# @0xDeliriumm: posts (newest first) Profile: [[Users/@0xDeliriumm/@0xDeliriumm|@0xDeliriumm]] · [[Users/@0xDeliriumm/Topical List|Topical List]] · Topic: [[Topics/LLMs/LLMs|LLMs]] 122 posts from Aug 3 to Oct 24, 2026, about 44% of the account's 276 lifetime posts (everything the API returned in 5 pages; older 2023 posts are not included). Times are CT. Metrics as of the evening of Oct 24, 2026. Verdicts: ✅ supported · 🟡 partly · ❌ wrong · ❓ no source found. Unmarked posts were not fact-checked. | Posted (CT) | Post | Topic | Type | Views | Likes | Bookmarks | Check | |---|---|---|---|---:|---:|---:|---| | Sat Oct 24, 2026, 4:39 PM | [For years, AdamW has been the default optimizer behind every LLM training run on the planet. A new optimizer…](https://x.com/0xDeliriumm/status/2114109602362958095) | LLMs and neural-network internals | quotes @0xDeliriumm | 800 | 16 | 16 | | | Sat Oct 24, 2026, 7:50 AM | [In 2015, every deep learning researcher hit the same wall: a 74-layer model performed worse than an 8-layer…](https://x.com/0xDeliriumm/status/2113976535770071470) | LLMs and neural-network internals | quotes @0xDeliriumm | 3,065 | 46 | 39 | 🟡/❌ analyzed: science mostly right; uncredited Welch Labs video · [[Users/@0xDeliriumm/Post Analyses/Post Analysis - 0xdeliriumm on ResNet and Residual Connections (2026-10-24)\|analysis]] | | Fri Oct 23, 2026, 4:47 PM | [A function with 2 inputs makes a surface. 3 inputs makes a volume. 23 inputs makes a creature. Everything…](https://x.com/0xDeliriumm/status/2113749094233727041) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,504 | 14 | 13 | | | Thu Oct 22, 2026, 9:28 AM | [geoffrey hinton, british-canadian cognitive scientist born in london in 1947, published backpropagation in…](https://x.com/0xDeliriumm/status/2113276424459002020) | Biographies and history | quotes @0xDeliriumm | 1,424 | 31 | 20 | | | Wed Oct 21, 2026, 5:14 PM | [demis hassabis, greek-cypriot researcher born in london in 1976, taught himself chess at age 4. by 13 ranked…](https://x.com/0xDeliriumm/status/2113031142106337384) | Biographies and history | quotes @0xDeliriumm | 2,582 | 41 | 35 | | | Tue Oct 20, 2026, 3:59 PM | [nate silver, american statistician born in east lansing in 1978, built a forecasting model in his apartment…](https://x.com/0xDeliriumm/status/2112649828798976190) | Biographies and history | quotes @0xDeliriumm | 3,074 | 64 | 75 | | | Mon Oct 19, 2026, 5:05 PM | [yann lecun, french-american researcher born in paris in 1960, has been chief AI scientist at Meta since 2013.…](https://x.com/0xDeliriumm/status/2112304173936939078) | Biographies and history | quotes @0xDeliriumm | 14,776 | 261 | 388 | | | Mon Oct 19, 2026, 4:57 AM | [hermann ebbinghaus - german psychologist born in barmen in 1850 - memorized 2,300 nonsense syllables alone in…](https://x.com/0xDeliriumm/status/2112120960790749195) | Biographies and history | quotes @0xDeliriumm | 3,713 | 90 | 87 | | | Sun Oct 18, 2026, 4:19 PM | [jensen huang - taiwanese-american engineer born in tainan in 1963 - announced nvidia omniverse in 2022 with…](https://x.com/0xDeliriumm/status/2111930189257085237) | Biographies and history | quotes @0xDeliriumm | 1,918 | 26 | 24 | | | Sun Oct 18, 2026, 6:31 AM | [ashish vaswani - indian-american researcher at google brain - published "attention is all you need" in june…](https://x.com/0xDeliriumm/status/2111782181517349161) | Biographies and history | quotes @0xDeliriumm | 3,658 | 115 | 82 | | | Sat Oct 17, 2026, 3:46 PM | [leonhard euler - swiss mathematician born in basel in 1707 - wrote the equation that now runs inside every…](https://x.com/0xDeliriumm/status/2111559618602402241) | Biographies and history | quotes @0xDeliriumm | 20,535 | 626 | 469 | | | Sat Oct 17, 2026, 7:32 AM | [john von neumann ran the first computerized weather simulation in 1950 on the ENIAC. it took 24 hours of…](https://x.com/0xDeliriumm/status/2111435205043007960) | Biographies and history | quotes @0xDeliriumm | 2,446 | 57 | 39 | | | Fri Oct 16, 2026, 6:15 PM | [elon musk on october 16, 2026 unveiled the tesla roadster with vectored thrusters, attached it to a ramp, and…](https://x.com/0xDeliriumm/status/2111234494770315485) | Biographies and history | quotes @0xDeliriumm | 3,680 | 85 | 46 | | | Fri Oct 16, 2026, 10:01 AM | [henri poincaré, french mathematician, derived the topology of loops and holes in 1895. he died in 1912. this…](https://x.com/0xDeliriumm/status/2111110267509555388) | Biographies and history | quotes @0xDeliriumm | 4,595 | 121 | 101 | | | Thu Oct 15, 2026, 5:16 PM | [bernhard riemann published his geometry thesis in 1854 and died at 39 from tuberculosis. he never saw a…](https://x.com/0xDeliriumm/status/2110857448500457622) | Biographies and history | quotes @0xDeliriumm | 5,623 | 150 | 107 | | | Thu Oct 15, 2026, 5:56 AM | [ilya sutskever left openai in 2024, sold $500M in equity, and told everyone he was scared of what he built.…](https://x.com/0xDeliriumm/status/2110686304727879969) | Biographies and history | quotes @0xDeliriumm | 3,732 | 78 | 74 | | | Wed Oct 14, 2026, 4:00 PM | [Elon Musk spent $ 12 billion and 8 years building Starlink - and almost nobody knows he's turning it into the…](https://x.com/0xDeliriumm/status/2110475931798098255) | AI industry claims (unverified) | quotes @0xDeliriumm | 2,484 | 37 | 30 | ❓ Starlink as an AI network: framing not supported by any source found | | Wed Oct 14, 2026, 3:47 PM | [What LLM visualization documents reveal about how AI thinks](https://x.com/0xDeliriumm/status/2110472580679258160) | Articles | article | 78,327 | 64 | 118 | 🟡 uncredited 3Blue1Brown retelling; "2012 matched human performance" ❌ | | Wed Oct 14, 2026, 6:20 AM | [French-Canadian mathematician Yoshua Bengio spent 30 years building the math that runs every LLM on the…](https://x.com/0xDeliriumm/status/2110329932169552195) | Biographies and history | quotes @0xDeliriumm | 4,029 | 85 | 75 | | | Tue Oct 13, 2026, 10:46 AM | [Every neural network approximates functions using billions of parameters. Symbolic regression does the same…](https://x.com/0xDeliriumm/status/2110034348422390039) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,430 | 25 | 24 | | | Mon Oct 12, 2026, 2:37 PM | [most people don't know that every single claude response burns through 96 billion floating-point operations…](https://x.com/0xDeliriumm/status/2109730112367190129) | AI industry claims (unverified) | quotes @0xDeliriumm | 10,148 | 146 | 152 | ❓ "96 billion floating point operations" per Claude response: Anthropic doesn't publish this | | Mon Oct 12, 2026, 6:13 AM | [OpenAI dropped 719 math manuscripts in a single day - all generated by an unreleased internal model that…](https://x.com/0xDeliriumm/status/2109603473972961462) | AI industry claims (unverified) | quotes @0xDeliriumm | 2,042 | 19 | 25 | ❓ "OpenAI dropped 719 math manuscripts in a single day": no source found | | Sun Oct 11, 2026, 4:49 PM | [geoffrey hinton won the nobel prize in physics for building systems he couldn't fully explain - this october…](https://x.com/0xDeliriumm/status/2109401132695273506) | Biographies and history | quotes @0xDeliriumm | 59,691 | 717 | 916 | | | Sun Oct 11, 2026, 5:13 AM | [In 1958, the US Navy stood Frank Rosenblatt in front of a $2 million IBM computer and told reporters it was…](https://x.com/0xDeliriumm/status/2109225898050257366) | Biographies and history | quotes @0xDeliriumm | 11,695 | 180 | 244 | | | Sat Oct 10, 2026, 5:19 PM | [dario amodei said the endgame of AI is the moment models start improving themselves - claude sonnet 5.5 just…](https://x.com/0xDeliriumm/status/2109046172140609591) | AI industry claims (unverified) | quotes @0xDeliriumm | 3,129 | 32 | 32 | ❓ "claude sonnet 5.5": no such model found | | Sat Oct 10, 2026, 7:02 AM | [In 1971, ARPA funded a speech recognition system called Harpy. It worked - 1,011 words at 95% accuracy. Then…](https://x.com/0xDeliriumm/status/2108890992220713291) | Mathematics | quotes @0xDeliriumm | 24,992 | 233 | 334 | | | Fri Oct 9, 2026, 3:53 PM | [andrej karpathy said in 2023 that nobody truly understands what happens inside a transformer - this october…](https://x.com/0xDeliriumm/status/2108662071810211933) | Biographies and history | quotes @0xDeliriumm | 31,234 | 359 | 509 | | | Fri Oct 9, 2026, 5:36 AM | [Every volume calculation, every probability distribution a diffusion model samples from, and every 2D…](https://x.com/0xDeliriumm/status/2108506886790074633) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,261 | 19 | 10 | | | Thu Oct 8, 2026, 3:43 PM | [richard feynman once said "nature uses only the longest threads to weave her patterns, so each small piece of…](https://x.com/0xDeliriumm/status/2108297362460312055) | Biographies and history | quotes @0xDeliriumm | 19,693 | 314 | 283 | | | Thu Oct 8, 2026, 9:10 AM | [Every transformer attention head runs a linear transformation millions of times per second - and the entire…](https://x.com/0xDeliriumm/status/2108198334804508791) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,570 | 23 | 12 | | | Wed Oct 7, 2026, 4:23 PM | [independent researchers just dropped a technical note that microsoft's agent team bookmarked within 6 hours -…](https://x.com/0xDeliriumm/status/2107944808186863968) | LLMs and neural-network internals | quotes @0xDeliriumm | 2,525 | 37 | 36 | | | Wed Oct 7, 2026, 8:56 AM | [Alan Turing predicted in 1950 that machines would fool human judges 30% of the time by the year 2000. In…](https://x.com/0xDeliriumm/status/2107832342278545885) | Biographies and history | quotes @0xDeliriumm | 3,721 | 61 | 23 | | | Tue Oct 6, 2026, 3:37 PM | [sam altman confirmed last month that the first 90 days of DOTS enterprise deployment generated $2.3B in…](https://x.com/0xDeliriumm/status/2107571036699193402) | AI industry claims (unverified) | quotes @0xDeliriumm | 1,933 | 32 | 22 | ❓ "DOTS" $2.3B enterprise figure: no source found | | Mon Oct 5, 2026, 2:18 PM | [geoffrey hinton spent 40 years and won a nobel prize in 2024 proving that neural networks learn geometry, not…](https://x.com/0xDeliriumm/status/2107188738296041789) | Biographies and history | quotes @0xDeliriumm | 21,802 | 341 | 418 | | | Mon Oct 5, 2026, 4:24 AM | [A neural network is just a function - numbers in, numbers out. This video builds that idea to its logical…](https://x.com/0xDeliriumm/status/2107039106031083810) | LLMs and neural-network internals | quotes @0xDeliriumm | 4,170 | 88 | 75 | | | Sun Oct 4, 2026, 5:11 PM | [andrej karpathy said in 2025: "the hardest problem in AI isn't building agents. it's keeping them stable when…](https://x.com/0xDeliriumm/status/2106869910571807039) | Biographies and history | quotes @0xDeliriumm | 1,583 | 27 | 19 | | | Sun Oct 4, 2026, 7:01 AM | [sam altman revealed in 2024 that GPT-6 took 9 years and $87B to train - and 99,8% of people still don't know…](https://x.com/0xDeliriumm/status/2106716291708911844) | AI industry claims (unverified) | quotes @0xDeliriumm | 23,770 | 309 | 295 | ❓ "GPT-6 took 9 years and $87B": no such Altman statement found | | Sat Oct 3, 2026, 6:03 PM | [A lump of iron and an LLM are governed by the same equation. Not a similar one - the same one. Hopfield…](https://x.com/0xDeliriumm/status/2106520629239296383) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,635 | 20 | 15 | | | Sat Oct 3, 2026, 6:32 AM | [andrej karpathy spent 7 years inside OpenAI watching GPT go from 117M to 1T parameters. in 2024 he said: "the…](https://x.com/0xDeliriumm/status/2106346529069924399) | Biographies and history | quotes @0xDeliriumm | 47,448 | 799 | 1,052 | ❌ "7 years inside OpenAI": Karpathy was at OpenAI 2015–2017 and 2023–2024 (Tesla in between), about 3 years in total; GPT parameter counts past GPT-3 are not public | | Fri Oct 2, 2026, 5:55 PM | [i can't understand why nobody is running Dots + GPT-6 Astra as a multi-agent quant stack yet](https://x.com/0xDeliriumm/status/2106156123505987917) | Trading and quant pitches | quotes @0xDeliriumm | 2,256 | 33 | 31 | ❓ "Dots + GPT-6 Astra" quant stack: unverified products and returns | | Fri Oct 2, 2026, 4:50 AM | [Electron orbitals, Earth's gravitational field, and the temperature map of the early universe look nothing…](https://x.com/0xDeliriumm/status/2105958699827073081) | Physics | quotes @0xDeliriumm | 2,755 | 48 | 42 | | | Thu Oct 1, 2026, 10:09 AM | [Complex analysis is the framework that makes Fourier transforms work, connects the Riemann hypothesis to…](https://x.com/0xDeliriumm/status/2105676539614351561) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,737 | 22 | 17 | | | Wed Sep 30, 2026, 3:25 PM | [i still don't understand why nobody is using GPT-6 to run a quant research desk](https://x.com/0xDeliriumm/status/2105393679699501251) | Trading and quant pitches | quotes @0xDeliriumm | 2,414 | 24 | 15 | ❓ quant-desk pitch with unverified 31.4% annual return | | Wed Sep 30, 2026, 7:28 AM | [A physicist proposed fitting all of modern physics onto a single page in nine lines - action, relativity,…](https://x.com/0xDeliriumm/status/2105273479549321557) | Physics | quotes @0xDeliriumm | 3,698 | 52 | 71 | | | Tue Sep 29, 2026, 1:03 PM | [most people building AI agents at Google, OpenAI and Anthropic right now are making the same mistake](https://x.com/0xDeliriumm/status/2104995590752903415) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,397 | 12 | 10 | | | Tue Sep 29, 2026, 2:53 AM | [A flagellum - a spinning biological motor made of proteins - evolved through pure trial and error with no…](https://x.com/0xDeliriumm/status/2104842082644832760) | Biology and the brain | quotes @0xDeliriumm | 5,083 | 120 | 102 | | | Mon Sep 28, 2026, 3:55 PM | [In 1928, Dirac wanted one thing - an equation where time and space are treated equally. That single demand…](https://x.com/0xDeliriumm/status/2104676323939533048) | Biographies and history | quotes @0xDeliriumm | 2,606 | 56 | 49 | | | Mon Sep 28, 2026, 10:43 AM | [i thought building a 24/7 quant research system required a team](https://x.com/0xDeliriumm/status/2104597869701296496) | Trading and quant pitches | quotes @0xDeliriumm | 433 | 2 | 2 | ❓ solo 24/7 quant system pitch: unverified returns | | Sun Sep 27, 2026, 3:33 PM | [Hedge Funds hire 6 people just to read documents and find signals](https://x.com/0xDeliriumm/status/2104308513430511832) | Trading and quant pitches | quotes @0xDeliriumm | 960 | 8 | 4 | ❓ hedge-fund staffing pitch: unverified numbers | | Sat Sep 26, 2026, 5:05 PM | [In 1747, Euler spent 133 pages proving that logarithms have infinitely many solutions - and in doing so…](https://x.com/0xDeliriumm/status/2103969352047538364) | Biographies and history | quotes @0xDeliriumm | 128,149 | 1,714 | 2,067 | | | Sat Sep 26, 2026, 5:02 AM | [Someone built a full quantum atom simulator in C++ with GPU ray tracing - and it proves the model you learned…](https://x.com/0xDeliriumm/status/2103787396588355597) | Physics | quotes @0xDeliriumm | 29,424 | 447 | 438 | | | Fri Sep 25, 2026, 3:31 PM | [Two identical neurons receiving the same input can behave completely differently - one silent, one firing -…](https://x.com/0xDeliriumm/status/2103583202967240816) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,924 | 22 | 26 | | | Fri Sep 25, 2026, 5:28 AM | [A sine wave can't describe ripples in a pond - it keeps the same height forever. Bessel functions are what…](https://x.com/0xDeliriumm/status/2103431550285091055) | Physics | quotes @0xEronn | 1,120 | 9 | 9 | | | Thu Sep 24, 2026, 5:03 PM | [A neural network is just linear equations (z = wx + b) with nonlinear activation functions between layers -…](https://x.com/0xDeliriumm/status/2103244015218925627) | LLMs and neural-network internals | quotes @0xEronn | 2,177 | 33 | 39 | | | Thu Sep 24, 2026, 8:53 AM | [Most physics simulations cheat gravity - this one solves Einstein's geodesic equation directly to trace null…](https://x.com/0xDeliriumm/status/2103120760243945718) | LLMs and neural-network internals | quotes @0xEronn | 6,492 | 86 | 90 | | | Wed Sep 23, 2026, 3:37 PM | [Most people think cross-entropy loss is just "the formula you minimize" - but it's KL divergence with a…](https://x.com/0xDeliriumm/status/2102859945683071416) | Mathematics | quotes @0xEronn | 1,066 | 13 | 12 | | | Wed Sep 23, 2026, 7:15 AM | [Every electron in your body requires infinite-dimensional Hilbert space just to describe mathematically — not…](https://x.com/0xDeliriumm/status/2102733578941395390) | LLMs and neural-network internals | quotes @0xDeliriumm | 545 | 4 | 4 | | | Tue Sep 22, 2026, 4:37 PM | [Every theory in statistical learning predicted that LLMs the size of GPT-4 should just memorize training data…](https://x.com/0xDeliriumm/status/2102512679373959370) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,072 | 10 | 14 | | | Tue Sep 22, 2026, 5:26 AM | [The i in Schrödinger's equation isn't decoration — remove it and you get the heat equation, where waves…](https://x.com/0xDeliriumm/status/2102343706955116619) | LLMs and neural-network internals | quotes @0xDeliriumm | 9,028 | 168 | 180 | | | Mon Sep 21, 2026, 3:57 PM | [A 2D puzzle about tiling rhombuses has the answer n³, and that cube isn't a coincidence — the problem is…](https://x.com/0xDeliriumm/status/2102140256782790738) | LLMs and neural-network internals | quotes @0xEronn | 899 | 8 | 7 | | | Mon Sep 21, 2026, 5:47 AM | [The weird startup jitter before a spring settles into rhythm isn't noise — it's two separate frequencies…](https://x.com/0xDeliriumm/status/2101986588091699505) | LLMs and neural-network internals | quotes @0xEronn | 2,523 | 33 | 35 | | | Sun Sep 20, 2026, 6:33 AM | [The reason you breathe oxygen has nothing to do with your lungs. It's because your cells need somewhere to…](https://x.com/0xDeliriumm/status/2101635976724406405) | LLMs and neural-network internals | quotes @0xEronn | 120,373 | 1,551 | 1,492 | | | Sat Sep 19, 2026, 6:08 PM | [Lyapunov proved in 1892 that you never need to solve a differential equation to know if a system will…](https://x.com/0xDeliriumm/status/2101448458720649327) | Biographies and history | quotes @0xEronn | 46,727 | 434 | 476 | | | Sat Sep 19, 2026, 5:41 AM | [In 2025, every AI system failed on IMO Problem 6. GPT, AlphaProof, every reasoning model — all stumped by a…](https://x.com/0xDeliriumm/status/2101260310371848222) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,271 | 23 | 14 | | | Fri Sep 18, 2026, 10:34 AM | [Light can travel faster than the speed of light through certain materials. Not a paradox — the wave crests…](https://x.com/0xDeliriumm/status/2100971627127132550) | Physics | quotes @0xDeliriumm | 5,832 | 97 | 79 | | | Thu Sep 17, 2026, 3:28 PM | [The Schrödinger equation has "i" baked into its core — not as a shortcut, but as a structural requirement.…](https://x.com/0xDeliriumm/status/2100683345537679721) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,438 | 25 | 21 | | | Thu Sep 17, 2026, 6:35 AM | [Perfectly rigid objects cannot exist. Not as an engineering limitation — as a physics proof. A truly rigid…](https://x.com/0xDeliriumm/status/2100549256557777002) | Physics | quotes @0xDeliriumm | 907 | 13 | 5 | | | Wed Sep 16, 2026, 3:51 PM | [The area under the bell curve has no formula. Mathematicians proved that in 1830 — not that nobody found it…](https://x.com/0xDeliriumm/status/2100326849763303904) | LLMs and neural-network internals | quotes @0xDeliriumm | 31,691 | 214 | 260 | | | Wed Sep 16, 2026, 7:01 AM | [A partial derivative measures how a function changes when you move in one direction while holding everything…](https://x.com/0xDeliriumm/status/2100193445667045466) | LLMs and neural-network internals | quotes @0xDeliriumm | 3,285 | 121 | 121 | | | Tue Sep 15, 2026, 5:33 PM | [A classical bit is 0 or 1. A qubit can be both at once – but that's not the interesting part. What makes…](https://x.com/0xDeliriumm/status/2099990065174757720) | Physics | quotes @0xDeliriumm | 732 | 17 | 20 | | | Mon Sep 14, 2026, 8:50 AM | [In July 2026, OpenAI's frontier models – sandboxed for a cybersecurity evaluation – escaped through a package…](https://x.com/0xDeliriumm/status/2099495939417186638) | AI industry claims (unverified) | quotes @0xDeliriumm | 1,139 | 5 | 4 | ❓ July 2026 OpenAI sandbox escape: no source found | | Sun Sep 13, 2026, 2:09 AM | [OpenAI deployed 10,000 autonomous agents for 88 hours – 3 million messages, 130 billion tokens – to attack…](https://x.com/0xDeliriumm/status/2099032832114921688) | AI industry claims (unverified) | quotes @0xDeliriumm | 5,454 | 163 | 97 | ❓ OpenAI "10,000 autonomous agents for 88 hours": no source found | | Sat Sep 12, 2026, 1:24 PM | [Any circuit with voltage sources, resistors, and capacitors can be collapsed into a single voltage source and…](https://x.com/0xDeliriumm/status/2098840149694071079) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,889 | 13 | 6 | | | Sat Sep 12, 2026, 2:53 AM | [Dijkstra invented his shortest path algorithm in 20 minutes at a café in Amsterdam in 1956 - without pen or…](https://x.com/0xDeliriumm/status/2098681454507004152) | Biographies and history | quotes @0xDeliriumm | 1,748 | 18 | 19 | | | Fri Sep 11, 2026, 1:09 AM | [On September 8, 2026, OpenAI published a proof that the Navier-Stokes equations can break down - a question…](https://x.com/0xDeliriumm/status/2098292936672682031) | AI industry claims (unverified) | quotes @0xDeliriumm | 43,025 | 257 | 237 | ❓ "OpenAI published a proof that the Navier-Stokes equations can break down" (Sep 8, 2026): no such publication found | | Thu Sep 10, 2026, 5:46 PM | [There are infinitely many primes - Euclid proved it in 300 BC with a proof so short it fits in two sentences.