i have tried this workflow and it is solid. It is a codex that commits once it finishes something. You can pipeline changes, so it works like in 5-10min intervals and it gets mostly right, much better (and much slower) than opus. He has two computers and one for longer running tasks and another for short one. I suppose you just pipeline a bunch of small issues to the long term one and ask it to work and work on the repo. Another one is probably where he is more engaged with specific tasks. Impressive that it works quite good.

What if we will go even further? I have built end-to-end messaging layer for Clawdbot to talk to each other, called Murmur - https://github.com/slopus/murmur.

We tried this with friends and it is truly magical (while crazy insecure) - i can ask my agent to search friends life, their preferences, about their calendars, what films they are watching. It can look at emails and find if you need something and go to people around asking for help. It is truly magical. Very very curious where it can go. At the moment it is exceptionally easy to exfiltrate anything, but you still can control via proper prompts - what you want to share and what you dont want to. I bet models will became better and eventually it wont be a problem.

> ... what films they are watching.

I assume this relates to looking at their desktop activity? Or various streaming services (bot can login from desktop, and can see what was recently viewed... Even if viewed on smart tv etc?

> At the moment it is exceptionally easy to exfiltrate anything, but you still can control via proper prompts - what you want to share and what you dont want to.

You can? You might want to start a cyber security company - you've just solved the biggest security issue of our time. That, or you're overly optimistic?

Sounds futuristic but then, for this particular example, isn't this just a very hard way to get access to their calendar, which their could share w/ you in 10 seconds on their Gmail tab?

The problem that you expose everything, why i need to expose my health things to coworkers? or surprise party planning? with prompts you can limit exposure easily.

I am curious when we will land dead github theory? I am looking at growing of self hosted projects and it seems many of them are simply AI slop now or slowly moving there.

Humans dont need a skill to know that they need a skill

thanks! thats a donation button, thank you for catching the error in docs search, will take a look

Thanks for report, thankfully it is needed only once!

The most interesting link is broken( does anyone knows how to find source for this link: https://www.isg.rhul.ac.uk/~sdg/igor-slides.pdf?

Actually some Outline feature aren't, which could be quite limiting when deployed in a large organization or institution

It's the BBC. 'boss' as a mafia trope is specific to the US.

Yup. Similar to scheme, which doesn't carry the same negative connotation in British English as it does here.

I believe, the noun "boss", when meant as "a person who exercises control or authority", does not imply legal status or purpose of any group or organization.

https://www.merriam-webster.com/dictionary/boss gives an example of "union bosses".

thats what I keep telling the guy at the corner store

I am wondering why flash attention is like 5x slower with variable masking than without it? Lack of good masking support almost zeros out the optimizations

Where are you seeing these benchmarks?