Nathaniel Bechhofer retweeted
Yesterday I gave a talk on “Context Fails & Fixes”, that functioned a bit on a general “State of Context Engineering”: where it came from, where it's gone, and where I think it's going. dbreunig.com/2026/10/24/the-…
Nathaniel Bechhofer retweeted
Worst case of " 'exponential' means 'a lot' " I've seen so far.
If you had an exponential amount of compute you could solve simulate biology exactly and wouldn't need to run any experiments.
Some of you will think "of course, but that's an unrealistic assumption" and others will think "no, you need experiments even with that much compute. That's just How It's Done".
You would of course need evaluations once you have a solution, but you don't in principle need experiments to find it: Experiments are a way to save compute, dating from a time when it was much more expensive.
And I think the presentation is exponential but real systems aren't irreducibly exponential. AlphaFold worked, after all.
Nathaniel Bechhofer retweeted
iOS apps that have baked-in secrets are in trouble.
The right solution is using AppAttest attestations to authenticate to services instead of baked in API keys:
developer.apple.com/document…
If you're on MacOS 27.0.1 and make a fresh Apple ID. You can now decrypt IPAs on MacOS and after a one-time enrollment also on Linux or Windows if needed. github.com/drwhax/ipatool more builds will be supported soon. No jailbreak needed.
Nathaniel Bechhofer retweeted
We should be more ambitious with the software we build:
danluu.com/ambitious/
![Since early this year, Michael Malis (of pgrust) has been showing me interesting things he's done and saying
[W]e should be more ambitious1 with the software we build
There is, understandably, a lot of skepticism about projects like pgrust because there are so many low-quality AI-generated projects out there. If I didn't know Malis and didn't look into it, I would just assume that pgrust was another vibe slopped project that's never going to work. But I know Malis well enough to trust his work without looking into it, and if you look into it a bit, you'll see that in testing pgrust, he's testing thoroughly enough that he's even uncovering real bugs in Postgres, such as this numerical bug that can cause valid SP-GiST operations to fail, this format cache bug that can cause incorrect results from JSONPath, this pgcrypto bug that can cause later operations to fail, multiple memory safety bugs in the dictionary loader, this money overflow bug, etc.
This isn't to say that pgrust i...](https://pbs.twimg.com/media/HVbOGdKboAAAbtO.png?name=orig)
ALT Since early this year, Michael Malis (of pgrust) has been showing me interesting things he's done and saying [W]e should be more ambitious1 with the software we build There is, understandably, a lot of skepticism about projects like pgrust because there are so many low-quality AI-generated projects out there. If I didn't know Malis and didn't look into it, I would just assume that pgrust was another vibe slopped project that's never going to work. But I know Malis well enough to trust his work without looking into it, and if you look into it a bit, you'll see that in testing pgrust, he's testing thoroughly enough that he's even uncovering real bugs in Postgres, such as this numerical bug that can cause valid SP-GiST operations to fail, this format cache bug that can cause incorrect results from JSONPath, this pgcrypto bug that can cause later operations to fail, multiple memory safety bugs in the dictionary loader, this money overflow bug, etc. This isn't to say that pgrust i...
Nathaniel Bechhofer retweeted
it says so much that gyges, in a highly critical post, suggests that the basic bayesian term “priors” originates in or owes anything to yudkowsky (cf. Edwards, Lindman, Savage 1963 below). what of rationalism isn’t like this?
An attempt to document important facts concerning AI Safety, Eliezer Yudkowsky, Rationalism and San Francisco Effective Altruism
verysane.ai/p/ai-safety-is-m…
Nathaniel Bechhofer retweeted
These are calibration plots — which says nothing about accuracy.
These are nice pictures, but they don’t prove accuracy
Nathaniel Bechhofer retweeted
It also gets the Apple story wrong. The right version of that history is that Apple—and Foxconn—tapped into existing manufacturing capabilities in China. Before Apple, China already had a massive network of component producers and an incredibly skilled labor force. Apple repurposed and added some of these capabilities to iPhone production on top of that baseline. Apple should get the credit for jump starting handset production in China, but the spillover story is in the other direction as well. Apple went to China because of China’s existing general manufacturing capabilities in the first place and it was able to retool and upgrade these existing capabilities for its own production. That is not Arizona story at all.
1/ Dan Wang’s NYT essay on TSMC's Arizona plant argues that TSMC is doing for America what Apple did for China. But the analogy rests on two assumptions and it's not clear that either one of them is holding.
nytimes.com/2026/10/23/opini…
Nathaniel Bechhofer retweeted
Nepal reduced extreme poverty almost as fast as China. It did not have much FDI, manufacturing, the agricultural share of the economy remained high, and there was an active civil conflict.
