Wildroot

Queries may use an external AI service. Details

Wildroot Browser

TodayStory on Bluesky

“Most modern transformer-based LLMs use the same broad…”

Posts
159
People
152
Time
–
Mostly
Politics

159 of the story’s 215 posts, best first: each person’s best post, then their other posts. We leave out exact copies and posts from people whose settings keep them off pages like this one. Page 1 of 4. The title is the name people mention most, or the opening line of the best post.

  1. @lea-linux.org

    "Most modern transformer-based LLMs use the same broad structure: tokenization, embeddings, positional encoding, stacked transformer layers (each with multi-head attention and a feed-forward network), residual streams, layer normalization, and next-token prediction."

    Most modern transformer-based LLMs use the same broad structure: tokenization, embeddings, positional encoding, stacked transformer layers (each with multi-head attention and a feed-forward network), residual streams, layer normalization, and next-token prediction. What changes between models is: The trained weights themselves, learned from different training data at different scales. The configuration: number of layers, vocabulary size, head count, parameter count, MoE or dense. The post-training: instruction tuning, learning from human feedback, safety controls applied on top of the base model. Tiny explainer: weights Weights are the learned numbers inside the model. Training changes those numbers until the model predicts text well. The 2023-2025 “modern transformer” stack converged on a common set of choices across many serious frontier and open-weight models, even though different teams arrived at them independently. Pre-norm placement. RMSNorm. RoPE. SwiGLU. Grouped-Query Atten
    InfoAI
  2. @jennburrill.bsky.social

    Top national security advisers accidentally texted war plans to The Atlantic’s editor in chief, Jeffrey Goldberg Their Signal messages were set to auto-delete @theatlantic.com reporting from June 2026

    Info
  3. @petisia.bsky.social

    John Grisham says that Scott Colom’s (D) defeat of Sen. Cindy Hyde-Smith (R) challenges a political calculation that has shaped Mississippi politics for decades: Black voters largely support Democrats, white voters overwhelmingly support Republicans, and statewide Democratic candidates lose. #PINKS

    Photo: A social media quote graphic featuring a main photograph of U.S. Senator-elect Scott Colom, a Black man in a blue blazer and white collared shirt, standing behind his young daughter in a purple long-sleeved top who is raising a small American flag. A larger American flag hangs out of focus in the background. In the upper right corner is a circular inset headshot of author John Grisham, an older white man with short graying hair wearing a dark blazer and light blue collared shirt. Centered at the bottom in bold white and light green text is the quote: “As a kid, I once watched the Ku Klux Klan burn a cross in my hometown in Mississippi. Sixty years later, I look forward to shaking Mr. Colom’s hand and telling him I’m proud he’s my senator.” Below the quote, smaller text reads, “- John Grisham, on Scott Colom.” A white square logo for “QUOTE CATALOG COM” sits in the top left corner, and vertical Getty Images photo credits run along the lower right edge. • The quote in the image is
    Info
  4. @dzaladan.eurosky.social

    I'm a naturally curious person, so I like to explore any kind of topic. This thread here gives a precise explanation about how LLMs work. In a nutshell, confirms what has been said already: it's controlled randomization of preexisting data. So... math, not chemistry.

  5. @premthakker.bsky.social

    Here’s what Pramila Jayapal said on the strategy to advance multiple stock trading ban bills, instead of one all-encompassing ban:

    InfoPopular today
  6. @swizzard.pizza

    i'm struggling to explain the breadth of loss ai has inflicted on programming as a trade like imagine if 6 months after the first cement-mixer truck debuted everyone just forgot about rebar

    OpinionAI
  7. @chtkghk.bsky.social

    Human mind: "From logic, reason, experience and observing reality. I know for a fact that 1+1=2." LLMs: "I've been given examples of what 'numbers' are said to be. I've been given examples of what 'addition' is said to be. Therefore, when asked, I will say 1+1 might equal 2 or any other number."

    OpinionAI
  8. I think a chunk of the problem with discussing Democrats is that we erroneously split them into "progressives versus moderates" when there's actually *three* distinct groups: Progressives: The solidly left Moderates: The center-left Blue Dogs & their Senate equivalents: The center/center-right

    Opinion
  9. @1x1.adder.industries

    The thing about AI nonsense that makes me the maddest is that they've ruined cute animal photos online. I'm paranoid about every pic I see these days. (This is not the thing I'm maddest about; that will always be the wage theft. Allow a girl some hyperbole.)

