Radiology is a specialty that was long promised to be automated via ML, but that still persists today. There are fascinating studies on AI/human joint performance around it. I appreciate this article, and the distinction between benchmarks settings and the more situated nature of real work.

The rebar3 -> rebar4 kickstarter ends in a few hours! Late pledges stay open, but rewards (T-shirt, mug, stickers) disappear with the campaign. We're seeking a more sustainable future for #Erlang build tools, and this campaign helps support this goal: www.kickstarter.com/projects/pee...

I was doing some software history research and stumbled on this absolutely FASCINATING letter from 1964: dl.acm.org/doi/10.1145/... Some random defense contractor writes in to say "You should deliver a minimal prototype as fast as possible to get feedback and involve users at every stage of labor"

This analogy was in my head for months and I had to write about it. Treating incidents as "fix it and move on" is a losing proposition. It's more useful to see them as emergent outcomes of all the tradeoffs we make, and to use them as "landmarks" to orient ourselves in a solution space.

New SRE team swag is in

Software incidents are painful, and we're trying to help change that. If you deal with incidents in your work, please help us help you! Take the survey here: www.thevoid.community/survey

Proud to back the Rebar4 Kickstarter — moving the BEAM ecosystem forward with the community. 🙌

‘helm’ has a levenshtein distance of 1 from both ‘help’ and ‘hell’ and I think that’s on purpose

Fred's rule about posting: You can either write long, contextually rich texts that are unambiguous but also won't be read much because they're too long, or shorter but engaging texts that are going to be interpreted in ways you disagree with, and there's no middle ground.

"Behavioural science is unlikely to change the world without a heterogeneity revolution" is this week's paper, arguing that if we stopped chasing main effects and considered diverse context-dependent dynamics as the norm, policy research could get better models. Notes at: ferd.ca/notes/paper-...

The Leftover principle is a consequence of “automate what you can and what is left over is done by humans”, implying people are left with trivial tasks not worth automating and tasks too rare or complex to automate. Assuming this can be done without negative impact to leftover tasks is often wrong.

In my second post of key questions for developers, we dig through the insights from multiple @ferd.ca posts to determine how products evolve and the environments that facilitate that process. programmersstone.blog/posts/what-a... #ProductDevelopment #PsychologicalSafety

designing truthful bumper stickers

I’ve sowed white sweetclover for the last two years in the strip of soil next to my house and they’ve never even managed to sprout there. Yet these fuckers are growing wild right outside my fence, taunting me.

You know videogame bad guys are software folks because it starts with a tiny issue (your hero) & they keep delegating to straightforward minions but as the problem grows they have to churn ever fancier solutions until final stage where they must fight you themselves but by then you’re way too strong

I finally came to the realization that this was never supposed to work. It is meant to make me feel stupid. Someone at AWS successfully built a set of features whose sole purpose is to trap engineers into hating themselves, a self-service bullying machine, and they got me.

Skimming through a preprint on the risks of using LLMs as therapists, then reading tech commenters who were even less thorough. Aside from “the poor demand harmful crumbs”, there’s a surprising amount of folks claiming they’re too informed to be harmed—the “not me, I’m built different” defence.

Probably am at the point where coping with events by going “maybe this will finally make people react” doesn’t work anymore.

New paper hot off the press www.nature.com/articles/s41... We analysed over 40,000 computer vision papers from CVPR (the longest standing CV conf) & associated patents tracing pathways from research to application. We found that 90% of papers & 86% of downstream patents power surveillance 1/

using this to talk about llms and adaptive behaviors part of what people mean by "chat isn't intelligent" is that core pieces like "noticing that gassing your homie up is hurting them" are simply not present

I was seeing unusual weeds in my lawn, until I noticed they were actually lettuce. I grow a heirloom variety in my garden and I let them bolt late in autumn to harvest seeds for the following year. They’re apparently so well adapted to local conditions they just outcompeted dandelions and grass!?

Big fan of the phishing emails “intended for all ferd” that are announcing layoffs through the HR department of my own personal domain name. Guess I’m not gonna be allowed to write my own blog anymore then.

Looking at "Empirically derived evaluation requirements for responsible deployments of AI" The paper points out you should NOT evaluate AI in isolation, but empirically, with people, as a joint system, and with the whole range of results, including errors. notes: ferd.ca/notes/paper-...

