- edflsafoiewqLink: https://arxiv.org/abs/2512.24880
- Here's the relevant diff: https://github.com/python/cpython/pull/137968/files#diff-966...
Search is limited to 20 attributes and non-descriptors only to avoid arbitrary code execution.
I assume constructing AttributeErrors isn't highly performance sensitive.
- Can definitely think of some places I should use bytearray.take_bytes.
- Can screen readers emit their narration as text instead of / in addition to audio?
- The error from Rust's File::create basically only contains the errno result. So it's eg. "permission denied" vs "failed to create file: permission denied".
- Python's
is even more succinct, and the exception thrown on failure will not only contain the reason, but the filename and the whole backtrace to the line where the error occurred.f = open('foo.txt', 'w') - Marginal gains over AVIF.
(Also I am highly skeptical of the importance of these generation loss tests.)
- That's the whole internet now. That or Anubis.
- *distant sound of the sorcerer's apprentice is heard*
- Mutable releases are used for continuous/nightly builds.
- You recoup the saving of home automation immediately as additional leisure time. But for most people, work automation neither reduces your working time nor increases your wage.
- You're unlikely to get any radio signal that isn't specifically meant for you.
- I use an MVNO and I never needed any ID.
- Typically you'd also be in debt to the BNPL provider, not the merchant.
- How?
- > CAISI chose GPT-5-mini as a comparator for V3.1 because it is in a similar performance class, allowing for a more meaningful comparison of end-to-end expenses.
- The NIST report doesn't engage with training costs, or even token costs. It's concerned with the cost the end user pays to complete a task. Actually their discussion of cost is interesting enough I'll quote it in full.
> Users care both about model performance and the expense of using models. There are multiple different types of costs and prices involved in model creation and usage:
> • Training cost: the amount spent by an AI company on compute, labor, and other inputs to create a new model.
> • Inference serving cost: the amount spent by an AI company on datacenters and compute to make a model available to end users.
> • Token price: the amount paid by end users on a per-token basis.
> • End-to-end expense for end users: the amount paid by end users to use a model to complete a task.
> End users are ultimately most affected by the last of these: end-to-end expenses. End-to-end expenses are more relevant than token prices because the number of tokens required to complete a task varies by model. For example, model A might charge half as much per token as model B does but use four times the number of tokens to complete an important piece of work, thus ending up twice as expensive end-to-end.
- They compare with gpt-oss.
- > What decision-making process led to the idea of injecting human urine into a frog in the first place?
Hormones are basically messages sent through an animal's body to signal some change should take place. It was discovered that there was a hormone called hCG produced by the human placenta that triggers "you're pregnant" changes in the body. hCG is also present in the urine.
So if you want to detect a hormone, the idea is you inject it into an animal and see if it triggers the relevant changes (since the changes are usually internal, you generally need to kill the animal to check). So you would look for an animal that responds somehow to the hCG hormone, inject urine into it, and check for the response. Mice and rabbits were first used, but it was eventually discovered that certain species of frog that are highly sensitive to hormonal changes made for much simpler and faster testing.
- My takeaway from this story has always been that both MS and Plan 9 simply passively implemented Unicode as received. It was only IBM that had the vision to see that the encoding was wrong and they should make a new one.