    OpinionAI
  10. @cryptidworks.bsky.social

    Every economist worth their salt has been saying for years GenAI was a dangerously overpressured investment bubble that would cause the next dot com level crash in tech when it popped Now TechBros are acting shocked it finally happened

    OpinionAI
  11. @bruces.bsky.social

    *It sounds baffling and buzzwordy to *other scientists* who are in closely-related fields and might be muscling-in on your territory *There's also rather a lot of trendy one-upmanship in science jargon, which exists to propose that I'm with-it and you no longer are

    OpinionScience
  12. I'm not a mathematician so this is definitely not my lane but for me, AI generating a proof for the Fiemann hypothesis is fundamentally uninteresting. Okay, so the monkeys banged out Hamlet - it was inevitable based on the laws of probability.

    OpinionAI
  13. @do60een.bsky.social

    BREAKING: Speaker-designate Hakeem Jeffries unveils Democrats’ bold new agenda as they prepare to take the majority: — End Citizens United — Universal healthcare — Ban gerrymandering — Congressional stock trading ban — Raise federal minimum wage — End the Iran War — Repeal the Trump tariffs.

  14. @philipncohen.com

    Incorrect first names in otherwise correct citations probably just means someone extracted citations using a style that doesn't spell out first names and then used a chatbot to convert them to a different style without realizing it would fabricate the names.

    OpinionAI
  15. @lydz.bsky.social

    one of the things that has me convinced the LLMs are going to be the entirety of coding people are sniffing farts is seeing shit like "these days it's got a pretty limited range of outcomes" like the chatbots always ensure dece results when every new website i've touched over the last year is trash

    OpinionAI
  16. @pnuttgallery.bsky.social

    Raskin is one of those Democrats that I actually believe hates these people on a visceral level and wants to see them behind bars for the rest of their lives for what they’ve done. I just feel it in his words whenever he speaks. Ossoff has this quality as well.

    Opinion
  17. @tshirt.bsky.social

    despite my policy against it. i'm tempted to grade some of these clearly AI essays students have turned it very highly. because it is kind of performance art that basically everyone who turned in one chose the topic (out of their own free will) of why human writing is better than AI writing

    OpinionAI
  18. @joshuafaile.bsky.social

    Some Democrats are considering passing a stock trading ban by public officials that will exempt the public officials doing most of the stock trading? This is like proposing a ban on agents of the federal government covering their faces that exempts ICE.

    Opinion
  19. @sean.thenewells.us

    I'm deeply uncomfortable even hinting at an equivalence between current LLMs, NN, RNNs et. al. to biological brains. (Even the fly) We modelled these software systems after a very juvenile and flawed and even then simplified version of what we understood our brain to be doin up there.

    OpinionAI
  20. Van Hollen says it's still a "wide-open race" for majority leader Sen. Chris Van Hollen (D-Md.), who is challenging Senate Minority Leader Chuck Schumer (D-N.Y.) to lead the body’s Democratic caucus in the next Congress... https://thehill.com/homenews/senate/6235600-van-hollen-schumer-senate-race/

    InfoPopular today
  21. There is no “centrist” or “moderate” stance between the Republican and Democratic positions on the rule of law. Either you believe the law should apply to the Executive branch or you don’t.

    Opinion
  22. @kevinmkruse.bsky.social

    Trump muddied the water on the "weaponization of law" bullshit that Democrats will have to keep stressing that holding criminals accountable for clear violations of federal law is not remotely the same thing as the shit he pulled.

    OpinionPopular today
  23. @jamesbicycle.bsky.social

    Okay, let's clarify. From a Canadian, Australian etc and European perspective, the USA has no political left, only centre-right, and more extreme right (currently very very). The excitement about some newly elected politicians there, and a mayor- these are in the political centre by world standards.

    Opinion
  24. @laurelann.bsky.social

    This is how out of touch and arrogant this man is: in the same month Wisconsin voters swept Democrats into power, Ron Johnson, who is up for reelection in 2028, called Social Security a "ponzi scheme".

