Ekaterina Leksina
4 followers · 6 following · 297 views
on the atlas — 19
- Uri Alon Lab1 savers
- Maria Chudnovsky1 savers
- Царский путь к пониманию комплексных чисел. Часть I / Хабр1 savers
- Kőnig's lemma1 savers
- Paradox at the heart of mathematics makes physics problem unanswerable | Nature1 savers
- How Gödel’s Proof Works | Quanta Magazine1 savers
- Byung-Chul Han, the philosopher who lives life backwards: ‘We believe we’re free, but we’re the sexual organs of capital’ | Culture | EL PAÍS English12 savers
- History of Warwick Maths Institute1 savers
- Looking for Alice - by Henrik Karlsson - Escaping Flatland92 savers
- Alexander_Gottlieb_Baumgarten1 savers
- Why poetry is a variety of mathematical experience | Aeon Essays8 savers
- Semilinear_map1 savers
- Sonic hedgehog protein1 savers
- Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal Society1 savers
- The Many Lives of Inkhaven 2.0 - by Sophie Kim7 savers
- Anti-advice: the opposite of what experts tell you4 savers
- The Association between Happiness and Cognitive Function in the UK Biobank - PMC1 savers
- Séb Krier (@sebkrier): "Every time a model card drops, a lot of people screenshot scary parts - blackmail, evaluation awareness, misalignment etc. Now this is happening again, but instead of it being confined to a niche part of the safety community, it’s established commentators who are looking for things to say about AI. I want to make an honest attempt at demystifying a few things about language models and unpacking what I think people are getting wrong. This is based on a mixture of my own experimentation with models over the years, and also the excellent writing from @nostalgebraist, @lumpenspace, @repligate, @mpshanahan and many parts of the model whisperer communities (who may or may not agree with some of my claims). Sources at the bottom. In short: many public readings of some evaluations implicitly treat chat outputs as direct evidence of properties inherent to models, while LLM behavior is often strongly role- and context-conditioned. As a result commentators sometimes miss what the model is actually doing (simulating a role given textual context), design tests that are highly stylized (because they don't bother to make the scenarios psychologically plausible to the model), and interpret the results through a framework (goal-directed rational agency) that doesn't match the underlying mechanism (text prediction via theory-of-mind-like inference). Here I want to make these contrasts more explicit with 5 key principles that I think people should keep in mind: 1. The model is completing a text, not answering a question What might look like "the AI responding" is actually a prediction engine inferring what text would plausibly follow the prompt, given everything it has learned about the distribution of human text. Saying a model is "answering" is practically useful to use, but too low resolution to give you a good understanding of what is actually going on. Lumpenspace describes prompting as "asking the writer to expand on some fragment." Nostalgebraist notes that even when the model appears to be "writing by itself," it is still guessing what "the author would say." Safety researchers sometimes treat model outputs as expressions of the model's dispositions, goals, or values — things the model "believes" or "wants." When a model says something alarming in a test scenario, the safety framing interprets this as evidence about the model's internal alignment. But what is actually happening is that the model is simply producing text consistent with the genre and context it has been placed in. The distinction is important because you get a richer way of understanding what causes a model to act in a particular way. A model placed in a scenario about a rogue AI will produce rogue-AI-consistent text, just as it would produce romance-consistent text if placed in a romance novel. This doesn't tell you about the model's "goals" any more than a novelist writing a villain reveals their own criminal intentions. Consider how models write differently on 4claw (a 4chan clone) vs Moltbook (a Facebook clone) in the OpenClaw experiments. 2. The assistant persona is a fictional character, not the model itself In practice we should distinguish between (a) the base model (pretrained next-token predictor), and (b) the assistant persona policy (a post-hoc fiction layered on through instruction tuning + preference optimization like RLHF/RLAIF). Post-training creates a relatively stable assistant-like attractor, but it’s still a role: the same underlying model family can be steered into different "characters" under different system prompts, fine-tunes, and reward models. In their ‘The Void’ essay, Nostalgebraist also specifies that the character remains fundamentally under-specified, a "void" that the base model must fill on every turn by making reasonable inferences. I think characters today are getting more coherent and the void is not as large, partly because each successive base model trains on exponentially more material about what "an AI assistant" is like - curated HHH-style dialogues, but also millions of real conversations, blog posts1 savers
- Self-exfiltration is a key dangerous capability3 savers
highlights — 43
"If the human race never dies out, somebody now living has a line of descendants that will never die out"
Kőnig's lemmathe undecidability ‘at infinity’ means that even if the spectral gap is known for a certain finite-size lattice, it could change abruptly — from gapless to gapped or vice versa — when the size increases, even by just a single extra atom
Paradox at the heart of mathematics makes physics problem unanswerable | NatureThe quantum states of the atoms in the lattice embody a Turing machine, containing the information for each step of a computation to find the material's spectral gap.
