Leonard Stimpfle @leonardstimpfle.eurosky.social · 02/10/2026The speedup for `estimate_u_using_random_sampling` is magical! 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 20/09/2026Thanks! I don't quite understand the "no independence" critique. Isn't "a test of marginal name sensitivity on one document" what you'd want to do when screening from a (non-randomly drawn) pool of candidates? 1100
Reposted by Leonard StimpfleTerence Tao @teorth.bsky.social · 11/09/2026A group of 25 Fields Medalists, including myself, have made a joint declaration on Math and AI: mathandai.org . We welcome additional signatories. See also this article in the Economist announcing the declaration: www.economist.com/science-and-...mathandai.orgDeclaration — Math and AIRead the declaration and add your name. 422060932
Reposted by Leonard StimpfleGlyph @glyph.im · 30/08/2026the social contract that allows the web to exist at all is literally collapsing in real time people.kernel.org/monsieuricon/cree…people.kernel.orgCreepy crawliesComments 654126
Leonard Stimpfle @leonardstimpfle.eurosky.social · 28/08/2026Coming from the Python world this is surprising: I believe astral's typechecker ty relies heavily on salsa and there was/is a lot of hype around Python tooling built in Rust 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 20/08/2026Any recommended reading on the paper's shortcomings (papers, blogs, etc)? 230
Leonard Stimpfle @leonardstimpfle.eurosky.social · 19/08/2026This @duckdb.org blog post illustrates this nicely duckdb.org/2024/11/22/r... 020
Reposted by Leonard StimpfleRichard McElreath 🐈⬛ @rmcelreath.bsky.social · 13/08/2026These ratio outcome variables are a plague in the sciences. If your analysis treats 1/2 and 10/20 as the same evidence, then you have discarded uncertainty in the outcome. Model what you measure, not what you construct. Use the denominator as an exposure instead. 612237
Leonard Stimpfle @leonardstimpfle.eurosky.social · 07/08/2026OpenAI made me come back: There's a "Cambrian explosion" of "collective intelligence" including some "(real quotes)" youtu.be/87DyyMV0kCY?... 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 06/08/2026This is very cool, thanks! Minor nitpick: I think it would clearer if the y-axis wasn't automatically scaled and always started at zero (see for example ets.bruegel.org/installation...) 000
Reposted by Leonard StimpfleŁukasz Langa @lukasz.langa.pl · 05/08/2026You can help shape the future of #Python typing with the power of having an opinion! Fill out the annual Python Type System and Tooling Survey: www.surveymonkey.com/r/python_typ...surveymonkey.comPython Type System and Tooling Survey 2026The Python Typing Survey offers a comprehensive look at the current state of Python’s type system and developer tooling. It captures the evolving sentiment, challenges, and opportunities around Python... 275
Leonard Stimpfle @leonardstimpfle.eurosky.social · 02/08/2026It's what type theorists refer to as setlist 071
Reposted by Leonard StimpfleHenry Yuen @henryyuen.bsky.social · 01/08/2026Some initial thoughts, and a complicated mix of feelings. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not. 232972
Leonard Stimpfle @leonardstimpfle.eurosky.social · 06/07/2026As a non-pianist I learned a lot from this piano channel: www.youtube.com/@WalkThatBass 000
Leonard Stimpfle @leonardstimpfle.eurosky.social · 07/06/2026But it looks good, no?https://bsky.app/profile/seema.bsky.social/post/3lbkfet44l22a 000
Leonard Stimpfle @leonardstimpfle.eurosky.social · 06/06/2026Love it (because of him) en.wikipedia.org/wiki/Hamilto...en.wikipedia.orgHamilton de Holanda - Wikipedia 110
Leonard Stimpfle @leonardstimpfle.eurosky.social · 23/05/2026I've moved back to writing from scratch and asking Claude for feedback once I've got something working. Even then I often end up exasperated at how convoluted Claude seems to have become 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 23/05/2026Learned a lot, thanks! If natural refrigerants are "better in almost every way", what drove adoption of CFCs/HFCs in the first place? Is it toxicity/inflammability? 130
Reposted by Leonard StimpfleMax Parzen @maxparzen.bsky.social · 17/05/2026#Open grid data is a public good improving security and not a threat. Support by signing & sharing the petition. 🔗 to the petition: tally.so/r/xX08eE ➡️ blog by @nworbmot with details: nworbmot.org/blog/open-gr... ➡️ impact examples mapyourgrid.org/impact/ #freethemodels 021
Leonard Stimpfle @leonardstimpfle.eurosky.social · 02/05/2026Any suggestions for further reading on the "don't use relative imports" point? 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 29/04/2026Out of curiosity: When you say "didn't find anything" do you mean anything better than hairballs? Any alternatives to consider in 2d? 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 29/04/2026More excited than reasonable about sentinel values. It'll be a very nice quality of life addition. The `reverse_entropy_in` example though just shouldn't have a default value instead of a sentinel, no? 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 26/04/2026Great work, thanks! I'm wondering if the framing in terms of effective/ineffective schemes is helpful though? What does "only nine effective schemes remain" mean in view of constant tinkering by policymakers and massive changes in carbon prices (see, e.g., the EU ETS)? 110
