Reposted by Calvin McCarterarXiv q-bio.GN Genomics @qbiogn-bot.bsky.social · 01/09/2026Calvin McCarter: Confounder-Aware Feature Correction for Single-Cell Batch Integration arxiv.org/abs/2608.28849 arxiv.org/pdf/2608.28849 arxiv.org/html/2608.28849 031
Calvin McCarter @calvinmccarter.bsky.social · 31/07/2026Finding animal videos that I'm willing to show to my toddler has become a tricky business. If it's not the BBC etc, or an old video, or an old channel with a consistent history, I'm very suspicious. 110
Calvin McCarter @calvinmccarter.bsky.social · 09/07/2026A video of my talk at the ML4PE seminar, "How to Make the Most of Your Masked Language Model for Protein Engineering", is now online: youtube.com/watch?v=Lcmo...m.youtube.comHow to Make the Most of Your Masked Language Model for Protein Engineering - Calvin McCarterYouTube video by ML for protein engineering seminar series 000
Calvin McCarter @calvinmccarter.bsky.social · 09/06/2026(which, aside, ESMC + mean-pooling + TabICLv2 is extremely close to the Kermut state of the art on ProteinGym: github.com/OATML-Marksl...)github.comAdd ESMC+TabICLv2 supervised by calvinmccarter · Pull Request #106 · OATML-Markslab/ProteinGymThis PR adds ESMC-TabICLv2, a supervised variant-effect predictor for the DMS substitution benchmark: mean-pooled ESMC-600M embeddings, PLL, and WT marginal scores, fed to a TabICLv2 tabular founda... 000
Calvin McCarter @calvinmccarter.bsky.social · 09/06/2026i actually would be mostly convinced if there's low test-set real vs scrambled AUROC for TabICLv2 given mean-pooled embeddings. 110
Calvin McCarter @calvinmccarter.bsky.social · 09/06/2026it's more the top-2 tSNE that worries me. if (hypothetically) a model has a single channel for indicating whether "this sequence looks like nonsense to me" (which would be reasonable behavior, imo) top-2 tSNE wouldn't necessarily pick that up. 110
Calvin McCarter @calvinmccarter.bsky.social · 09/06/2026I think it's more accurate to say that top-2 tSNEs of mean-pooled ESM-C representations is unable to distinguish real from scrambled CDRs. 100
Calvin McCarter @calvinmccarter.bsky.social · 09/06/2026can we estimate the fraction of claude code and codex active users that use openrouter? 000
Calvin McCarter @calvinmccarter.bsky.social · 06/06/2026i wonder whether milton is also thinking of jesus being a virgin, while also speaking of gathering the people like a mother bird gathers her chicks. 020
Calvin McCarter @calvinmccarter.bsky.social · 06/06/2026in this particular case, i think it is a direct quote of genesis 1, where in the hebrew, the spirit is literally brooding like a mother bird. (the spirit of god is female in the hebrew scriptures, but then becomes male in the new testament, due to the virgin birth narrative.) 120
Calvin McCarter @calvinmccarter.bsky.social · 05/06/2026i think Milton and Blake were actually pretty similar on the former? they were both pioneers in the married incel apologia genre 141
Calvin McCarter @calvinmccarter.bsky.social · 12/05/2026the nucleus sampling paper was really good at this imo arxiv.org/abs/1904.09751arxiv.orgThe Curious Case of Neural Text DegenerationDespite considerable advancements with deep neural language models, the enigma of neural text degeneration persists when these models are tested as text generators. The counter-intuitive empirical obs... 020
Calvin McCarter @calvinmccarter.bsky.social · 11/05/2026there's a startup working on this (advised by George Church, with funding from Tyler Cowen) helenarosengarten.com/how-to-sleep...helenarosengarten.comSleep Compression - HelenaHi, I am Helena! 050
Calvin McCarter @calvinmccarter.bsky.social · 02/05/2026so, ime, there was a barbell-like mix of wanderlust / high-agency formerly big city escapist people and phlegmatic people in alaska. eg one tour guide retired from broadway, and just moved there. but then anchorage esp feels like it has lots of people who have no reason to live there. 010
Calvin McCarter @calvinmccarter.bsky.social · 02/05/2026it was a great trip, 11/10. seward and denali are especially great; almost made it to whittier (which one of the videos is about) but ran out of time. 110
Calvin McCarter @calvinmccarter.bsky.social · 01/05/2026as an alaska tourist, i noticed that the locals were a bit... different, and also randomly drove by non-touristy neighborhoods that made me wonder "what are these people doing here?", and the series answered some of my questions. 110
Calvin McCarter @calvinmccarter.bsky.social · 29/04/2026the battle of empel is another one where the spanish were in a sticky situation but encouraged themselves with a self-fulfilling victory omen 110
Calvin McCarter @calvinmccarter.bsky.social · 11/04/2026yeah so, assuming that where civs meet, they just stop expanding and respect their neighbors, you get the following distribution over civ size. because the x axis is log-scaled, this implies a fairly high level of inequality in the distribution of civ size. 131
Calvin McCarter @calvinmccarter.bsky.social · 08/04/2026i think i was subconsciously combining "my own personal vietnam" with bsky.app/profile/norv... 020