I bought a pen that ended up shaped to only be comfortable with the dynamic tripod grip, but I have always written with the lateral tripod method. So I’m doing the most logical thing and I decided to learn a whole new writing grip, messing with decades of muscle memory.

ooh, I like this thread from Justin that, from a research work perspective, adds some nuance to my last blog post regarding how even if you have a good explicit frame to what you want, re-contextualizing as it happens while doing the work is often critical, and hard to scaffold for.

Why do some have a shit time with LLMs for programming while others love it? To succeed, the latter group tacitly creates tons of scaffolding and gain weird new skills. While it works, this posts explains how doing all that is an incidental consequence of bad interaction design in coding AI agents.

The other thing I’d say that pops to mind is that systems with this inadequate type of designs are constantly popping up and find ways to be commercially successful, they just also happen to probabilistically fail and cause harm in predictable ways.

Reading yet another post asserting all you need is a human doing careful continuous supervision of the automation and it’s gonna be perfectly fine, as I grind my own teeth into dust while 40+ years of research and industry reports flash through my mind like war footage in a ptsd movie scene

James Reason taking no prisoners in Managing the Risks of Organizational Accidents

Just starting to read Dara Z. Strolovitch’s “When Bad Things Happen to Privileged People” and it’s obvious early on that it’s gonna rule

same but medical reasons, so the debate boils down to "should disabled people have jobs?" without anyone realising it

The RTO conversation always feels sad from my living-in-a-remote-place point of view. Doing my job without remote work implies any of: - uproot & drop everything and people in my life - forego more than 2/3 of my compensation - switch careers There’s not much to debate for me in this situation.

I must once again advocate we use the term “stochastic errorism”

The more theory you read the more you grow a collection of cool models for all things you can use to reason about and apply to various situations. But also the more it feels akin to doing ethics by just picking whatever framework is most convenient/satisfying right now, and maybe you’re not good?

Took some time out of vacation today to publish Rebar3 3.25.0 for #Erlang 28.0 compatibility: github.com/erlang/rebar...

MCP rules because all the tech orgs trying to embrace AI for their board/investors but couldn’t find a way to make it useful to end users can instead ship an MCP server and now it’s a legit AI thing they’re doing to satisfy the demands without having to deliver new product value (although it could?)

I realized something recently, which is that we're building AI tools completely backwards. While I was aware that there's some de-skilling happening already, I finally clicked a bunch of pieces together and decided to write this to talk about how we can do *better* hazelweakly.me/blog/stop-bu...

This is also why I push for engineering organizations to have EVIDENCE strategies not just "measurement" or "metric" strategies. Evidence strategies include qualitative insight! Different FORMS of evidence and understanding whether we are or aren't getting them is evidence strategy!

corgi is the plural of corgus

Gonna have to create a new kind of Quebec-centric sci-fi dystopian genre focused on AI, inequity, and societal collapse, and it will be called “ciboire punk”

The best customers are ones that make your product better. Huge thanks to the @honeycomb.io team for co-designing new PagerDuty insights with us! Special shout-outs to @ferd.ca and @another-engineer.bsky.social in particular. 🙏

SRECon Americas videos are up, which means you can watch the closing plenary @charity.wtf and I gave—an open letter to AIOps vendors if they are to write better tools. It also turns into an open letter to SREs, because we don't expect vendors to all listen to us. www.usenix.org/conference/s...

I was trying to write a quick post here about how I'm taking more time off my usual online stuff after how demanding the last few months were, only to find myself writing more and more and going "this could be a good work blog post" and oh yeah of course I'm in this situation, I just keep doing this

overhearing LLM programming chat and this is all I can think about

While people can feel like they’re getting better and more productive with tool assistance, it doesn’t necessarily follow that they are learning or improving. @ferd.ca www.honeycomb.io/blog/ai-wher...

I know all this power and identity stuff is hard. But don't you just want to TALK talk about what it means to be in tech right now? What it really all means and why this is happening to us? Because I do. We need the talk story. It is the only way we can find something different

I really am growing tired of the "be careful" words of warning from the folks running the knives-without-handles factory.

Who's coming to SRECon this week?? @ferd.ca and I are doing the closing keynote! 👯 And @honeycomb.io is hosting a happy hour on Wednesday along with our friends from @buildkite.bsky.social, @spacelift.io, Oso and Cloudsmith. Register here 👉 events.spacelift.io/off-call-on-... ... see you soon!!

This week's paper, digging further into Work-as-Imagined and Work-as-Done through Lucie Cuvelier's text, comparing them in what I loosely translate as "when resilience engineering questions ergonomics once again," which compares Resilience Engineering's perspective with French Ergonomics concepts.