yeah I’m gonna start blocking people for being shitty to Will Stancil, a person who I do not follow who does not follow me. the slashed tire is fully nuts and if you’re part of the problem from here you’re likely contributing to whatever fever swamp conditions caused it
potatolicious
@potatolicious.bsky.social
I make ML stuff at . Opinions my own.
I can attest from first-hand experience this is true. RTO flexibility in tech is now uneven and largely reflects political clout of specific managers and orgs. Informal power structures have never been more critical for managers to wrangle.
the stick approach also inevitably incentivizes well-meaning managers to break the rules to accommodate their reports’ needs, which leads to widespread inequities across the org, which is a pretty obvious outcome of a dumb hammer approach
this is just extremely plainly cartel behavior
OPENAI, ANTHROPIC STAFF CIRCULATE PETITION URGING US TO SUPPORT INTERNATIONAL EFFORT TO 'DELIBERATELY PACE' FRONTIER AI DEVELOPMENT, WARNING OF RISK IT ADVANCES FASTER THAN HUMANS CAN CONTROL; LETTER ALSO SHARED WITH GOOGLE DEEPMIND STAFF, EXPECTED PUBLIC THIS WEEK
Michel Foucault: I am describing the philosophical concept of the Panopticon as a warning about authoritarianism and it's effects on society The 21st century: oh boy, I can't believe we finally democratised the Panopticon, this will only have positive consequences
Everyone besides the couple involved in this does not know the love of god nypost.com/2026/07/24/u...
Working a factory floor in London, Ontario as an engineering intern is honestly low-key one of the most formative experiences of my life. And I wasn't even on the line all day every day.
this is part of the reason complaints about “bullshit jobs” and “email jobs” have never resonated with me, i have *had* the “real” jobs people fantasize about and they largely suck
unironically: this is the kind of thinking that comes from doing an INCREDIBLE, bona-fide-unbelievable, how-are-you-even-alive amount of drugs and calling it a "stack" in the interest of becoming the guy in Limitless you're not a visionary, you're coked out of your fucking skull Chad
JFC. Chad Jones, economist who'll be joining Anthropic: "In other words, with log utility it is optimal to take a 1 in 3 chance of ending human existence in exchange for a 2/3 chance of dramatically raising living standards by a factor of 55." www.ft.com/content/bb04...
Good thread. I think two things can be true at once: - Highly skilled, competent, and conscientious human labor is largely better than whatever a LLM can give you. - 99% of the world has no access to the above, and never has.
I had strong priors against LLMs for medicine. There are a lot of doctors in my family and I grew up viewing doctors as careful, skilled professionals. I had plenty of bad medical experiences, but I thought it would be hard to do better. Then an LLM found a cure for my 2 decade chronic condition...
Oh, bishonen Hamilton, bishonen hamilton, America waits for you
whoah this Apple tech is cool they run a 20B model in-memory, but that’s far too big to actually fit in memory, so they use a tiny classifier to select which experts to load, once per inference instead of per output token machinelearning.apple.com/research/int...
Introducing the tanuki ball sack bag you never knew you needed until today ("his balls get bigger as you shop"). You're welcome.
Another day, another bunch of entitled people yelling at free software maintainers. The older I get the more impressed I am about how nothing ever changes. apenwarr.ca/log/20211229
The Gift of It's Your Problem Now
Recently a security hole in a certain open source Java library resulted in a worldwide emergency kerfuffle as, say, 40% of the possibly hund...
apenwarr.ca
As much as I understand the impulse, *any* feature that prevents a context window from being fully clean is bad and needs to be fired directly into the sun. The ability to *fully* reset to a clean context window is fundamental to almost every LLM product.
New -- I wrote about ChatGPT's April 2025 cross-chat memory update, which many people whose lives were upended by ChatGPT obsessions point to as an inflection point. It “takes people into cul-de-sacs... [my wife] is like a Waymo driving around in a circle endlessly.” futurism.com/artificial-i...
Vermeer’s Girl With The Pearl Earring will be shown in Osaka, later this year, and the press conference was as Japanese as you’d expect #Vermeer
A really dumb thing here is that even if we accept this framing, the building part is what’s been heavily automated and everyone becomes a “measurer”.
Cloudflare’s CEO explains why despite over 30% revenue growth, they are laying off 20% of staff. He argues that there are 3 types of workers; builders, sellers and measurers (middle managers, legal, HR, operations, etc). AI makes it possible to automate the last group and he expects everyone will
Network procedurals in general are A+ TV. Elementary is a particularly good example but man the whole genre doesn’t get its due. Full length seasons of charming ensemble casts! Retvrn!
God, Elementary is so much better than Sherlock.
100% this. LLMs are an augmentation, and generally not automation (especially in high-skill uses). Knowing your shit still matters.
what i actually think, having seen at least some of the front lines of this, is that ai is going to be bad for the lowest tier but actually quite good for the middle or higher tier worker. ai *increases* the skill premium. ceos hate this fact lmao
Infrastory is back with more posts this week - today’s is about how Toronto completely inverts expectations of what is possible for transit in car-oriented areas. Much to learn here for cities in North America and across the world! open.substack.com/pub/infrasto...
The Toronto Model for Transit
Induced demand is real for transit too
open.substack.com
Reinforces two of my strong suspicions: 1 - Small fine-tunes are *way* under-explored. 2 - Getting LLMs to speak JSON is increasing an architecture smell. In almost every situation you are getting degraded performance vs. asking it to speak XML, YAML, etc.
We parsed 3.6 million historical names with 96% accuracy by moving away from expensive frontier models to fine-tuned Qwen 3.5 models. We also found that switching our output format from JSON to YAML dramatically increased stability.
On the one hand the US can de facto have a flag carrier again. On the other hand it’s Spirit.
*US COULD OWN UP TO 90% OF SPIRIT AS PART OF $500 MILLION RESCUE
This is the greatest line read in history and I will never miss an opportunity to remind people of it.
this is kind of a crazy story! the headline predictably understates the role of the scientists the guy worked with as well as his own expertise, but the article is great and it does seem pretty clear that he would never have been able to do it without LLMs www.theaustralian.com.au/business/tec...
Sentence. Carriage return, carriage return. Sentence. Carriage return, carriage return. Sentence. Carriage return, carriage return. These people think their writing style is so badass and emphatic, when even their cadence is a cliche. I'm so sick and tired of this tired format.
did u know new sata ssds are mostly empty inside but just the right size for a custom enclosure
🐦 💘 It's the season for lovebirds! 💘 🦅 But ... which bird? From the iconic goose to the snowy owl, from monogamous to poly, there’s got to be one out there for you. Our Valentine's Day quiz will reveal the Canadian bird that best meets your relationship needs.
Valentine’s Day quiz: discover your Canadian lovebird | The Narwhal
Take our Valentine’s Day quiz to find the nest-mate of your dreams — and get a card to share
projects.thenarwhal.ca