OOohhh AI coding.
An integration test for a product failed because I added a new product.
Turns out the integration test was reading the documentation (??) which is shared between products and the inserted product gave it an off-by-one.
I have soooo many questions.
OPPO MVP Ruler, brought to you by the reliable Cisco network, is gonna snipe the UberEats Herald (MasterCard Priceless Moment??) making way for the Red Bull Baron to spawn.
Should companies publish their threat model?
It feels like open source vs. closed to me--security through obscurity isn't really security.
That said, what do you gain or lose by publishing in the people's eye? Maybe tech people trust you more? But it might scare non-tech peeps?
@rearden_primus I'm less worried about them once setting up a lot of tests. They can fool around all they want so long as I have sufficient test coverage.
The hard part now is on me to assure sufficient coverage.
Just a few days ago I mentioned how auth was by far the hardest part of all the work I've been doing.
Humble Bundle just vibe coded their auth and ruined their 2FA.
I feel so justified.
Also, turns out there *IS* skill in properly steering the AI. Senior engineers, rejoice.
I'm working on a product and building the components separately.
So far, auth is the hardest part by a long shot.
It's been maybe 3x harder than the product itself?
Man, I know GitHub has had a massive usage spike after AI coding became a thing. I'm sympathetic.
And I also get that people have been blindly hating on GitHub, which isn't fair.
But it feels like every day I have an issue.
@richgel999 I still get irked with every C++ project of mine bundling a CRT where 99% is unused.
Plus the parts that I do use having secret implementation details and being unable to separate...
So then I'm tempted to write my own every project. :/
@_Buttonizer Right, this wasn't that.
It was a lot of me saying "But if we do that this breaks, right? Am I missing something?"
I deliberately try to get the AI to challenge me or fill my knowledge gaps, rather than pad my ego.
Maybe I encouraged it to challenge me *too much*.
FRIGGIN' AI...
<AI> "We shouldn't do it your way. My way is better because ___."
<me> "I just don't believe that. Can you check what Amazon, X, Facebook, and Google do?"
<AI> "They all do it your way. You're right."
*updates AGENTS.md to tell it to research before challenging*
I should add some context because that maybe makes me sound like an arrogant ass.
I'm happy to be challenged and get pushback.
But I just spent an hour going back and forth.
All of that could have been avoided if the AI checked itself as much as I checked myself.
Back in August of last year, I won @NVIDIAGeForce's raffle to then buy a 5090.
I tried to buy it but the purchasing system failed.
Reaaaaally wish I was able to buy it back then :'(
Previously, I did not understand why putting smart home devices on a spearate VLAN helped. If a hacker gets into that device, the only pivot they can make is the other already-vulnerable things.
I get it now though. So many things trust internal networks. Read: are vulnerable
An Anthropic Pro account is $20/month and gets you SOOO much.
But that isn't what's going on here. Once you max out your Pro for the week, put $100 of credits on there and see how quickly you burn through it.
It's surprisingly easy.
NEW: Apple is cutting employee AI spending budgets even as it pushes workers to use the technology more. Engineers have seen daily limits cut from $300 to $200, with further reductions to $100 planned. Other teams could see limits fall to $25. https://t.co/4tnOIaxpDz
@GamerGoldyTV This.
Same with Issue 1 a few years back.
I blame Frank LaRose who approved wording that was clearly intentionally misleading. Not just "oops, I could see how..."
People think they voted for X but they actually voted for Y.