Thomas Read

100 posts

Thomas Read

Thomas Read

@thjread

Interpretability researcher at UK AISI

London, UK Inscrit le Kasım 2018
233 Abonnements103 Abonnés
Thomas Read
Thomas Read@thjread·
First paper from my team at UK AISI! Excited to have this out there - we have some really great model organisms of conditional underperformance, and tried a lot of different detection techniques to see what works in an adversarial setting
Jordan Taylor@JordanTensor

NEW PAPER from UK AISI Model Transparency team: Could we catch AI models that hide their capabilities? We ran an auditing game to find out. The red team built sandbagging models. The blue team tried to catch them. The red team won. Why? 🧵1/17

English
0
1
5
106
Thomas Read
Thomas Read@thjread·
@tim_hosgood to add context to the other answers: I had no idea before I looked it up but aluminium foil is crazily thin, typically around 0.016 mm according to wikipedia
English
0
0
1
0
Thomas Read
Thomas Read@thjread·
Not been very active on here so time for an update: I'm excited to be starting a PhD at the University of Warwick in October!
English
1
0
7
0
Thomas Read
Thomas Read@thjread·
@_julesh_ @sarah_zrf @grassmannian Think so - a compact manifold is homotopy equivalent to a finite CW complex, so the fundamental group is a quotient of the free group on finitely many generators, hence countable
English
1
0
2
0
julesh
julesh@_julesh_·
@sarah_zrf @grassmannian Guess I can easily think of examples, but they all involve deleting a bunch of points. Is it famous that this isn’t true for manifolds, where you can’t delete just a point?
English
3
0
1
0
Thomas Read
Thomas Read@thjread·
@Birdyword One factor: think US vaccinations are less concentrated among the elderly than Israel's were, so you'd probably expect a larger effect on transmission in the US
English
0
0
1
0
Thomas Read
Thomas Read@thjread·
@dzackgarza @SpookySpctrlSeq Hatcher Prop 0.18 should be useful - shows that attaching a cell by homotopic attaching maps gives homotopy equivalent spaces. Idea is to construct a space with both spaces as deformation retracts, by attaching a sort of "thickened cell".
Thomas Read tweet media
English
1
0
0
0
Thomas Read
Thomas Read@thjread·
@syzygay1 Maybe some kind of wolf spider? Apparently there's a lot of different species which are hard to distinguish
Thomas Read tweet media
English
0
0
1
0
syzygay
syzygay@syzygay1·
Anyone know what type of spider this is
syzygay tweet media
English
2
0
3
0
Aɴᴏɴʏᴍᴏᴜs Eʟʏsɪᴜᴍ
Aɴᴏɴʏᴍᴏᴜs Eʟʏsɪᴜᴍ@lex_with_a_hex·
I have a vague brain-worm about there being an object like some kind of vase that holds dirt and when the vase disintegrates the dirt forms the shape of the vase and there's a poetic quote about the dirt being the new vase. Does this ring any bells?? For either quote or object??
English
2
0
2
0
Thomas Read
Thomas Read@thjread·
@isbellduality A somewhat technical one: a reflexive subcategory is equivalent to the category of algebras of the associated idempotent monad. This gives a proof that the inclusion of a reflexive subcategory creates all limits.
English
0
0
1
0
jord!
jord!@isbellduality·
Okay here’s a question: do any of you know an interesting example of an algebra for a monad that isn’t on Set? I just realized that I don’t know of any and I haven’t come up with anything trying to think about it at the moment. Am I missing an obvious one?
English
7
0
7
0
Thomas Read
Thomas Read@thjread·
@EpistemicHope Officials / ethics people in the UK have actually seemed to be somewhat open to challenge trials since the beginning - this particular one has been in the works since September
English
0
0
1
0
Thomas Read
Thomas Read@thjread·
@Iceland_jack A presheaf on a topological space is a contravariant functor from the poset of open subsets of a topological space to Set - that is, it specifies data attached to subsets of a space. Given the title of the book, I imagine he'll talk about this in detail later!
English
0
0
1
0
Dad×2_jack
Dad×2_jack@Iceland_jack·
What does author Daniel Rosiak mean by functors "specifying data locally", is this a notion from algebraic topology Sheaf Theory Through Examples (arxiv.org/pdf/2012.08669…)
Dad×2_jack tweet media
English
4
0
2
0
Thomas Read
Thomas Read@thjread·
@dzackgarza @Category_Fury Problem is universal constructions come with specified morphisms, and these won't be unique. E.g. there are multiple ways to define product projections from a 6 element set to a 2 element set and a 3 element set. So all you gain is now things are unique up to unique automorphism.
English
0
0
0
0
Thomas Read
Thomas Read@thjread·
@isbellduality More simply, we say a rank k vector bundle E->X is orientable if you can pick generators of H^k(E_x, E_x \ 0) for every x in X such that the generators vary locally trivially. Then what you described is pretty much that the trivial bundle is orientable but the Mobius band isn't
English
0
0
1
0
Thomas Read
Thomas Read@thjread·
@isbellduality There's a fancy story here, which goes something like: real line bundle on S^1 => double cover of S^1 (put an inner product on the bundle and take all norm 1 points) => homomorphism pi_1(S^1) -> Z/2 (theory of covering spaces) => element of H^1(S^1; Z/2) (universal coefficients)
English
1
0
2
0
jord!
jord!@isbellduality·
Question: is there a way to use some sort of cohomology to distinguish, for instance the Möbius band and the cylinder as fiber bundles over S¹? Every intuitive explanation of cohomology talks about the failure of a passage from local to global solutions and it seems like (1/4)
English
4
0
10
0
Thomas Read
Thomas Read@thjread·
@TomChivers To understand the Birmingham data, it's interesting to look at Cambridge's PCR test asymptomatic screening - prevalence was ~1% all term before plummeting in the last couple weeks, which would explain the LFT data. Perhaps students were just really careful before going home!
Thomas Read tweet media
English
0
0
2
0
Thomas Read
Thomas Read@thjread·
This site has a clever suggestion - use CO2 levels as a proxy for indoor COVID risk, since both depend on ventilation and number of people in a similar way. Could be a good metric to produce quantitative guidance for schools / offices / etc in the future.
English
0
0
0
0
Thomas Read
Thomas Read@thjread·
Once again wishing someone would formalise all of undergrad maths, so that I would have a reliable source for correcting sign errors in lecture notes
English
0
0
0
0
Thomas Read
Thomas Read@thjread·
@s8mb @cjsnowdon Amazed by that PHE paper that the gov's airport testing policy was apparently based on - they assume everyone is tested before being allowed to board a flight, but fail to mention this until the end of the appendix
English
0
0
2
0

Découvrir

@tim_hosgood @_julesh_ @sarah_zrf @grassmannian @Birdyword @dzackgarza @syzygay1 @lex_with_a_hex