Ch(e)at GPT? - Computerphile
recommended viewing if you wonder how anthropic is embedding watermarks in a model's text output (possibly, maybe, probably something like this, maybe not)
Armin's handler at https://t.co/B05ybKGkzx. Old man yelling at Claudes. https://t.co/Q1wG57v1yc https://t.co/mnOoWUr0TO https://t.co/8i5vIRE0Wn
Unofficial aggregated profile. This creator hasn’t claimed their profile yet.
recommended viewing if you wonder how anthropic is embedding watermarks in a model's text output (possibly, maybe, probably something like this, maybe not)
hey @dillon_mulroy they are stealing code mode https://patentsgazette.uspto.gov/week26/OG/html/1547-5/US12670045-20260630.html
also fable at 50% context window
this is such a confused company. breathtaking.
ft.com
recommended viewing. warmly so!
new next token! we talk about mental health and burnout around ai and not being a human vessel for claude - tune in and let us know what you think!
feeling the AGI at 50% context window.
"make your phone as dumb as possible" is the best advice in 2026.
you will regret remote coding. stop it.
someone on the team vibe slopped "Figma but for TUI" and this timeline is killing me.
especially nice when working with local models.
The Cache hit ratio with @pidotdev for Deepseek v4 flash is insane!! In other harnesses as well I typically see 94-97% but in Pi it reaches 99.x%+ all the time!

ngl tired of starring at markdown.
well, that didn't take long. that's a no. but there's already an extension if you need this.
based on the error messages, we are about to get a new GPT model.
had the closed lab sota models come up with a durable agent harness design on their own for fun. multiple tries. they were all laughably bad. don't @ me.
recommended reading.
blog.senko.net
recommended reading. similar observations on my end, albeit at much smaller scale.
Article
AI Adoption is a Myth
i did it. i made sol go in thinking cirlcles forever. and all it took was 1 derranged design doc.
why did he have to become a freaking nerd.
well, hello there. welcome to the club. cookies over there. don't file any bugs.
Today we’re launching MiniMax Code 2.0. Rebuilt on the open-source @pidotdev Pi Agent framework, MiniMax Code 2.0 delivers a smoother, more reliable experience across everyday conversations, office work, and long-running complex tasks. Here’s what’s new

the benefit of having a "transparent" harness is that i can see how context handoff to subagents goes entirely wrong. the subagent has no idea about the proposed manifest changes. sol didn't give it that info (it's not in any of the referenced files, which the orchestrator read)
ah, the great filter.
theguardian.com