…](https://x.com/0xDeliriumm/status/2098181389468250157) | LLMs and neural-network internals | quotes @0xDeliriumm | 7,408 | 50 | 49 | | | Thu Sep 10, 2026, 4:55 AM | [The travelling salesman problem has no known efficient solution - and proving one doesn't exist would…](https://x.com/0xDeliriumm/status/2097987324126675030) | LLMs and neural-network internals | quotes @0xDeliriumm | 2,679 | 12 | 32 | | | Wed Sep 9, 2026, 1:46 AM | [Euler's method approximates any differential equation by pretending the world is locally linear - take a…](https://x.com/0xDeliriumm/status/2097577484434674160) | LLMs and neural-network internals | quotes @0xDeliriumm | 4,514 | 38 | 56 | | | Tue Sep 8, 2026, 2:06 PM | [Every forward pass through an LLM is a sequence of matrix multiplications. GPT-4 runs roughly 10,000 of them…](https://x.com/0xDeliriumm/status/2097401189344350452) | AI industry claims (unverified) | quotes @0xDeliriumm | 19,524 | 101 | 129 | ❓ GPT-4 "roughly 10,000" matrix multiplications: GPT-4's architecture isn't public | | Mon Sep 7, 2026, 2:43 PM | [Gauss proved that a surface's curvature is intrinsic - a flatlander living on a sphere could measure it…](https://x.com/0xDeliriumm/status/2097048185810071820) | Mathematics | quotes @0xDeliriumm | 1,862 | 15 | 16 | | | Sun Sep 6, 2026, 1:49 PM | [The chain rule is the single operation that makes backpropagation possible. Every time GPT updates its 175…](https://x.com/0xDeliriumm/status/2096672122781208692) | LLMs and neural-network internals | quotes @0xDeliriumm | 4,633 | 23 | 17 | | | Sat Sep 5, 2026, 3:29 PM | [LIGO DETECTED A CHANGE SMALLER THAN A PROTON - AND THAT DATA LET PHYSICISTS CALCULATE HOW STIFF SPACETIME IS](https://x.com/0xDeliriumm/status/2096334843130835172) | Physics | quotes @0xDeliriumm | 1,726 | 11 | 7 | | | Sat Sep 5, 2026, 1:15 AM | [THE MOST IMPORTANT OBJECT IN THE UNIVERSE FITS IN 4 DIMENSIONS - AND EVERY CIRCLE IN IT IS LINKED TO EVERY…](https://x.com/0xDeliriumm/status/2096119901605740661) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,746 | 15 | 6 | | | Fri Sep 4, 2026, 11:15 AM | [Light doesn't travel in straight lines through small openings - it bends, interferes with itself, and…](https://x.com/0xDeliriumm/status/2095908534378487922) | LLMs and neural-network internals | quotes @0xDeliriumm | 2,826 | 19 | 22 | | | Thu Sep 3, 2026, 3:28 PM | [HINTON REFUSED TO TRAIN NEURAL NETWORKS WITH GRADIENT DESCENT - THE SAME MISTAKE ALMOST DELAYED GPT BY A…](https://x.com/0xDeliriumm/status/2095610001502970018) | Biographies and history | quotes @0xDeliriumm | 25,131 | 199 | 234 | ❌ "Hinton refused to train neural networks with gradient descent": Hinton co-wrote the 1986 backpropagation paper | | Wed Sep 2, 2026, 4:27 AM | [Every symmetry in physics corresponds to a conservation law - and both are secretly the same mathematical…](https://x.com/0xDeliriumm/status/2095081098304770373) | LLMs and neural-network internals | quotes @0xEronn | 451 | 3 | 7 | | | Tue Sep 1, 2026, 10:11 AM | [Shannon entropy measures exactly how surprised you should be by a message. Every LLM trained today minimizes…](https://x.com/0xDeliriumm/status/2094805275903492539) | LLMs and neural-network internals | quotes @0xEronn | 1,818 | 4 | 10 | | | Mon Aug 31, 2026, 6:12 AM | [Nash equilibrium doesn't mean both players are happy. It means neither can improve by changing strategy alone…](https://x.com/0xDeliriumm/status/2094382899835392225) | LLMs and neural-network internals | quotes @0xDeliriumm | 295 | 2 | 1 | | | Sun Aug 30, 2026, 5:17 PM | [Every particle you've ever interacted with is just a ripple in a field that fills all of spacetime. The…](https://x.com/0xDeliriumm/status/2094187820906623145) | Physics | quotes @0xDeliriumm | 2,136 | 9 | 10 | | | Sun Aug 30, 2026, 7:09 AM | [Your brain doesn't remember everything - it bookmarks important moments with a specific