Reposted by Leonard StimpflePaul Goldsmith-Pinkham @paulgp.com · 21/04/2026Python implementation of `rdrobust` that is significantly faster! github.com/leostimpfle/... These runtimes are in logscale!! 3273
Leonard Stimpfle @leonardstimpfle.eurosky.social · 21/04/2026Would love that! I initially just wanted to get it to run with `numpy>=2.0` but along the way realised that there were a few easy wins for speed which really mattered for my use case 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 21/04/2026JIT compiling the nn matching with `numba` is indeed the largest driver but there are gains in other places too (attached are benchmarks for `hc0` plug-in residuals) 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 20/04/2026Contributions and bug reports very welcome at github.com/leostimpfle/... 000
Leonard Stimpfle @leonardstimpfle.eurosky.social · 20/04/2026I have started to work on `pyrdd` for fast estimation of RDDs in Python. It is essentially a refactored `rdrobust` which is great but struggled with the data I was throwing at it and also seems incompatible with the newer versions of `numpy`. Some early benchmarks attached (note the log scale). 210
Reposted by Leonard StimpfleAlex @s3alfisc.bsky.social · 23/03/2026I am very excited that PyFixest 0.50.0 is on PyPi, including a new graph-based solver for demeaning that makes fixed effects estimation in PyFixest significantly faster for "sparse" fixed effects structures. 1157
Leonard Stimpfle @leonardstimpfle.eurosky.social · 15/02/2026Yes! Plus git repositories for the underlying code/data free for anyone to fork and play around with. One can dream after all... 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 15/02/2026What are your thoughts on how an alternative (publication) model should look like? 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 25/01/2026Similar issues exist for paper and chemicals installations which receive free allocation for heat that may be produced in separate CHP plants. 000
Leonard Stimpfle @leonardstimpfle.eurosky.social · 25/01/2026Great overview, thanks! One comment re Figure 2: The aggregations based on activity types can be misleading because they fail to capture free allocation boundaries. For example, steel is short since 2013 when accounting for transfers of waste gases from blast furnaces to power plants. 100
Leonard Stimpfle @leonardstimpfle.eurosky.social · 23/01/2026"To ascertain the influence of ChatGPT outside the academic context and beyond scripted speech, we analyzed 771,591 podcast episodes spanning diverse topics." 130
Leonard Stimpfle @leonardstimpfle.eurosky.social · 23/01/2026The authors do acknowledge this channel: "despite 56% of YouTube academic talks containing conversational elements, these linguistic shifts stem primarily from direct machine-to-human transmission—such as speakers reading from ChatGPT-co-authored materials" 110
Leonard Stimpfle @leonardstimpfle.eurosky.social · 18/01/2026Along similar lines, @robinlinacre.bsky.social recently shared his take on "respectful" AI use which really resonated with me 130
Leonard Stimpfle @leonardstimpfle.eurosky.social · 05/01/2026It would be very cool if this helped keeping published work up to date at low cost. I thought this was also interesting from a replication/reproducibility perspective more generally 010
Leonard Stimpfle @leonardstimpfle.eurosky.social · 05/01/2026Thoughts on this? www.linkedin.com/posts/andrew...linkedin.comHere's proof that Claude Code can now write an entire empirical polisci paper. Over the weekend, I had Claude Code fully replicate and extend an old paper of mine estimating the effect of univer...Here's proof that Claude Code can now write an entire empirical polisci paper. Over the weekend, I had Claude Code fully replicate and extend an old paper of mine estimating the effect of universal ... 120
Reposted by Leonard StimpfleAlex @s3alfisc.bsky.social · 03/01/2026If you are aware of any other high quality implementations of "modern" estimators in Python, please let us know! 111
Reposted by Leonard StimpfleHanno Böck @hanno.hboeck.de · 30/12/2025Was there a massive leak of a dangerous greenhouse gas in Iceland🇮🇸 in 2011-or was it a data📈 reporting error? Emission databases show that the Norðurál aluminium factory🏭 released 60 tons of SF₆ in 2011 — but no such emissions in any other year. 🔌💡 industrydecarbonization.com/news/errors-... 🧵industrydecarbonization.comErrors and Inconsistencies in European Emission DataFactories that suddenly have emissions a thousand times lower than before, unexplained data gaps, and inconsistent CO₂ emissions from different data sources are some of the issues one finds when havin... 164
Leonard Stimpfle @leonardstimpfle.eurosky.social · 09/12/2025Great to finally see Ruff integrated! 000
