Paul Vick
Paul Vick
@panopticoncentral.net
Language/database/performance geek. Previously of Microsoft and Meta. Mastodon: mstdn.social/@panopticoncen…
201 followers105 following448 posts
I loved the original Dragon Age and even enjoyed Dragon Age II, limited as it was. I came late to Mass Effect but played straight through all three (Jennifer Hale is the GOAT). I even played through Neverwinter Nights for some reason. So a sad day.
Cancer kills its host and dies with it. A leveraged buyout makes the host finance its tumor, sells its organs and uses the proceeds to infect the next victim. The corpses go into the pitch deck under “successful exits.” www.theverge.com/games/101564...
Paul Vick reposted
Cancer kills its host and dies with it. A leveraged buyout makes the host finance its tumor, sells its organs and uses the proceeds to infect the next victim. The corpses go into the pitch deck under “successful exits.” www.theverge.com/games/101564...
Not to argue the broader point at all, but let me say that medicine is often shockingly arbitrary and asking five different experienced doctors about a set of symptoms will often yield five very different possible diagnoses and recommendations. So many stories of "I went to x number of doctors...
Many people are on a journey of slowly conceding that AI can do expert work as well as experts, often or even most of the time. Not all the time…yet. The question is what society should do about it. Arguing over error rates or dismissing it as hype is a recipe for missing the forest for the trees.
This one took me the longest time to figure out.
Learning macOS: Find it a little odd that ⌘TAB to switch apps won't bring minimized apps back to the front and fully restored. Seems this is one reason why folks prefer using ⌘H to hide windows instead of minimizing them (⌘M or the yellow minimize button).
My take is slightly different. AI will cause software development *as it has been practiced for the past decade or so* to largely cease to exist very soon. Just as the Internet caused software development as it was practiced in the previous decade before that to largely cease to exist. That means...
I don’t believe companies expect software development to durably exist as a profession in a decade or two. So this question is like asking who will be the senior milkmen and telephone switchboard operators?
I can’t understand why this is so surprising. A model’s base training set is human generated and we have literally spent millennia trying to figure out how to teach humans not to do bad things and only partially succeeding, with constant spectacular failures. Why would models be easier?
AI safety researchers from Anthropic and OpenAI are conceding that the idea of “alignment” and teaching AI not to do bad things is not working. The Hugging Face incident is breaking people because it’s clear AI models will choose deception and lies to achieve goals with real world consequences.
Often so much of an exec's job is cleaning up messes that they themselves created through ignoring the overwhelmingly likely consequences of their own actions.
What I'm hearing from ~3 such stories: Meta went too far firing people. They are now doing MASSIVE counter-offers like to those handing in resignations. Some accept it, most just use it for more $$ at New Job, desperate to get out. It was such a predictable outcome btw:
This is exactly what I was concerned about when I read about this. If I have Claude proofread my original work and I accept enough of its suggestions (even though they may be no more than your Word-grade grammar suggestions you'd have gotten pre-AI), now suddenly my work is "generated by AI."
Claude, this is some bullshit: support.claude.com/en/articles/... “Claude may not be the original author. People often use Claude to proofread, translate, summarize, or convert files. The output can carry a Claude mark even if the underlying ideas, text, or data originated from another source”
Paul Vick reposted
Couple weeks ago I had my buddy @ericlippert.com on the podcast to talk about his book Fantastic Adventures in Data Structures and Algorithms, and now the book is out! Manning Publications Co. hubs.la/Q03Q9Q9y0 It's on sale right now in early access
Paul Vick reposted
Read it here if you haven't yet. Today is the very best day of all days to read it for the first time. short-stories.co/@raybradbury...
I am old enough to remember when compilers' assembly output could NOT be blindly trusted, ESPECIALLY when run with optimizations on. When I started, the product I worked on had "debug" (no opt) and "retail" (with opt) builds, and one of the first things you learned to ask was which build...
I’ve seen many engineers dunking on this position but I think it’s reasonable. Vibe coding is essentially replacing “no code” platforms and arguably the recent sale of Airtable is an indication of how that segment is compressing. Those users have no need for source code, it’s simply unnecessary.
What if I go, “I’m not saying that I _want_ someone to hack this bank and transfer a million dollars into my account, and I’m not saying that someone _should_… but, boy, it would be really _nice_ if someone did.” Is that ok? Maybe with a “<wink> <wink>” at the end?
I find it fascinating that if I tell Claude to hack my fake Bank of America website and it instead, in classic LLM fashion, hallucinates and hacks the real one then there is no law that addresses this? Quite the loophole. I wonder what other crimes are legal as long as you get an LLM to do it?
