What stable mechanisms are in place today that capture “Comprehension Assets”? It’s beneficial when one team member recalls a load bearing assumption, but that person may also be on vacation.

“The ability to look at a diff and immediately know which behaviors are load-bearing. To remember why that architectural decision got made under pressure eight months ago.”
https://www.oreilly.com/radar/comprehension-debt-the-hidden-cost-of-ai-generated-code/

I have been using email for 40 years. It used to work.

As an (independent) academic researcher, I need to contact new people, primarily in universities, to ask questions.

I refuse to use Google, Microsoft or the other American IT giants.

But they are increasingly preventing refuseniks from sending email at all.

I know what RFC, DNS, MX, SPF and DMARC mean. My email goes through small British companies with intelligent, friendly and helpful staff.

mxtoolbox.com says that I must have DMARC to send email to M$. So I set it up. I now get a dozen copies of the same report from G or M$ for each email that I send out.

They show that my email gets to G and M$ sites, but then it is marked as spam.

The stupid senior management of numerous universities has surrendered their staff email to M$.

Web searches and AIs preach about spam. I don't send spam - I want to contact my colleagues.

Rumour has it that previously unknown senders are treated with suspicion and their emails are sent to spam. In other words, it is impossible to **initiate** communication with someone.

Let's be blunt about this. They are a mafia that is enforcing an **oligopoly**. It's got nothing to do with reducing spam --- I have no doubt that they let through emails from "trusted partners", ie companies that bribe them enough to send their spam.

The result of this is that it will only be possible to send emails by paying M$ to do it, and then it will only be allowed to express "approved" opinions.

What can we do about this?

At the very least, those of you with senior positions in universities can tell your management to revert to competent standards-based email systems hosted on Linux systems.

"We taught models to compress their own chain-of-thought mid-generation. Peak KV cache drops 2–3x, throughput nearly doubles, and the erased reasoning blocks leave traces in the KV cache that the model still uses."

https://www.microsoft.com/en-us/research/articles/memento-teaching-llms-to-manage-their-own-context/ via Microsoft Research

this question is coming up a lot so i'll pin it [again after migration to treehouse]

• replacement for github: https://codeberg.org/
• replacement for github pages: https://grebedoc.dev/
• easy ci runners for codeberg: https://codeberg.org/whitequark/nixos-forgejo-actions-runner