Julian Dunn's Stack @juliancdunn - Tumblr Blog

Posts

Titus Winters writes in Software Engineering at Google (O’Reilly) “Software engineering is programming integrated over time.” Expect the impact of your software to stick around.

(Excerpted from Tanya Reilly’s “The Staff Engineer’s Path”)

#software

Another great episode from Cal Newport. He posits that the biggest impact AI will have on work is not massive white collar job loss, but making those jobs miserable. That's because white collar managers often already evaluate their employees' productivity using bullshit proxy metrics that incentivize busyness, or the appearance of productivity: how many e-mails you are sending today, how "present" you are in Slack, whether you are visibly working at all hours, and so on. It's already exhausting and about to get more so if these AI tools are here to stay.

I wish Newport didn't focus so much on what individuals can do, but in the absence of more systemic change I guess that will have to do. His five suggestions to resist the urge to jump on every stimulus and ultimately get nothing done are:

Plan weekly. On Monday morning, ask yourself what important things that create unambiguous value for your organization must get worked on that week, and prioritize them by finding & protecting time on your calendar. That might involve canceling or rescheduling certain meetings to do so.

Create and maintain an ongoing portfolio. This is a document listing all important initiatives, projects, or accomplishments you are responsible for, so that you can counter the narrative that you don't "look busy enough". The doc should list accomplishments by month, quarter, etc. and include impact / positive consequences. It's a tool you can bring to quarterly reviews, or show to your managers, or use as an instrument to ask for feedback on what should change.

Avoid what AI can do. If a task is something that you could largely automate with LLM queries, then that is a task not making the best use of your unique skills, training, and knowledge. If most of what you are doing is just automated by AI or could soon be automated by AI, then you are vulnerable and bringing it upon yourself.

Pursue upskill projects. Always have some sort of new skill that you are acquiring that's valuable and relevant to your job that will make you more rare and valuable. If possible, connect this to something you are working on for your job. The harder the skill is to acquire, the more you can escape the trap of "AI-accelerated pseudoproductivity" because it's hard-won value that AI can't replicate.

Care about your writing and write well. This is the most straightforward way to differentiate yourself from the LLMs. Take the time to write well; make your emails, reports, or any professional written document extremely clear, succinct, and well-crafted. When everyone else is sending out long reports with bullet-point lists and emojis attached to everything with convoluted AI-generated language, you are sending out clear and concise communications that have a point and for which its meaning is crystal clear. Set yourself up as an alternative.

Newport argues convincingly that this is the time to stop relying on visible activity as your main marker of value, and instead rely upon actual hard-won accomplishments that you did and can point to. Do the hard work of doing the hard work.

#ai #Youtube

I Profile Celebrities for a Living. Nothing Prepared Me for Tilly Norwood.

https://www.nytimes.com/2026/05/31/magazine/ai-actress-tilly-norwood.html?unlocked_article_code=1.mlA.YX-h.f6FNjSsACQJZ&smid=url-share

Surprisingly not as terrible a "profile" as I thought it would be. The writer, Taffy Brodesser-Akner, found that "interviewing" the AI made her feel exhausted, like being on a computer all day. Some key observations I had:

Most of the AI's responses were not particularly insightful or compelling, which you wouldn't expect them to be since it's a text extrusion machine that outputs the mean, but maybe people would find them more compelling if they didn't know they were coming from an LLM? Not every interview subject (and celebrities, certainly!) says things that are insightful, but we often ascribe more insight to them because they are coming from a human.

If this is true, then I believe this will lead to more content producers hiding the fact that output is AI-generated, because people get that ick factor/uncanny valley effect when they realize they are interacting with a robot. Such producers -- Hollywood executives, for example -- have no incentive to clearly label output as AI, because if AI-generated content can engender emotional reactions in audiences (so long as they don't know it's AI), then this is a margin-maximizing machine. The only risk is of detection, and AI-generated content is getting harder and harder to detect, so many producers may decide that the risk of a PR disaster is worth it.

SAG-AFTRA's head, Sean Astin, had such a milquetoast reaction to AI "actors" as to be laughable: "The reason that a synthetic construct, an algorithmic output, will never take the place of a human actor is because it is not a human actor.” Yeah dude, that's tautological; you need to say more. You'd think he would do something like explaining that only human actors can create unique emotional experiences and reactions in audiences, unlike language models that essentially just parrot existing patterns in their training data, etc. Or even saying something like audiences will demand to see humans and not AI because humans value connection to other humans, or something like that.

