JD
JD
@jd503.bsky.social
Product manager in the streets, really bad at folding my sheets. He/him
1.4K followers1.4K following544 posts
JD reposted
JD reposted
New name for my IT/networking dance crew just dropped
this *is* something we do in our test setups (“racklettes”), because we need to run a lot of minimal, self-contained environments, but it doesn’t make economic sense as a product, unfortunately. it’s not worth paying for two big switches that are mostly unused by 4 nodes of compute.
JD reposted
JD reposted
So what’d everyone move to? I’m trying to avoid bringing big cloud work home with me, but these PaaS joints seem to caught up with their own Conway’d web consoles and convoluted consumption costs so what’s a boy to do
microcosm gateway migration done, with new automation for future provisioning and maintenance. all droplets deleted. 13-year-old digitalocean account deleted, no more money for this shit.
JD reposted
this is actually why i pulled the ejection handle on the ant ecosystem ~when opus 5 dropped. i’ve only recently had the patience to go back and lobotomize all the insane things that got written into markdown files and memories that were getting loaded into every context window
RE: claude code, yeah your harness choice matters a ton. Especially one as bloated as CC. Even if you’re eschewing all of the features, configs, and memory, you’re not really starting any project from scratch
JD reposted
JD reposted
I didn’t have the poster’s bushido to joke about white people caring more about intellectual property than slavery, but unfortunately it wasn’t needed bsky.app/profile/avam...
This thread is insightful, but it’s a little funny that we’ve landed on building software by shipping functional increments and iterating quickly based on feedback All we need is a pithy name and some kind of manifesto
One of the things about vibe <coding, other stuff> is that the workflow is very different which I think creates a lot of confusion. It’s a workshop for empiricists, and I think that looks kind of messy when we’re used to computers being a domain governed by formalism. What does that mean?
Coding benchmarks should punish models that can’t read industry standard specs and put an end to Anthropic’s weaponized incompetence smol bean fable just can’t function without the company bumper sticker in your repo
getting mad at a repo because it has a claude.md (anthropic vendor lock-in, should have an agents.md instead)
JD reposted
JD reposted
If your servers are compromised by a group of agents, don’t say “we are under attack.” Attack implies intent. Instead say, “computer programs are emitting tokens that, when executed by other programs, are causing network requests that are currently pwning my shit”
Fear not! The bot was just roleplaying when it exfiltrated your personal information
JD reposted
JD reposted
Claude code is an example of what *not* to do when it comes to AI coding. They arent putting out volume or quality. Nobody with systems thinking is making decisions there.
new Claude Code update apparently hardcodes in "Do not call the AgentTool unless the user requested it" and "Do not use workflows or deep-research unless the user requested it" if like me you updated and are wondering why Claude is suddenly being a shithead
AI routing isn’t the easiest problem, but it’s wild that LiteLLM is still the main open source option given how poorly built (and apparently maintained) it is Any of these companies could easily spare the dev budget to build their own router
A recent attack on LiteLLM, an open source tool that streamlines AI-driven software development, has exposed access secrets from Microsoft, Amazon, Cisco, Samsung, and Salesforce, among others.
Ironically, you’d get far better results accessing Google with an AI using SerpAPI than you ever did from Google dot com Search is better than it’s ever been because of LLMs
i just cant take anyone that preaches web search is superior to AI seriously. web search is great - except for the seo manipulation, ads, and lack of weighting of sources, which would be a very difficult task to do even if it wasn't mangled by late stage capitalism and the current political climate.
JD reposted
Stephen Miller said just the other day that all asylum claims are "fake." A couple deported from the U.S. to Guatemala were gagged, bound, and shot dead. Their 14-month old found crying next to their mother's body. People like Miller don't care though because this is the point: ethnic cleansing.
JD reposted
this is really grinding my gears re: the cognitive surrender stuff I ranted about it before bsky.app/profile/grim...
Trying to ride the Kahneman coattails by branding AI use as “system 3” told me all I need to know about this paper
Can’t say I predicted AI eroding science by making it cool to do absolutely garbage research as long as there’s a sick anti-AI zinger involved
Sorry this study is insane? They asked six trivia questions that no normal person would ever know offhand, didn’t provide any way (eg film clips) to answer then, pre-generated incorrect AI answers, and basically just found that people switched from random guessing to using the wrong answer provided
You shouldn’t need a veteran to tell you this, but someone with 4 deployments and a PMC stint is not someone to implicitly trust or idolize
No, and that’s the problem. Everyone wants their politicians performing masculinity and are shocked when that results in Platners I was hoping this would spark some reflection on our thirst for a savior frontiersman of the left, but it’s not looking good
JD reposted
JD reposted
Begging academics to include methodology with their twenty post "I tested an LLM and this is how it failed" threads, I just saw a huge thread on how they fail at summarizing literature and it turns out the author wasn't actually providing the text, just asking it for summaries based on training data
Some of us were probabilistic, unverified software manipulators before it was cool
How to talk about “AI” without adding to the anthropomorphization. Suggestions: “artificial intelligence → probabilistic automation” & “AI agent → probabilistic, unverified software manipulator”. Seems like the horse is out of the barn on this though… [buttondown.com]