dar-mit

dar-mit

@dar-mit.bsky.social

Just getting started…

20 followers37 following65 posts

dar-mit
@dar-mit.bsky.social

@anthropic.com Just a little note of thanks for being able to release Fable 5, and Sonnet 5, and for resetting weekly totals. Your team catches a lot of flack, so I thought so balance was justly deserved.

dar-mit reposted

Ceinwen Langley
@ceinwenlangley.bsky.social

I know Calvin’s dad is already having a moment, but every terrible “outsource precious and fleeting moments with your children to tech so that you can spend more time on your work” pitch makes me think of that one strip I never really cared about as a kid, but care a lot about as an adult.

This post is unavailable.

dar-mit reposted

Electronic Frontier Foundation
@eff.org

There are billions of reasons not to turn Meta’s customers into a distributed surveillance machine. This whiplash behavior proves exactly why we cannot rely on the "good will" of Big Tech to protect our digital rights. www.eff.org/deeplinks/2...

eff.orgVICTORY: Meta Strips Facial Recognition Code From Smart Glasses AppJust days after a damning WIRED report exposed that Meta had quietly embedded facial recognition technology (FRT) code into millions of phones, the tech giant has quietly acquiesced in demands to
dar-mit
@dar-mit.bsky.social

@anthropic.com Please keep #Claude.AI, Claude: Me: If I had a dollar for everytime I've heard, "exactly the bug this phase fixes" in the last two weeks I'd have like 10 dollars. TEN, I say...TEN DOLLARS! CAI: $10. A full Hamilton's worth of "exactly the bug this phase fixes." 😄

dar-mit
@dar-mit.bsky.social

CAI: In fairness, this one earned its keep — the guard literally deny'd my own <<<'... rm x' probe mid-investigation, which is about as on-the-nose as a bug demo gets. Soon to be $10 and _a working lexer_, which beats Alexander Hamilton's actual outcome.

dar-mit reposted

AndersOSU.bsky.social
@andersosu.bsky.social

Replying to @lessbadproblems.bsky.social

Here’s a good example of where transformers are playing a significant role in scientific advancement arstechnica.com/ai/2026/06/o... The article does a good job discussing what the latest models do well (synthesis of ideas from distant fields) and what they do poorly (develop new methods)

arstechnica.comAn OpenAI model solved a famous math problem that stumped humans for 80 yearsI tried to explain OpenAI’s solution more clearly than OpenAI did.
dar-mit
@dar-mit.bsky.social

Well FUCK YOU Microsoft! I our haves a fully licensed suite of apps with the knowledge that they couldn’t be upgraded to the next versions. Which was okay. This is not okay!

Slashdot@slashdot.org

Microsoft Deliberately Bricking All Office For Mac 2019/2021 Installations https://tech.slashdot.org/story/26/06/02/1725210/microsoft-deliberately-bricking-all-office-for-mac-20192021-installations?utm_source=rss1.0mainlinkanon&utm_medium=feed

tech.slashdot.orgMicrosoft Deliberately Bricking All Office For Mac 2019/2021 InstallationsMicrosoft Office 2019 and 2021 for Mac will reportedly drop into "reduced functionality mode" on July 13, 2026, when a license-validation certificate expires, leaving perpetually licensed apps able to open files but not edit or save them. Slashdot reader joshuark shares a report from OSnews: "Micro...
dar-mit
@dar-mit.bsky.social

3. Including robust auto: error checking and recovery, full Git|Hub bidirectional flow, & a workshop for Skills, Hooks, Agents with testing and deployment protocols. 4. Memory/ is a full Code, Library, Documentation, Reference depo & a 6-part LLM Wiki with…

dar-mit
@dar-mit.bsky.social

…auto ingest, classification, atomization, reliability ranking, memory tiers and supersedes|superseded tracking to support auto-lookup at SessionStart for context injection. 5. Memory/ uses the same error checking as System/. 6. Plus two launchd crons to support overnight Memory ingest. IN 4 DAYS!

dar-mit
@dar-mit.bsky.social

#MyLifeWithClaude Day 16•Husky CONFIRMED! Sigh… Spent a fair amount of time researching why #Claude.ai creates code, fails, jumps to a different path, and repeats this pattern. It’s trained w/o doubt. Updated CLAUDE.md to stop it from derping like that, and to better structure planning sessions.

dar-mit
@dar-mit.bsky.social

#MyLifeWithClaude Day 17•Started coding this morning and blew through my 5 hour limit of tokens in <50 min.! Wrote 30 lines of code. Turns out having a 19KB CLAUDE.md ain’t great. Learned a lot about how tokens are used w/ .md file. Current md file now <5KB. Then found .xml won’t work for task. 🤨

dar-mit
@dar-mit.bsky.social

#MyLifeWithClaude Days 1–15 Day 1•Dear GODS he’s like a golden lab who’s so happy to help, loves you to bits, knows just about everything and is dumb as a stump. (AKA My 1st chatbot) Days 2–10•It’s like talking to the internet! Great chat RE: QFT, Null Geodesics, Photonic Messages. Started Coding…

dar-mit
@dar-mit.bsky.social

#MyLifeWithClaude Days 11–15•The worlds of DOMs, JSX, ExtendScript & XLIFF crashed HARD! He knows code, he doesn’t know how to think ahead. Labs are like that… I hope he’s not a Husky! We had a LONG chat tonight. Updated 3 CLAUDE, Default- & Dev-KB MD files. XLIFF > XML Hashed Style Sheets.

You are up to date.