AI Development general thread.

What my codex has been burning tokens on for the last 15 mins...

"The defect is cross-machine clock skew: SQL Server stamps the committed media row about 155 ms ahead of the application host, while variant freshness was stamped from the application clock. I’m fixing commit to reload the authoritative persisted media timestamp before generating variants, preserving the stale-variant security check across separate hosts."

🤦‍♂️
 
Been opening so many PRs on my personal project have run out of included minutes for the Github runners. Thankfully have a Proxmox host and just spun up a few small VMs for self-hosted runners. Not only doesn't cost me anything but also faster, those GH runners are painfully slow.
 
L-Dog said:
Anyone know a website where I can make a free AI video ?
Click to expand...
Not much in the way of "free" video sites. I think Google Omni / Gemini... will let free users do a few videos but they're generally 10 seconds. If you're looking to do more involved complex stuff then you should look at specialised sites like Higgsfield or OpenArt, though their free tiers probably won't let you do much. If you really want free then you need to get to grips with running local models on a decent PC.
 
Well that's new. I tasked Codex with some data analysis. This mofo found my Ollama instance and now is using it. And when it wasn't as fast as it expected it changed it to a smaller model. Wtaf.
 
I was having a discussion with a viber yesterday who is fighting stability issues and I told them have the agent you using give a basic diagram of how its handling storage and we can discuss the handling and architecture. The immediate response was to pick up a point and then start trying to action it and I said, "No wait, hold on, lay it out first!".

I think a big problem is almost like gambling addiction, the endorphin of seeing "something" being done, without the consequence of what and how its doing it, that's a large problem.
 
Kosmik said:
I was having a discussion with a viber yesterday who is fighting stability issues and I told them have the agent you using give a basic diagram of how its handling storage and we can discuss the handling and architecture. The immediate response was to pick up a point and then start trying to action it and I said, "No wait, hold on, lay it out first!".

I think a big problem is almost like gambling addiction, the endorphin of seeing "something" being done, without the consequence of what and how its doing it, that's a large problem.
Click to expand...

I see it often in things, "its working, it does what it needs to do"... but HOW is it doing it, does it meet all sorts of basic logical requirements for a thing to be considered "working" in a production type world.
 
Kosmik said:
I was having a discussion with a viber yesterday who is fighting stability issues and I told them have the agent you using give a basic diagram of how its handling storage and we can discuss the handling and architecture. The immediate response was to pick up a point and then start trying to action it and I said, "No wait, hold on, lay it out first!".

I think a big problem is almost like gambling addiction, the endorphin of seeing "something" being done, without the consequence of what and how its doing it, that's a large problem.
Click to expand...
Another issue is being fully dependent on your IDE to do everything, including the thinking, for you. Something goes wrong in prod and you don't have access to your IDE for whatever reason - now what...
 
ToxicBunny said:
I see it often in things, "its working, it does what it needs to do"... but HOW is it doing it, does it meet all sorts of basic logical requirements for a thing to be considered "working" in a production type world.
Click to expand...
What kills me is the lack of interest in why. I mean, ask your agent to give a explanation overview to help understand, especially if you struggling but many don't do that.

And the current verbal diarrhea of some of the agent families is pretty bad. Conspiracy me leans towards the whole ai watermark implementation because its just excessive.
 
Kosmik said:
I was having a discussion with a viber yesterday who is fighting stability issues and I told them have the agent you using give a basic diagram of how its handling storage and we can discuss the handling and architecture. The immediate response was to pick up a point and then start trying to action it and I said, "No wait, hold on, lay it out first!".

I think a big problem is almost like gambling addiction, the endorphin of seeing "something" being done, without the consequence of what and how its doing it, that's a large problem.
Click to expand...
It's quite interesting with Codex specifically. As things have progressed, not necessarily model-based, if you start something new it can take on a whole extra layer of concerns and awareness of security and best practice whether you asked for it or not. It would seem that they're trying to curate a path for development requests thats somewhat hidden in the engine. I'm guessing at some stage "Oom Kosmik" will be a layer to bypass.
 
Mike Hoxbig said:
Another issue is being fully dependent on your IDE to do everything, including the thinking, for you. Something goes wrong in prod and you don't have access to your IDE for whatever reason - now what...
Click to expand...
hmmm, I would expect ide to be miles away from prod, that's the territory of logs, analytics, app insights etc. The inverse does kill me though, treating pseudo code in a non ide when the basics of the ide are there to ensure stupid things like syntax errors and other fallicies don't crop up prior to compiling.
 
quovadis said:
It's quite interesting with Codex specifically. As things have progressed, not necessarily model-based, if you start something new it can take on a whole extra layer of concerns and awareness of security and best practice whether you asked for it or not. It would seem that they're trying to curate a path for development requests thats somewhat hidden in the engine. I'm guessing at some stage "Oom Kosmik" will be a layer to bypass.
Click to expand...
lol 😅 :crying:
 
Kosmik said:
hmmm, I would expect ide to be miles away from prod, that's the territory of logs, analytics, app insights etc. The inverse does kill me though, treating pseudo code in a non ide when the basics of the ide are there to ensure stupid things like syntax errors and other fallicies don't crop up prior to compiling.
Click to expand...
It has context of what your app does, how it works, and therefore how to reproduce and fix it. They wouldn't have a clue if the thing was even hallucinating or not. I've seen some idiots give their agent direct SSH access because they are too lazy to learn some basic docker commands. That's how you end up with the PocketOS disaster...
 
Mike Hoxbig said:
It has context of what your app does, how it works, and therefore how to reproduce and fix it. They wouldn't have a clue if the thing was even hallucinating or not. I've seen some idiots give their agent direct SSH access because they are too lazy to learn some basic docker commands. That's how you end up with the PocketOS disaster...
Click to expand...
yeah, I mean a couple of times I caught mine going into az cli, which is alright as long as it asks permission and at least I KNOW what its trying to do but quite a few folks have it on full approval and trust the modesl to do the right thing.

"Oh you absolutely right, I shouldn't have done that!"
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X