem where the pip solver is coming up with different answers given the same local inputs. I've been beating my head against the wall for a few hours so PEBKAC is just as likely.

221

Russell Pierce @russellpierce.bsky.social · 03/12/2024

Maybe yes? Heck, it solves one of the problem current approaches have... they're searchable... and it is the wrong medium for durable authorship. See my other posts this morning about the importance of forgetting.

000

Russell Pierce @russellpierce.bsky.social · 03/12/2024

Atlassian: we have created JIRA Me: you’ve fucked up a perfectly good spreadsheet is what you’ve done. Look at it. It’s got anxiety. (Originally @julesfern.gay)

1122

Russell Pierce @russellpierce.bsky.social · 01/12/2024

> this is one of the main drivers for ever new "To Do" apps: the existing app gets full because deleting is too hard, so we start fresh with a new app. The app isn't the point, the starting fresh is. Ugh too much truth for Sunday morning mp (news.ycombinator.com/item?id=4228...)

news.ycombinator.com

Yes...but it goes even deeper. For example, in physics, the paradox of Maxwells ... | Hacker News

151

Russell Pierce @russellpierce.bsky.social · 29/11/2024

My villain origin story: the Zoom Lens theory of attention (Eriksen & Yeh, 1985) reports a sig main effect in one condition and non-sig in another... and goes about interpreting it as an interaction to support their theory. Reanalysis shows not even a particularly noteworthy interaction effect size.

000