James Wilson (@scotchandcode) header
James Wilson (@scotchandcode) profile picture

Reformed CTO turned podcaster at https://teapawt.net/t.co/ZkzEkkUTlJ

Sydney, Australia
Joined November 2023
Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)Photo by James Wilson (@scotchandcode)
But overall, what we've got here are autonomous context graph generation machines. Even if you don't let ARTEX loose to hack yourself first, at least use it to thoroughly if not continuous be scouting and threat hunting.
2
949
Scope it well at the outset and it stays in its lane. Forget to do this and you're in for some explaining. Hopefully not to a judge.
1
3
1,088
But, the simple "always be working on something" autonomy implementation quickly sees it go off in the weeds... or into WorkOS as I observed!
1
3
1,217
ARTEX itself is quite simple, almost naive. If what you've got and what you want at the goal aren't that far apart in problem space, then it'll do just fine.
1
5
1,240
I've been using ARTEX, and recorded a deep dive solo podcast about it, CAIRN, and AntiFox. There's a lot to learn about harness design and turning offensive tools into defensive machinery. Hour long. Link below!
8
21
233
17,075
I’ve been loving Dot, but over the last 24 hours it’s become useless due to dropped / failed calls. What gives @thsottiaux
1
234
Replying to @IceSolst
Yessssssss
2
344
🚨 Heads up‼️ Multiple crypto wallet projects are being targeted in a coordinated software supply chain attack. The threat actor is targeting @Zcash, @ZcashFoundation, @vizorwallet, @cosmoswallet, and @keplrwallet using a novel Rust dependency confusion attack, multiple malicious GitHub pull requests, and a phishing campaign that has targeted at least two OSS contributors. opensourcemalware.com/blog/c…
2
244
I’m using it to ensure demarcation points in my security harness are being used appropriately. For example on creation of a “memory” Jev is used to ensure that memory has real lasting purpose and value and isn’t just a stashed status update (which models love to do when left unchecked). It’s a useful nudge to get a larger model back on track when they stray.
1
2
81
Replying to @IceSolst
Close… it’s actually how many pieces of unsolicited feedback was the dev forced to deal with before shipping. How do I know? I wrote public APIs at Apple. That’s pain. Quality through pain.
1
1
70
Cooking up some pizzas and was joined by one of the Mourning Doves that likes to call my back yard home this time of year.
3
583
Three days and 350+ agents on Sol 6 Ultra (Day Break Blue) and I think I’m close to being able to validate some of the agent persistence ideas I had in the last podcast. Or, I will have inherited a branch that can only be rm -rf’d. Soon find out!
2
806
Thanks @IceSolst for the follower bump! Admittedly I haven’t posted much here yet, but that will change!
1
8
670
Just in: Binance rebranding to Minance after losing billions. Squatting domain Kinance just in case holding drop further.
1
448
Uh that's weird. My table number is exactly the same as my SSH private key.
413
Because ultimately, I want my agents to hack like their do ... but without the cheating!
4
And so I decided to look into what it might take to understand and engineer the relentless determination we see in a frontier lab's agents when faced with maximising a benchmark score.
1
5
It made me realise that my current use of AI is as a force-multiplier of myself. Problem is, when my input drops, so does the LLM output.
1
4
This began with my own curiosity as to what it is about these LLM benchmarks that drives agents to achieve the results in sometimes very unexpected and even novel ways... meanwhile I'm stuck babysitting a bunch of Ralph Wiggum's that constantly need nudging.
1
15