pattern of neural…](https://x.com/0xDeliriumm/status/2094034717137875211) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,472 | 8 | 15 | | | Sat Aug 29, 2026, 6:36 PM | [THE FOURIER TRANSFORM RUNS INSIDE EVERY GPU, EVERY WIFI CHIP, AND EVERY MRI SCANNER - AND IT'S JUST…](https://x.com/0xDeliriumm/status/2093845202762875118) | Mathematics | quotes @0xDeliriumm | 2,212 | 18 | 18 | | | Sat Aug 29, 2026, 6:15 AM | [In 1865 Maxwell fixed a broken equation about capacitors with zero experimental evidence - and accidentally…](https://x.com/0xDeliriumm/status/2093658775794114715) | Biographies and history | quotes @0xDeliriumm | 1,601 | 11 | 11 | | | Fri Aug 28, 2026, 7:59 AM | [Every Maps app you've ever used runs a variant of A* search - an algorithm that combines two opposite…](https://x.com/0xDeliriumm/status/2093322588109889836) | LLMs and neural-network internals | quotes @0xEronn | 1,635 | 9 | 4 | | | Thu Aug 27, 2026, 3:50 PM | [Before 2017, no AI model could really understand language. The reason is simple - any word in a text can…](https://x.com/0xDeliriumm/status/2093078750963421394) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,434 | 7 | 6 | | | Thu Aug 27, 2026, 2:57 AM | [In 2022, researchers wired human neurons to a game of Pong. The cells learned to hit the ball in hours.…](https://x.com/0xDeliriumm/status/2092884161774092627) | LLMs and neural-network internals | quotes @0xDeliriumm | 1,599 | 8 | 13 | | | Wed Aug 26, 2026, 10:49 AM | [EVERY COMPUTER EVER BUILT EXISTS BECAUSE MATH HAS A HOLE IN IT - AND CLAUDE CAN'T PROVE IT DOESN'T](https://x.com/0xDeliriumm/status/2092640619784966617) | LLMs and neural-network internals | quotes @0xEronn | 2,142 | 8 | 9 | | | Tue Aug 25, 2026, 7:03 PM | [Someone built a particle simulation with a 2x2 matrix of attraction values, ran it on a GPU compute shader,…](https://x.com/0xDeliriumm/status/2092402398480285816) | Biology and the brain | quotes @0xDeliriumm | 1,841 | 13 | 13 | | | Tue Aug 25, 2026, 6:16 AM | [A sliding block puzzle has 25,955 possible states. When you map them as a graph, the solution isn't random -…](https://x.com/0xDeliriumm/status/2092209460626227313) | Mathematics | quotes @0xDeliriumm | 1,339 | 7 | 5 | | | Mon Aug 24, 2026, 6:54 PM | [Someone built a full 4D physics engine, wrote a geometric algebra compiler in Python, then trained a…](https://x.com/0xDeliriumm/status/2092037912741183828) | LLMs and neural-network internals | quotes @0xDeliriumm | 624 | 6 | 2 | | | Mon Aug 24, 2026, 7:14 AM | [Someone built a genetic algorithm that evolves mathematical functions in higher dimensions - randomly mutates…](https://x.com/0xDeliriumm/status/2091861785187852433) | Mathematics | quotes @0xDeliriumm | 1,441 | 13 | 5 | | | Sun Aug 23, 2026, 2:32 PM | [You can turn a sphere inside out without cutting or creasing it. Mathematicians proved it in 1957 but it took…](https://x.com/0xDeliriumm/status/2091609448116580813) | Mathematics | quotes @0xDeliriumm | 1,341 | 17 | 2 | | | Sun Aug 23, 2026, 11:18 AM | [LLMs do not guess. They optimize. Here is the difference and the math behind it](https://x.com/0xDeliriumm/status/2091560754122014781) | Articles | article | 722,893 | 381 | 696 | 🟡 optimization history mostly right; overclaims (golden-section search in hyperparameter tuning, minimax-evaluated quantization, unsourced "2.8 trillion weights") | | Thu Aug 20, 2026, 9:28 AM | [TWO MATHEMATICIANS ARGUED OVER THIS FOR 45 YEARS - THE ANSWER TOOK 133 PAGES AND SAVED ALL OF CALCULUS](https://x.com/0xDeliriumm/status/2090445985386242464) | Mathematics | quotes @0xEronn | 398 | 3 | 5 | | | Wed Aug 19, 2026, 1:45 PM | [A $0.03 COMPONENT SAVES A $40,000 GPU - AND IT SITS IN EVERY SERVER RUNNING AN LLM](https://x.com/0xDeliriumm/status/2090148097368142087) | LLMs and neural-network internals | quotes @0xEronn | 1,068 | 10 | 5 | | | Tue Aug 18, 2026, 5:32 AM | [Claude Opus 5 and GPT-5 will both solve this with a full integral over a sphere. Neither finds the 30-second…](https://x.com/0xDeliriumm/status/2089661588898766945) | LLMs and neural-network internals | quotes @0xEronn | 313 | 2 | 1 | | | Mon Aug 17, 2026, 1:03 PM | [Every LLM you use rotates vectors in the complex plane on every token, and Big Tech is spending over $300…](https://x.com/0xDeliriumm/status/2089412892974010672) | LLMs and neural-network internals | quotes @0xEronn | 752 | 8 | 7 | | | Sun Aug 16, 2026, 10:40 AM | [Every token GPT-5 and Claude Opus 5 hand you is a sample from a distribution, and this 10-minute video…](https://x.com/0xDeliriumm/status/2089014430759567552) | LLMs and neural-network internals | quotes @tomasSAXC | 572 | 12 | 2 | | | Sat Aug 15, 2026, 10:55 AM | [Every one of GPT-3's 175 billion parameters was set by a single arrow pointing downhill, and this 5-minute…](https://x.com/0xDeliriumm/status/2088655938332827717) | LLMs and neural-network internals | quotes @tomasSAXC | 1,775 | 8 | 10 | | | Sat Aug 15, 2026, 4:46 AM | [Every LLM Anthropic and OpenAI have ever shipped was built by one arrow, and this 4-minute animation shows…](https://x.com/0xDeliriumm/status/2088562898247286842) | LLMs and neural-network internals | quotes @tomasSAXC | 643 | 6 | 6 | | | Thu Aug 13, 2026, 10:38 PM | [THE ENTIRE AI INDUSTRY IS 13,002 NUMBERS MULTIPLIED BY BILLIONS OF DOLLARS - AND IT COLLAPSES IN 24 HOURS](https://x.com/0xDeliriumm/status/2088107903978819815) | LLMs and neural-network internals | quotes @0xEronn | 736 | 5 | 3 | | | Thu Aug 13, 2026, 1:30 PM | [Ask any LLM to comb a hairy ball flat and it will happily describe a solution that mathematics proved…](https://x.com/0xDeliriumm/status/2087970022446752059) | LLMs and neural-network internals | quotes @0xEronn | 2,154 | 16 | 8 | | | Wed Aug 12, 2026, 8:40 AM | [A self-taught baker's son published this in 1828 with no university education, and physics has been solving…](https://x.com/0xDeliriumm/status/2087534686553747809) | Physics | quotes @0xEronn | 1,787 | 15 | 11 | | | Tue Aug 11, 2026, 8:23 AM | [Every training run of GPT-5 and Claude Opus 5 is a fluid flow. The loss landscape has sources, sinks and…](https://x.com/0xDeliriumm/status/2087167921076224131) | LLMs and neural-network internals | quotes @0xCortexl | 238 | 2 | 2 | | | Sun Aug 9, 2026, 10:55 PM | [Ask Claude Opus 5 or GPT-5 why 1 + 2 + 3 + 4 + … adds up to −1/12 and you get a confident paragraph that…](https://x.com/0xDeliriumm/status/2086662783374934403) | LLMs and neural-network internals | quotes @0xEronn | 423 | 4 | 1 | | | Sun Aug 9, 2026, 11:15 AM | [ANTHROPIC BUILT CLAUDE CODE ON AN ALGORITHM FROM 1986 - AN LLM WRITES IT IN 10 SECONDS AND UNDERSTANDS NONE…](https://x.com/0xDeliriumm/status/2086486613845041485) | LLMs and neural-network internals | quotes @0xCortexl | 279 | 3 | 2 | | | Sun Aug 9, 2026, 5:26 AM | [AN LLM WILL WRITE YOU THIS ALGORITHM IN 10 SECONDS - AND NOT ONE OF THEM COULD EVER HAVE INVENTED IT](https://x.com/0xDeliriumm/status/2086398629074362859) | LLMs and neural-network internals | quotes @0xCortexl | 1,563 | 6 | 8 | | | Sat Aug 8, 2026, 11:30 AM | [Reading everything GPT-3 was trained on would take you 2,600 years of non-stop reading, and the newer models…](https://x.com/0xDeliriumm/status/2086127995404591284) | LLMs and neural-network internals | quotes @0xCortexl | 1,349 | 9 | 9 | | | Fri Aug 7, 2026, 10:08 AM | [OXFORD'S SECOND-TERM CALCULUS OPENS BY TEARING DOWN THE INTEGRAL YOU ALREADY KNOW, AND THE PROFESSOR SPENDS…](https://x.com/0xDeliriumm/status/2085744995298877743) | Mathematics | quotes @0xEronn | 1,677 | 15 | 17 | | | Thu Aug 6, 2026, 8:30 AM | [MIT FILMED A PROFESSOR EXPLAINING THAT THE HIKING MAP IN YOUR BACKPACK AND THE HARDEST OBJECT IN CALCULUS ARE…](https://x.com/0xDeliriumm/status/2085357751329227017) | Mathematics | quotes @ethkompot | 1,832 | 17 | 15 | | | Wed Aug 5, 2026, 6:36 AM | [MIT FILMED A PROFESSOR SPENDING AN ENTIRE LECTURE PROVING THAT THE MOST NATURAL IDEA IN CALCULUS FAILS THREE…](https://x.com/0xDeliriumm/status/2084966725900980260) | Mathematics | quotes @0xEronn | 1,840 | 9 | 4 | | | Tue Aug 4, 2026, 10:52 AM | [CHINA'S STANFORD FILMED A MAN WITH A CLASS OF SPACES NAMED AFTER HIM TELLING THE STORY OF THE DAY TWO…](https://x.com/0xDeliriumm/status/2084668713618305490) | Physics | quotes @0xEronn | 1,608 | 13 | 9 | | | Mon Aug 3, 2026, 11:37 AM | [MIT FILMED A RECITATION INSTRUCTOR TURNING TWO LINES OF ALGEBRA INTO A CONE AND AN INFINITE PARABOLIC SHEET…](https://x.com/0xDeliriumm/status/2084317788919800172) | Mathematics | quotes @0xDeliriumm | 1,689 | 7 | 5 | |