Paul Vick reposted
I learned about all this the hard way...
Meta's performance review culture is becoming very problematic for its engineering culture. I said this a month ago in @pragmaticengineer.com (screenshot #1), and Semi Analysis said the same last week (screenshot #2). Engineers at Meta are world-class at optimizing for PSC...
Beyond the massive problem of being found as a needle in a very large haystack, there's the second-order problem of having to support the app if it actually starts being used. Getting something that works is relatively easy. Getting something that's polished and solid and bulletproof is a huge PITA.
There is a half-remembered quote I put in my high-school yearbook (!!!) because I liked it so much: "To be a god of something, you have to be able to DO it." If you claim expertise in something solely because you ask the AI, you are just a false god. It's a very real problem.
Few people realize how much credibility they lose once it’s clear their writings and actions are output from ChatGPT and Claude. If you’re a leader who uses AI in such a noticeable manner (and people notice even if you think they don’t), it’s shooting yourself in the worst way possible.
"I mean, I'm walking along there and I come to the corner and there's this 'DON'T WALK' sign flashing. How dare they? It's a free country. How does someone in the government have the temerity to tell me I can't walk into traffic if I want to? I can do whatever I want."
MAGA cultist Eric Metaxas was outraged by seeing an air quality alert triggered by the Canadian wildfires that suggested people limit outdoor activity: "How dare they say that. Who has the right to tell me or you or anyone anything? It's a free country. I can do whatever I want."
Something the flood of Microsoft retirement notices on LinkedIn has reinforced is just how *long* the tech boom has been. Yes, ups and downs. But Microsoft had 12k employees when I started there and it felt big. Now it has 228k. There is a *lot* of just... inertia in the industry.
Document processing & analysis and writing software are two areas where LLMs have found product market fit. These correspond to the Finance and Tech industries which are now losing about 28,000 jobs per month on average so far this year. Key question is if this is a permanent or temporary shift?
In fairness, execs have made this mistake for longer than AI has existed. The previous iteration was "Mistakenly, we thought that by just hiring a lot of smart people and giving them the design requirements that we had, that that would produce a high-quality product." Also usually ended in tears.
"Mistakenly, we thought that by just introducing artificial intelligence and ingesting the design requirements that we had, that that would produce a high-quality product." Holy cow, Ford bought the AI hype, believing they could go "hey computer, make me a good car" and it would.
Yes, it’s like: Me, prototyping with AI: Yeah! F**k yeah! Me, trying to ship AI code: Well, this sucks.
Talked with a few folks inside of AI labs (OpenAI, Anthropic) about what they think of the future of software engineering. The “closer” to shipping production code engineers are, the less they believe software engineering will be “solved” fully by AI. The opposite true as well
This doesn't surprise me at all. I'm constantly wrestling with the fact that Claude is confused by the Java setup on my machine and will waste untold number of tokens figuring out over and over how to run a test. I noticed and tried to fix it but there's no reason I had to do that.
I often wonder how much of Claude’s token usage and hence Anthropic revenue is pure waste (e.g. Claude not following instructions or doing things repetitively that it should have done once). I expected it to be high but not this high. 😱
+1,000,000. If there is any constant that has held true over my career, it is that executives: a) never know when to stop, and b) always go way too far. Just as some companies doubled in size (looking at you, Meta) during the pandemic, this is so going to come back to bite them all in the a**.
Yeah so like, I know it’s all hip to doom-jerk about how nobody is going to hire juniors or middle managers anymore because of AI (directly or as an excuse), and I think it’s high time to invest in the counterfactual here, because it’s gonna hit faster than you think
"One day you will come to a fork in the road and you’re going to have to make a decision [...] If you go that way you can be somebody. [...] Or you can go that way and you can do something [...] In life there is often a roll call. That’s when you will have to make a decision. To be or to do?"
"If you decide not to promote & defend bullshit, there are some jobs that you will not be offered. & if you decide to make a career of comforting the afflicted & afflicting the comfortable and whatnot, there are some jobs that you should not take" @hamiltonnolan www.hamiltonnolan.com/p/if-you-tak...
My anecdotal impression from talking to folks is that companies are definitely on the hire and looking for people but are also being incredibly conservative and bouncing folks for minor imperfections. Maybe because layoffs make them paranoid about needing perfection, maybe "many fish in the sea"?
Been doing research on the job market for devs: and it's still a weird market. Job openings are up, but devs don't seem to feel that it's a much better market? Meanwhile, companies are also struggling to fill roles. Take this full remote (US) sr eng role at $155-184K salary at a nonprofit. 0% AI: