2026 needs a new CSV format that supports multiple sheets and that doesn't require escaping. ASCII US (Unit Separator) for comma ASCII RS (Record Separator) for newline ASCII GS (Group Separator) between worksheets *maybe* ASCII FS (File Separator) for embedded images Three header rows: 1. name of worksheet 2. Names (like usual) 3. Types (string, int, date, etc) Editors should treat Unit Separator like tab, except that it should be calculated to the data-width and data can wrap within the column based on maximum display width or column width. RS should obviously represent rows, and GS for tabs. Binary data must be hex encoded (because it compresses better than base64). Any data containing those control characters or other non-printing, non-UTF characters is considered binary. Metadata can be stored in a group before the worksheet begins. Lines starting with `#` and followed by `.` and no other separator until GS can be used to pad groups to a rounded standard size to avoid full-file rewrites on simple metadata or single-value update. No formulas or user-oriented spreadsheet stuff. This is just a data format for APIs and AIs.

What kind of favors has Anthropic been pulling to finally get the regulation they've so desperately needed? And what will this mean for local Ai? Will China retaliate and we no longer get access to the models that are 10 to 100 times more efficient? No more competition? Instant monopoly?

I have no other word for this than simply "good". m.youtube.com/watch?v=Ts5Z64…

One of the reasons that I'm much happier with Pi Qwen is that when it does really stupid things like "Actually, it's in .gitignore, that's okay, I'm not supposed to commit this file. Actually, I'll just use `git add -f IGNOREME.md` ..." at least I'm not paying for it.

I must say the stochastic nature of working with agents by far the worst. Yesterday, I could describe my problem and it would implement it near minimal changes. Today I describe my problem and it rampage throughout my whole codebase...

Summed up pretty well by @ThePrimeagen: See timestamp 14:02. youtu.be/pkndFYSTr0Y?t=842

Is there an extension I can get to show the branch name in the PR list on Github?

Exactly. Either there's always a way for an attacker to probe an email anyway, or all of your infrequent users have to play which-email-is-it bingo.

Despite being a local-first guy, I honestly - all bias aside - did not expect this: I prefer Pi Qwen 3.6 a3b over Claude Opus 4.6. I've retooled my M2 Studio 64gb such that it's only job is serving llama.cpp via port-forward and I use it from everywhere - including on my most recent flight. This is good.

An important different take on the situation. (that he was saying "learn it so you don't repeat my mistakes") youtube.com/watch?v=tlQ7EoJD…

They're having the customer support Ais have foreign accents and say "uh" and "um" and draw out certain letters to sound more real. It's wild. I'm not 100% sure that this is a bot... but I'm like 95% sure. I want to ask "how many Rs are in strawberry?" or "can you write a python program that counts to 10?" but... what if it's a person. Maybe I could start with "how's the weather out there?" Ladies and Gents we are deep in uncanny valley.

We *can* do easy things. We really, really can.

And through a little binary search bingo I found that the max context size for llama.cpp w/ Qwen 3.6 35b on 32gb M1 MBP is 102500. It may be a few tokens more, but it can’t go to 105000. Over the limit and it comes back with "compute error". At the limit auto-compaction takes hold. Moral of the story: That's usable for agentic coding.