Calvin McCarter @calvinmccarter.bsky.social · 01/04/2026I've previously found interesting results from using -log(Euclidean distance) attention. On shallow networks, it seems to train more easily and learn interpretable keys. arxiv.org/abs/2310.18805 github.com/calvinmccart...arxiv.orgInverse distance weighting attentionWe report the effects of replacing the scaled dot-product (within softmax) attention with the negative-log of Euclidean distance. This form of attention simplifies to inverse distance weighting interp... 000
Calvin McCarter @calvinmccarter.bsky.social · 29/03/2026exactly. (having said all this, wearing my substack special purpose vehicle shareholder hat, i am rather pleased...) 030
Calvin McCarter @calvinmccarter.bsky.social · 29/03/2026so I'm purely a browser user, but yes it does (screenshot). the problem is that the feed surfaces "substack notes" and replies thereto, but not "substack articles" and replies thereto. 240
Calvin McCarter @calvinmccarter.bsky.social · 29/03/2026it's also hard for me to get the full list of people i follow -- like what if i decide i want to subscribe instead? 130
Calvin McCarter @calvinmccarter.bsky.social · 29/03/2026another bad thing is that if i follow someone, i don't find out about their comments on various articles. an even worse thing is that if someone follows me, they don't find out about my comments on various articles. 140
Calvin McCarter @calvinmccarter.bsky.social · 26/03/2026they've reached parity on the dual problem (unfortunately with degenerate solutions) 131
Calvin McCarter @calvinmccarter.bsky.social · 26/03/2026Of all sad words of LLM, The saddest are these: "Failed to fetch arxiv.org" again 000
Calvin McCarter @calvinmccarter.bsky.social · 22/03/2026yes, allegedly the Munich Conference was good because it bought time for GB to prepare for war, while clarifying that the next takeover on Eastern Europe would bring GB to intervene 110
Calvin McCarter @calvinmccarter.bsky.social · 22/03/2026interesting article: big for Neville Chamberlain if true 111
Calvin McCarter @calvinmccarter.bsky.social · 21/03/2026speaking of which, it would be cool for TMLR to be a place which tries out various improved reviewing ideas! 000
Calvin McCarter @calvinmccarter.bsky.social · 21/03/2026a similar tweak would be to require submissions be posted on arxiv (to be verified in some automated fashion), while still keeping the venue itself double-blinded. authors shouldn't be able to force reviewers to read something, if they don't want anyone else to see that they wrote it. 110
Calvin McCarter @calvinmccarter.bsky.social · 20/03/2026tangential, but "set the mood for people to have communal religious experiences at church" would make almost certainly make Bach flip out. if we're going to insist that his motivations matter, then we should at least be honest about what those were 220
Calvin McCarter @calvinmccarter.bsky.social · 19/03/2026well, but the lesson from that is that you need to time the market correctly in order to make money; and normal people shouldn't try to time the market. 120
Calvin McCarter @calvinmccarter.bsky.social · 17/03/2026a metaphor for x (formerly twitter) @norvid-studies.bsky.social 152
Calvin McCarter @calvinmccarter.bsky.social · 17/03/2026both of these have wikipedia articles featuring what TR thought about it -- coincidence? 120
Calvin McCarter @calvinmccarter.bsky.social · 16/03/20262026: celebrating 250 years of having taxes that are "very low by european standards" 040
Calvin McCarter @calvinmccarter.bsky.social · 14/03/2026what's the incentive here? there are no creator payouts... 230
Calvin McCarter @calvinmccarter.bsky.social · 14/03/2026i suspect that GDM has a solution that people in academia are either unable to invent or unable to afford (guessing the latter) 010
Calvin McCarter @calvinmccarter.bsky.social · 13/03/2026it's hard to learn what CDRs do due to their messy structures and the lack of evolutionary info about them in standard ab sequence datasets 100
Calvin McCarter @calvinmccarter.bsky.social · 13/03/2026obnoxiously tagging @kevinkaichuang.bsky.social @delalamo.xyz @austinjtripp.bsky.social about our new PLM paper -- hope you find it interesting and/or useful! 210
Calvin McCarter @calvinmccarter.bsky.social · 13/03/2026When supervised models were included via multi-objective guidance, we achieved a 100% synthesizability-and-binding success rate in vitro. But a cautionary note: guidance improved target objectives at the cost of reduced humanness. All the details are here: arxiv.org/abs/2603.10302arxiv.orgHow to make the most of your masked language model for protein engineeringA plethora of protein language models have been released in recent years. Yet comparatively little work has addressed how to best sample from them to optimize desired biological properties. We fill th... 041
Calvin McCarter @calvinmccarter.bsky.social · 13/03/2026Another surprise: ESM-2, trained on generic proteins, was highly competitive with antibody-specialized models for antibody optimization. Meanwhile, AbLang2 — trained on human antibodies — sometimes produced less human sequences than ESM-2. Training data ≠ output bias in the ways you'd expect. 6/7 100
Calvin McCarter @calvinmccarter.bsky.social · 13/03/2026Perhaps the most provocative finding from the in vitro experiments: choice of sampling algorithm matters at least as much as choice of model. Beam search consistently outperformed Gibbs across every model where both were tested. 5/7 110