How did Nepal do it? Migration, largely to the Gulf: indevelopmentmag.substack.co…
Nathaniel Bechhofer retweeted
Having used Gemini since 3.1, as of 3.8, it is good enough now for most work, I think most models are, provided you are a domain expert and can steer the model.
This bug was first found and fixed by Gemini Flash 3.7, and most experiments in the paper were done by Flash 3.8. Gemini 4 Argon wrote the first draft of the paper and figures, which I aggressively nitpicked until the content was solid but the writing was not. Then I fed Argon a handful of my first-author papers from the last few years and had it do a rewrite in my style (which worked astonishingly well), and then I manually did a rewrite (the hands + keyboard variety) after that. I really liked this workflow! If you like my writing style feel free to throw some of my old papers in your context window and do the same.
Nathaniel Bechhofer retweeted
tidycensus 2.0.0, one of the milestones from this summer's R Consortium grant, is now on CRAN.
The package works the same as before, but with a modernized httr2 backend and lots of internal improvements to make data assembly faster and more resilient to failures.
Plus I was able to clear out almost all of the outstanding issues, adding support for several new datasets / geographies and adding some long-requested features like auto-labels for variables.
The 2025 1-year ACS drops next week, so update today and get ready for the new data.
Not an R user? Tell your AI agent to "use tidycensus" and you'll find your agentic Census work so much smoother.
Nathaniel Bechhofer retweeted
Thank god. Finally
Any form of security through obscurity is fully over. This barely worked as a method with humans, but just simply is over in a world of autonomous systems that will look through every crevice and try every permutation of attack possible.
As a result, years and years of legacy enterprise infrastructure will need to be upgraded, security practices modernized, data pipelines cleaned up, access controls better managed, and much more. Agents will help on the defensive side because they can both accelerate this work and equally simulate the risk vectors.
But there’s a long road ahead to prepare for this future. This is also partially why diffusion timelines will take longer than some realize.
Nathaniel Bechhofer retweeted
I think there’s two interpretations of this claim that imply very different world views.
1. If you run many A/B tests, you can accumulate a lot of small effects that add up.
2. If you run many A/B tests, you get enough draws from a heavy tailed distribution to find big wins.
Nathaniel Bechhofer retweeted
Twice this week I've cloned a repo comparing decision models and significantly improved their results by running DSPy's ReAnchor. It does not take much. cmpnd.ai/blog/building-jev-p…
Nathaniel Bechhofer retweeted
We released DuckDB 2.0 on Wednesday, leading with the Python, Java and command-line clients, plus the core extensions.
By Friday, the rest of the ecosystem caught up with the new version. All primary clients support DuckDB 2.0, including Go, Node.js, R and Rust.
Most community extensions have also been updated and rebuilt for DuckDB 2.0, and there are even new ones out there
For example, @MistralAI released their new Timestral time-series model family, and there’s already a DuckDB extension for it, contributed by Sylvain Utard! Timestral forecasts that if you’re based in Europe, you’ll have an extra hour this weekend – perfect for trying it out using our new compact syntax:
INSTALL AND LOAD timestral FROM community;
Nathaniel Bechhofer retweeted
Turns out you can use @ethanarbuckIe's research to modify sinf data for running iOS apps on Mac again! Didn't know when I saved App Store IPAs from 2021 that they'd be useful today 😆
Thanks to @roothideDev for FairFree.
@Little_34306 @khanhduytran0
Nathaniel Bechhofer retweeted
theres a hidden fully functional ipod as a 404 page on apple music music.apple.com/us/jfdsalkfa…
Nathaniel Bechhofer retweeted
There is an (unpatchable?) supply-chain zero-day in Intel CPUs that allows a process to run *any* user-level instruction. Running a program on your computer can cause arbitrary code execution.
Funny historical timing that AI coding agents came along when PHP had declined enough such that it's quite rare for fully vibe coded websites to use PHP
Nathaniel Bechhofer retweeted
🦆 Announcing DuckDB 2.0.0 LTS “Cinnamon Teal”
We are proud to release DuckDB 2.0.0, named after the cinnamon teal (Spatula cyanoptera), a striking reddish-brown duck found in the western Americas. 🌎
DuckDB 2.0.0 is the biggest DuckDB feature release so far. It ships a polished version of the Quack protocol, which turns DuckDB into a full-fledged server. It improves performance and lowers costs through supporting asynchronous I/O, a first-class VARIANT type, and a dedicated “agent mode”. It also empowers extension developers with a new SQL parser that allows adding new statements, clauses, and keywords, a new API to build on, and a new mechanism to distribute extensions more easily.
In our announcement blog post, we summarize the highlights and then walk through the key changes and give guidance on how to migrate to 2.0. For the complete release notes, see the release page on GitHub.
duckdb.org/2026/10/21/announ…
#DuckDB