Like I said, I gagged less at reading this article than I thought I would, though as per usual I wished the journalist had been more skeptical of the AI industry's claims. No "Tilly Norwood" or any "AI actor" is capable of the types of misalignment that the AI industry spokespeople claim that it will (become "superintelligent" and kill us), yet Akner doesn't push back on this at all and instead makes what I assume are jokes about "Norwood" killing Meryl Streep. But these jokes are misguided and I think lend credibility to this ludicrous conjecture than is warranted.

#aihorseshit

Don’t want to link to the fascist microblogging site so took a screenshot instead. This is very true, coming from Aaron Levie (Box CEO) who would know.

#aihorseshit

This longtime con man was doing deepfakes long before that was a thing. He's a bot in human form

#ai

Meta Is Dying. It’s About Time.

https://www.nytimes.com/2026/05/08/opinion/meta-facebook-zuckerberg.html?unlocked_article_code=1.lVA.v9Aj.wFKKAFdFuB_Y&smid=url-share

#facebook

Signet records that you wrote, not what you wrote. Tamper-proof certificates of authorship for student writing in Google Docs.

First use case is for students to prove that they wrote papers, rather than copy-pasting from a chatbot, but I can see something like this becoming more widely used

#ai

“Book on Truth in the Age of A.I. Contains Quotes Made Up by A.I.” What a dumb own goal…

https://www.nytimes.com/2026/05/19/business/media/future-of-truth-ai-quotes.html?unlocked_article_code=1.lVA.97Uj.inQcWskitKAB&smid=nytcore-ios-share

Also,

“These A.I. errors do not, in fact, diminish the larger questions that the book raises about truth, trust and A.I. and its impact on society, democracy and editorial,” he added.

Well yes, in fact they intensify them.

#aihorseshit

Remaining vigilant about cognitive debt and atrophy.

“People who go all in on AI agents now are guaranteeing their obsolescence. If you outsource all your thinking to computers, you stop upskilling, learning, and becoming more competent.”

– Jeremy Howard, creator of fast.ai

#ai

Doesn’t look good for 50% of hyperscalers’ RPO to be made up of AI company spend… when that spend is basically coming from VC. (Source: The Information)

#aihorseshit

Subscription and open access journals from Sage, the world's leading independent academic publisher.

danah boyd on how social media became parasocial media.

I wonder: Will people simply abandon the Internet for real human connection now, since any tool seems like it will get co-opted by billionaires?

#social media

I have seen this too. Like Mitchell, I find it sad because people I previously respected as rational thinkers seem to have lost their minds over this technology.

While I’m not a complete AI skeptic (I have found some use for LLMs for software development) I think its utility is like 5-10% of what the hype would suggest is true.

#aihorseshit

AI in Computer Science Education — Oxide and Friends — Overcast

How Brown University professors are redesigning intro CS curriculum for the LLM era. The results may be more surprising than you would think (undergraduates who truly want to learn, once exposed to LLMs and their limitations, actually use them much more carefully).

#ai

Referenced by Valerie Veatch in Ghost In the Machine.

#aihorseshit

Valerie Veatch’s work will make you feel all kinds of things: dismayed, icky, but also empowered with the knowledge that we are not, in fact

Glad to see that "Ghost In the Machine" is getting more airtime and awareness. (I still haven't seen it, though.) This is a good quote:

"AI resistance is also actively being co-opted by big tech. Some AI doomers might look like they have things in common with AI resistance, but their whole point of view is that AI is going to become a super-intelligent godlike entity. It distracts from where the actual power sits and accommodates the actual harm being done—like horrifically racist predictive policing algorithms in justice systems. This isn’t existential; it is actual harm. These doomer groups get all of the funding and media attention, then issue vague statements about phoning your senator and regulating AI. It’s a waste of political will and energy. I think it will become increasingly clear where and how real AI resistance will be centralized and what it will look like. I think that by next year, we’ll look back at our 2024 and 2025 obsession with AI, and we’ll giggle. The limitations of these systems are already becoming clear. And in 10 years, certainly, it will be hilarious."

I am also in complete agreement with the statement that "local compute" will be the future (and even if language models are still used in one's workflow, not depending upon enormous data centers is what will keep costs manageable): "I think that overall, not engaging a hyperscale data center in the production of your work and relying on local compute—and insisting on software that allows you to do that—is something we can all do."

#aihorseshit

#ai

Lessons from the 1990s human cloning debate

#aihorseshit

Trending Blogs

Recently Viewed Blogs

Julian Dunn's Stack