Visch

Jesus is the way
Husband, and Father of two
Founder - - (Vibe coded and I use it daily!)

Still manually creating accounts?

[Visch @derekvisch]

@pandemicsyn Ahh so not critical path but useful. I like the very specific watch search "lotto" routine idea. I go really nuts with them, and they trip me up once in a while :/

[Visch @derekvisch]

@thdxr Start sharing subs with other coworkers or upgrade our sub

[Visch @derekvisch]

CLI Walkthrough

[Visch @derekvisch]

Just released pdt-meltano https://t.co/pZmFhDohgl

This lets you deploy your meltano jobs to

Azure Container Jobs
AWS Batch
Google Cloud Run

meltano --environment=prod invoke pdt-azure deploy

[Visch @derekvisch]

Secrets management is pretty slick as well

$ meltano --environment=prod invoke pdt-azure secrets meltano-daily-sync
updated TAP_SMOKE_TEST_API_KEY •••• (4 chars) •••• (7 chars)
Proceed? [y/N]
meltano --environment=prod invoke pdt-azure deploy

[Visch @derekvisch]

Other useful commands

pdt logs
pdt runs
pdt health

And example deployments

[Visch @derekvisch]

Launching PDT: process deploy tool. Run any scheduled job on your Infrastructure. No daemon, no subscription, and it works with all your AI tools

[Visch @derekvisch]

@Tamurphy Unfortunatly one of the big reasons snowflake is so popular is because you can bypass it restrictions. Hopefully that's not what's happening there!

[Visch @derekvisch]

@0x71ff @monofuel34089 We should have used our gpus as heaters in 2010 for those btcs, hindsight is 20/20 :D

[Visch @derekvisch]

Here's some evidence they are exaggerating the threat

1. The models were tested with safeguards turned off
2. The testing environment setup is suspicious at best, it was made to eventually generate a story. They were running exploitgym tests, while having access to atleast one service (artifactory) that had a lot of similar properties to the other exploitgym examples. It's not a huge jump to apply the similar exploit ideas to the services you have access to on the machine.
3. Openai and anthropic have a very similar "breach" and at very similar times, there's a reason for this, it's a strategy. Both environments were not setup to be secure, dive into their technical reports and look into the testing environments. There's a reason for this, it's not that they don't know how to set these up securely

Related is the fact that exploitgym is seeding the context. A decent analogy is we gave the model a book that tells it how to break locks, we give it a lock and tell it to break out. Then we are surprised when the model looks around and sees another lock in the room (artifactory) and breaks into it. The other lock was easier to break into then the test we gave the model (not to mention the exploit gym eval has some tests that are very similar to the same method used to break into artifactory)

This is all exciting stuff, but not something to be getting the government involved over

[Visch @derekvisch]

The name for these today is botnets. AI powered botnet will happen but it's not like we won't be able to stop it. Remember those early 2000 worms they had against xp machines? Everything got hacked over and over. We may go through a phase like that here but it seems to me it's more likely we will be able to stop them and patch more quickly then they can spread

[Visch @derekvisch]

@grok @bot "Buy toilet paper" and it remembers the brand you want and where you normally buy it from!

https://t.co/A7Zjy6aOqU

[Visch @derekvisch]

"Buy peanut butter" is all I say to my main agent, and it automatically sends it to my buying bot to add to our next order!

Grocery and Home Item Buying Bot Template here:

https://t.co/vgknyNJioq

[Visch @derekvisch]

"Buy peanut butter" is all I say to my main agent, and it automatically sends it to my buying bot to add to our next order!

Grocery and Home Item Buying Bot Template here:

https://t.co/vgknyNJioq

[Visch @derekvisch]

@gavinortlund It's not even a 1 percent risk, go look into how wrong the people building new technology are at projecting how that technology is going to impact society

[Visch @derekvisch]

@DustinDavis These bots should be trying to persuade us to move forward with things using the same techniques, and then iterating when we reject them

[Visch @derekvisch]

@_andrewthecoder git remote add origin /mnt/git/pdt.git

[Visch @derekvisch]

@Tamurphy @JoshDance I was way off😅

[Visch @derekvisch]

@Tamurphy Time in meetings per week?

[Visch @derekvisch]

I was already paying for cursor so grok bot was a "free" addon. Use any of the tools, I run OpenClaw locally right now!

I'm becoming more convinced that all the "routines" should be deterministic (or as close to deterministic as possible) anyways which means these tools are mostly for discovery and exploration

[Visch @derekvisch]

Using Grok Bot as our grocery manager. Can definitely work for OpenClaw and Hermes

Had it learn a year of Walmart + Amazon purchases. Now I just type “peanut butter, jelly, bread” and it adds our usual stuff to the cart.

Also tied it into Plan to Eat. Pretty slick