Phil Steitz

Yearly donation to Wikipedia ✅

Please, do the same, great folks working there!

Out-and-out superb explanation.

A bittersweet story but glad to see a principled stance!

I've received an 'exclusive' invite to try Nature's new 'research assistant', which will burn down a forest to provide a 'summary of the paper' I'm reading and I have SUCH exciting news for them, that's called an 'abstract' and the actual authors already wrote it for me, no forest-burning required.

YouTube video by Devoxx Self Healing Rollouts: Automating Production Fixes with Agentic AI & Argo Rollouts by Carlos Sanch

My @devoxx.com talk "Self-Healing Rollouts" is up: auto creating issues from @argoproj.bsky.social Rollouts canary failures, that get in turn auto-fixed by code assistants like Google Jules or GitHub Copilot and then re-deployed 🤯 (second half is all demo) #devoxx #argo
youtu.be/ABjIUw7ojiU?...

Power and Difference in the time of responsible AI at University of Sheffield on FindAPhD.com | Susan Oman Do a funded PhD with me! We are looking for projects at the cutting edge of interdisciplinary enquiry into AI and inequalities! "Your doctoral research project will be designed by you to respond to...

Do a funded PhD with me! We are looking for projects at the cutting edge of interdisciplinary enquiry into AI and inequalities!

www.linkedin.com/posts/susan-...

www.schneier.com/blog/archive...

This is obviously sick and un-American, but the second order effects are really scary. Imagine what the US would be like if this crap had been going on for say the last 10 years. We really have to get rid of these guys.

The Great Software Quality Collapse: How We Normalized Catastrophe The Apple Calculator leaked 32GB of RAM.

techtrenches.substack.com/p/the-great-...

Nice rant.

Yeah the “receiving” app is Java. Kind of comical that the Java code generated by the python generator is better than most of the Java code I have been able to get Claude to write :).

Had fun with Claude today. I gave it a problem that is painful for humans but right in the LLM wheelhouse. Given an arrow ipc file, read its schema and write a java class that can create List<Thing> from the records in the file. Like what avro does, but for tables. Claude crushed it in python.

How do we measure AI productivity? Here is how Google, GitHub, Microsoft, Dropbox, Atlassian and others measure it:

Article: newsletter.pragmaticengineer.com/p/how-tech-c...

#ML #MachineLearning #ArtificialIntelligence #AI #MLOps #AIOps #DataOps

'I Tracked Amazon's Prime Day Prices. We've Been Played' - Slashdot "Next time Amazon hypes its Prime Days savings, remember this: The prices during the sale aren't always better," writes a Washington Post technology columnist. "I've got the receipts to prove it." I...

news.slashdot.org/story/25/10/...

This is every pricing team’s nightmare: consumers starting to watch the time series. An app to do the tracking and identify real discounts would be cool.

UT Austin Linguistics is hiring in computational linguistics!

Asst or Assoc.

We have a thriving group sites.utexas.edu/compling/ and a long proud history in the space. (For instance, fun fact, Jeff Elman was a UT Austin Linguistics Ph.D.)

faculty.utexas.edu/career/170793

🤘

Cf. OpenAI “Why LMs Hallucinate.” 70%+ calibration error rates on code generation tasks. Any distributional “intuition” is bound to fail. If we could hone intuition to predict errors, so could the LLMs. If the code matters, you need to look at it carefully.

Want to work on open source full time? The @roost.tools engineering team is starting to hatch! Come build OSS tools making a difference in Trust & Safety.

This is a fully remote role, though some schedule overlap with North American time zones is expected.

www.linkedin.com/jobs/view/43...

An unexpected challenge at the start of Naomi Saphra’s career shaped her research as a computer scientist. www.quantamagazine.org/to-understan...

Fred says, “Save animals, save the planet!”

Happy World Animals Day!

With supply-chain attacks on NPM rising, we disabled all install scripts (preinstall, postinstall, etc.) across our projects.

That broke builds for native modules, especially libpq, which we rely on heavily.

Our fix: pre-build libpq for Linux (x86_64 + arm64) and macOS, then publish it to NPM.

Oh yeah. Coco’s stylin’

Editorial Published in ASA Discoveries (Vol. 1, No. 1, 2025)

ASA Discoveries - new journal from the American Statistical Association and published by Taylor & Francis. Now online!

@amstatnews.bsky.social @tandfresearch.bsky.social

Check out the opening Editorial!
⬇️⬇️⬇️⬇️⬇️
www.tandfonline.com/doi/full/10....

Tweet by David Sacks: "Wikipedia is hopelessly biased. An army of left-wing activists maintain the bios and fight reasonable corrections. Magnifying the problem, Wikipedia often appears first in Google search results, and now it’s a trusted source for AI model training. This is a huge problem."

Someone should probably inform the White House's "AI & Crypto Czar" that no one is forcing AI companies to train their models on Wikipedia

This was accepted to #NeurIPS 🎉🎊

TL;DR Impoverished notions of rigor can have a formative impact on AI work. We argue for a broader conception of what rigorous work should entail & go beyond methodological issues to include epistemic, normative, conceptual, reporting & interpretative considerations

Beautiful morning in the Sonoran. Gentle rain and sun on our backs. Slightly softened desert hard pack on the trail. Perfect running conditions.

Interesting long-form argument from Rodney Brooks that methods similar to the ones from the deep learning era of AI that cracked image recognition, text generation, etc. are unlikely to crack humanoid robot dexterity in the near future.

The recent OpenAI paper (arxiv.org/pdf/2509.04664) on LLM hallucinations is interesting. Their recommendations on benchmarks are helpful. The probability analysis points to a bigger practical problem: LLM errors are "all over the place" in a distributional sense. Makes QA intractable.

stat.osu.edu/about/open-p...

The largest gain since 1989, the start date of the data, are among the top 0.1%.

Next up are gains in the top 1% of households.

Here is the link to the data:

federalreserve.gov/releases/z1/da…

Statistical Quantile Learning for Large Additive Latent Variable Models Large and complex datasets, emerging from technological advancements in fields such as genomics and brain imaging, hold ample promise for gaining new scientific insights. Yet, their inherent nonlin...

Statistical Quantile Learning for Large Additive Latent Variable Models www.tandfonline.com/doi/full/10....

Why do economists care about inequality in incomes and wealth?

1) Inequality tends to dampen overall economic performance. One of most immediacy effects is on consumer spending. Low-and middle-income households spend a larger percentage of each dollar that they earn when more spending is in…

Please take this as a warning and help out your local PBS and NPR stations if you can. Don't wait for a fund drive to donate - please just give what you can.
Help keep real journalism alive