@JWerner247 Illinois is proof that momentum can stay exactly the same
English
Eric Lee
75 posts
Purdue, which had lost four of six to end the regular season, is proof the momentum can turn in the postseason. Started with toughness and intensity on defense for the Boilermakers.
English
People want to rally *for* something
Sadly, we have created enemies amongst ourselves because we, as a society, have failed to show hard work has value and output to each individual
As a result, people rally *against* progress, for collectivism, etc
Ben Dziobek@BenDziobek
A data center in New Brunswick was canceled tonight when hundreds of residents showed up. When fight big tech and private equity we win.
English
Donovan Dent coast to coast with the game-winner.
UCLA beats Illinois in overtime.
English
UCLA has made 17 of its last 19 shots.
#illini finally cold from three, missng five straight from beyond the arc.
English
@JWerner247 I don't think they've contained any team off the bounce this season
English
It's a two-possession game for the first time since #illini led 13-10 at the 15:26 mark of the first half.
54-48 with 16:56 left.
UCLA has made 14 of its last 16 shots. Illini can't contain them off the bounce. Switching not working.
English
@YouJiacheng I use these as interview questions for new grad engineers where incorrect answer = instant rejection
English
AA-Omniscience's SWE section looks crazy.
what are you Measuring???
English
Kinda crazy lol
BURKOV@burkov
LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work. The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway. There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself. The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding. Read with AI tutor: chapterpal.com/s/1b15378b/pro… Get the PDF: arxiv.org/pdf/2512.14982
English
Prompt repetition is way overplayed
If you put the question first, the gains from repetition vanish or reduce dramatically
The biggest gains are on tasks where they didn't report question-first variants... I wonder why
BURKOV@burkov
LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work. The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway. There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself. The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding. Read with AI tutor: chapterpal.com/s/1b15378b/pro… Get the PDF: arxiv.org/pdf/2512.14982
English
That was great from Jeremy Fears Jr. Terrific in the game (26 points/15 assists). Then he handled the postgame interview well, all things considered.
English
@OfficialLoganK Funny, I spent the last week building something worth $200M... I'm sure yours is good too though
English
I spent the last week building what could easily be a $100M venture backed business… it’s truly wild how much leverage AI gives you.
English
Only 29 days after Pro: Gemini 3 Flash is just as smart (1477 LMSYS Elo), 4x cheaper, and sooo much faster! ⚡
English
Gemini 3 Flash is here. ⚡⚡⚡
Pro-grade reasoning with Flash-level speed and efficiency.
It’s rolling out today globally as the default model on Gemini App and Search AI Mode.
Learn more: bit.ly/4pTo5YU
English
No. If this came out in the pre-AI era, almost no one would notice it. People largely don’t care about visual art for its own sake. The reason this is getting so much attention has almost nothing to do with the work itself. x.com/chatgpt21/stat…
Chris@chatgpt21
Apparently this video has all of X in a frenzy. If it had come out before the AI era, people would be fawning over it as great art, but now they are so clicker trained that any mention of AI sends them into a verbiage frenzy and they anoint anything AI related as slop.
English
A formative work to my perspective on the AI art debate was a paper a few years ago where they tried to get people to guess whether poems were written by AI and they were worse than random, but when you dig into the data a bit it’s because average people hated the human poems
English
Kylan Boswell has 17 of Illinois' 40
They'd be getting hammered if not for him.
English
Using Google’s reported estimate of 0.24Wh per prompt, guesstimating ~500-1500 tokens/prompt puts them in the range of about 300-850 MW on AI
Demis Hassabis@demishassabis
We processed over 1.3 Quadrillion tokens last month - that's 1,300,000,000,000,000 tokens! or to put it another way that's 500M tokens a second or 1.8 Trillion tokens an hour... 🤯
English
You either die a sigmoid, or you live long enough to see yourself become the exponential.
English
While you're all still arguing about agi, I'm watching the joy on my kids' faces as they burn through my midjourney credits animating their own paintings.
English