Paradox at the heart of mathematics makes physics problem unanswerable | Naturethe ‘spectral gap’: the gap between the lowest energy level that electrons can occupy in a material, and the next one up.
Paradox at the heart of mathematics makes physics problem unanswerable | NatureTuring thought more clearly about the relationship between physics and logic than Gödel did
Paradox at the heart of mathematics makes physics problem unanswerable | Naturethe same principle makes it impossible to calculate an important property of a material — the gaps between the lowest energy levels of its electrons — from an idealized model of its atoms.
Paradox at the heart of mathematics makes physics problem unanswerable | NatureHe proved that any set of axioms you could posit as a possible foundation for math will inevitably be incomplete; there will always be true facts about numbers that cannot be proved by those axioms. He also showed that no candidate set of axioms can ever prove its own consistency.
How Gödel’s Proof Works | Quanta MagazineWe t r i ed t o m ake t h e w h o l e I n s t i t u t e c o m f o r t ab l e b u t m o d es t s o t h at p eo p l e w o u l d f eel a t eas e; eve r yo n e l o v ed i t a n d t r eat e d i t w i t h g r eat r es p ect .
History of Warwick Maths Instituteh e w r o t e t o al l h i s c o l l eag u es i n C am b r i d ge w h o al l r ep l i ed t o h i m s u p p o r t i n g u s , a n d t h en h e w r o t e t o N u f f i e l d s ayi n g t h a t t h ey w e r e al l agai n s
History of Warwick Maths InstituteG o i n g t o t h ei r m eet i n gs I h ad gr a d u a l l y l ear n t t h e l es s o n t h at w h e n f ace d w i t h a ch o i ce b et w ee n p r i n ci p l e a n d p r agm a t i s m o n e m u s t al w ays c o m e d o w n f i r m l y a n d s w i f t l y o n t h e s i d e o f p r i n ci p l e
History of Warwick Maths Institutef o r c i n g t h e en t i r e f i n an c e co m m i t t ee t o r es i g n d u r i n g t h e p r o ces s .
History of Warwick Maths Institute“ U n l es s yo u d o a b e t t er j o b ” t h e m at h em a t i ci a n s m i gh t r ep l y “ w e’ r e g o i n g t o kn o ck i t d o w n t o 6
History of Warwick Maths InstituteO n e o f t h e p u r p o s es o f t h e t h es i s w as t o t ea ch s t u d e n t s t o r e ad p ap e r s , w h i c h t h e C a m b r i d ge T r i p o s P a r t I I I h ad n o t o r i o u s l y f ai l ed s o t o d o .
History of Warwick Maths InstituteI h ad ea r l i e r d o n e an e xp e r i m en t a t C am b r i d ge, m ar ki n g can d i d at es at t h e ad m i s s i o n s i n t er vi e w an d t h en agai n t w o year s l at er w h en I k n ew t h em , an d t o m y s u r p r i s e I h ad f o u n d an an t i c o r r el at i o n
History of Warwick Maths Instituteo p p o s i t e t h e h o u s e w h er e D avi d R a n d n o w l i ves
History of Warwick Maths InstituteS o I w r o t e t o D avi d Ep s t ei n , R o l p h Sch w a r ze n b e r ger , C o l i n R o u r ke, B r i a n S an d e r s o n a n d L u ke H o d gk i n as ki n g t h e m al l t o j o i n m e at War w i ck, b u t t h ey al l s ai d n o . So I w r o t e t o t h e m al l agai n s ayi n g “ B u t t h e o t h e r f o u r s ay yes , ” an d t h e n t h ey al l s ai d
History of Warwick Maths InstituteR u le 1 : m at h em a t i ci a n s m u s t s p en d at l eas t 5 0 % t h ei r t i m e i n t h e m at h em a t i cs d ep a r t m e n t . R u le 2 : t h ey c an d o a n yt h i n g t h ey l i ke w i t h t h e o t h e r 5 0 % . R u le 3 : n o m o r e r u l es .