    Opinion
  25. @carl-bond.com

    1/15 When we talk about stopping or slowing AI development, I want us to also discuss what people should be able to build for themselves and how they can create value in their own communities. #AI #CommunityDevelopment #PublicInfrastructure

  26. @504spyboy.bsky.social

    "This system did not randomly fail on its own. They broke it. The upgrade work is what triggered the outage." Pete Buttigieg Former Secretary of Transportation On MS Now, November 18, 2026 youtu.be/89nMJXUc2Ho 1/

  27. @jmw-512.bsky.social

    tech.yahoo.com/science/arti... And we'd want to ................why? These people see everything in terms of quantity That's why $100 billion can't possibly ever be enough. They don't understand Quality. The purchase it, but see their purchases only as icons of quantity.

    OpinionAI
  28. @jamesgoodwin.bsky.social

    But let's not get ahead of ourselves. We'll still have to frame everything Democrats do, no matter how incremental or objectively reasonable, as "overreach." Sorry, I don't make the rules.

    Opinion
  29. @thoreau.bsky.social

    People before the midterms kept saying "we're so polarized, a victory like 2018 is not going to be possible" and i called this out multiple times. Dems started with a higher floor, and so their ceiling was always higher. Link to my call-outs in thread.

    Opinion
  30. @ethanlandes.bsky.social

    If anyone is looking for a weekend coding project, it would be stupidly easy in the era of LLMs to automatically compile predictions of pundits then compare them later against news stories.

    OpinionAI
  31. Yesss Like I said, frame all this as enforcing the rule of law. Keep saying it, over & over. The more it's hammered home that we're just enforcing rule of law (which is true even in many of the maximal desires!), the more it defangs a tit for tat, revenge, persecuting political opponents narrative

    Opinion
  32. @indivisible.org

    To win again in two years, we must make the differences between Democrats and Republicans as clear to voters as possible. We can’t do that without ending the filibuster.

    Popular today
  33. A philosopher is being quoted as an expert on human life expectancy? Not a biologist, or an expert on aging? Man, get the fuck of here. And when you get there, go outside and play hide and go fuck yourself.

    OpinionAI
  34. I'm honestly despairing a little bit at the fact that prominent people who could be making arguments for if and how we should be harnessing the amazing powers of LLMs safely and humanely are instead spending their time lying about how powerful LLMs actually are

    OpinionAI
  35. @ash243x.bsky.social

    it's funny how mad the south and western right get at new yorkers, especially as their ultraliberal neighbors and local sports rivals in the boston metro; ny is conservative from our perspective, and folks like mamdani are in a sense a return to traditional american values, but without the racism.

    Opinion
  36. @westernstyle.live

    i'm so tired of llm voice everywhere. it used to be so fun being able glean something about a person from the way they phrased stuff but not like 80% of any email or description is the exact same tone, the exact same patterns and the exact same pointless stylistic flourishes

    OpinionAI
  37. @operator1997.bsky.social

    Bolsonaro Sr. is in the ICU. He got a bowel obstruction in the small intestine.

    Info
  38. @mcbridetd.bsky.social

    Sen. Chris Van Hollen says ‘momentum is in our direction’ in battle for majority leader www.nbcnews.com/politics/con...

    Info
  39. @jekeep.bsky.social

    Saw the tech scammers pushing the idea that human lifespans will increase to over a thousand years, or to immortality, because of AI. Ludicrous, but exactly the kind of thing that will appeal to our vain, selfish, aging ruling class, to ensure they back AI no matter what.

    OpinionAI
  40. @tdverstynen.bsky.social

    Pro tip: Anything that follows “Philosopher Nick Bostrom estimates…” is very likely to simply be attention-grabbing clickbait with little connection to observable reality.

  41. @headdoc35.bsky.social

    Hopefully, with Democrats at the helm, the funds will be distributed. Other than saying that, I want to say Republicans can go suck a tailpipe.

    Opinion
  42. Imagine writing a law to prevent bad people from doing a bad thing but excepting bad people who are doing said bad thing. Really leans into the trope of the law applying equally to all.

    Opinion