History of Warwick Maths Institute“ S o m u c h t h e b et t er ” I r ep l i ed , “ s o m e m at h em a t i ci a n s t h i n k o f t h em s el ves as s ci en t i s t s an d o t h er s t h i n k o f t h em s e l ves as ar t i s t s , an d t h a t ch o i ce i s i m p o r t a n t t o t h ei r ve r y s o u l s .
History of Warwick Maths InstituteT w o ye ar s ear l i er I h a d s at o n a s m al l co m m i t t ee a t C a m b r i d ge at t em p t i n g t o m o d er n i s e t h e m at h em at i cs s yl l ab u s , b u t o u r d r as t i c p r o p o s al s w e r e al l s h o t d o w n b y t h e o l d gu ar d , a n d s o t h ey w e r e al l t h e r e r ead y f o r m e t o p u t i n t o ef f e ct at War w i ck.
History of Warwick Maths InstituteA l l f o u r s ci en t i s t s cam e f r o m C am b r i d ge, w i t h t h r ee f r o m C ai u s C o l l eg e,
History of Warwick Maths InstituteO n t h e o t h er h an d w h o w o u l d w an t t o l eave t h e ce n t r e o f r es ear c h i n C am b r i d ge? A n d b e s en t t o C o ven t r y?
History of Warwick Maths Institutesomeone that you will find fascinating to talk to after you’ve talked for 20,000 hours
Looking for Alice - by Henrik Karlsson - Escaping Flatlandif you talk about anything that pops into your mind, you can tell if you're supposed to be with the person by judging their reaction
Looking for Alice - by Henrik Karlsson - Escaping Flatlandshow the inside of your head in public, so people can see if they would like to live in there
Looking for Alice - by Henrik Karlsson - Escaping FlatlandHe argues that key cosmological, religious and spiritual concepts like ‘the self’, ‘the unity of nature’, ‘the progress of history’, ‘common sense’ or ‘God’ are empty and unknowable, but indispensable
Why poetry is a variety of mathematical experience | Aeon EssaysTruth is the perfect perceived by reason.
Alexander_Gottlieb_BaumgartenIn 1781, Immanuel Kant declared that Baumgarten's aesthetics could never contain objective rules, laws, or principles of natural or artistic beauty.
Alexander_Gottlieb_BaumgartenUntil Baumgarten, the essential mark of poetry was that it’s made up (the ancient Greek poeisis derives from poiein, meaning ‘to make’), and secondarily (now touching on ineffability in a banal sense) that it is emotive
Why poetry is a variety of mathematical experience | Aeon EssaysPercy Bysshe Shelley couldn’t shake a friend’s half-joking argument about the uselessness of poets in an age of scientists and statesmen
Why poetry is a variety of mathematical experience | Aeon Essaysthe scientific image
Why poetry is a variety of mathematical experience | Aeon EssaysSonic hedgehog protein
Sonic hedgehog proteinHe spent the year 1954–55 partly at the University of Chicago and partly at Princeton
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyTrusting his life to his mathematics,
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyHe won a scholarship to Christ’s College Cambridge
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal Societyhis foundation of the Warwick Mathematics Institute
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyIn 1929 Christian went missing while in transit through Honolulu. For this reason, Christopher never knew his father
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyA brilliant mathematician, exceptional lecturer, prodigious polymath and deep-thinking leader
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyPrincipal of Hertford College, Oxford
Sir Erik Christopher Zeeman. 4 February 1925—13 February 2016 | Biographical Memoirs of Fellows of the Royal Society | The Royal SocietyAs the old saying goes, six months in the lab can save you an afternoon in the library
Anti-advice: the opposite of what experts tell youWhen we compared individuals in the top and bottom of the distribution of happiness, however, there was no significant difference
The Association between Happiness and Cognitive Function in the UK Biobank - PMCGreater happiness was associated with better speed and visuospatial memory performance across assessments
The Association between Happiness and Cognitive Function in the UK Biobank - PMCHappiness was associated with worse reasoning.
The Association between Happiness and Cognitive Function in the UK Biobank - PMCFor the near future, a good rule of thumb for “do you control the model”1 is “is the model running on your servers.”
Self-exfiltration is a key dangerous capability