ive seen that a lot in recent gpts and bonsai/qwen models when they invoke their vision system/modality , or when they ask their harness to do so for them.
It happens. There use to be a joke during the first big DC build out phase that went like if you're ever going into the wilderness take a 1ft length of fiber optic cable with you. If you get lost bury it and a back hoe operator will appear and sever it within an hour. You can get a ride back with them.
>A man stranded in the bush in northern Saskatchewan was rescued last week after chopping down four power poles — knocking out electricity to surrounding communities. [...]
>But he had an axe and he knew SaskPower would have to check the downed line, so he went to work.
I remember a story from an acquaintance working in construction. In one case they had to dug up right where a fiber optic ran, so they had cleared it with its operator and waited for their go-ahead. And waited. And waited, until the boss said something to the effect of 'start digging, we have a schedule to keep'...
The token price isn't the only reason to run a model locally though. You can do additional training to specialize or remove censorship that may be a no-no per TOS with cloud GPUs.
This issue really separates the wheat from the chaff. The dirty secret is that a lot of academic published work contains errors but the authors also have fragile egos. It reminds me of when Data Colada exposed someone for fraud and they accused the exposers of "methodological terrorism" (something in psychology or social science). Just imagine what happens once papers get exposed at scale.
Now here there was no fraud just genuine error, but it will annoy people nonetheless and scrape their ego that someone uninitiated can just type some stuff in a magic box and conclude that they, the established published, tenured mathematician with awards and medals can be wrong.
There's going to be all kinds of weird stuff like this surfacing everywhere until after the mid-term elections in November. Youtube is full of it right now, it's just going to get more ridiculous and crazy for the next several week.
My grandfather was like this, grew up in the depression and fought in the pacific during ww2. His garage was full of coffe can after coffee can filled with random nuts and bolts. When I was a kid every meal with my grandparents was 50% leftovers and they had a giant garden. Waste not want not indeed.
> Maybe it's faster but explosive diarrhea is also a faster way to produce shit.
i'm in the midst of a very large global roll out of a new system including about 24 separate erp integrations. My team's system is dependent on another system coming online first then we go then the integrations come online then a massive data migration happens and ends with all the associated testing. So like a huge number of dependencies and I'm second in line. Today we get word that the first system is having issues with a vendor and will not come online as scheduled (our dry run deployment is tomorrow and spans all next week as each dependency gets their turn. Week after next is PROD.
We're not heavily integrated into that first system but we are in spots and so there's a reason why they go first. Writing around that system is feasible but we're looking around 5k lines which isn't a lot but there's still testing and sign off. My offshore team is in bed, meanwhile they, and onshore, all have other shit to do anyway so pulling them just puts everything else behind.
I tasked an agents to make the changes as soon as i woke up to the news, they got it mostly right. It took two rounds of fixes with some offshore testers who stayed up late but the PR got in and merged. We just did (literally like 5min ago) a validation run against our target system and we're green for tomorrow. I told the testers to goto bed and then i pulled up HN and opened a coke.
There's no way in hell that would have happened without the coding agents we had on tap. So, yeah, i find a lot of value in the rate at which AI can work on code even if it's not perfect every time.
True, it would not have happened back in the day. But oftentimes what would happen is that sane, sensible leadership would have a talk about delays on the critical path and adjust the schedule accordingly.
Maybe not as satisfying to people who enjoy being heroes. But it would not have been an automatic disaster.
> “…talk about delays on the critical path and adjust the schedule accordingly.”
And self-imposed deadlines are the worst—not saying that’s the case here. And when your leadership sets the goals, you do your best. But fast, good, cheap; pick two.
And if the deadlines really are hard, then you should be actively managing the critical path from the get-go. The very existence of upstream critical path dependencies that, as the parent poster’s story demonstrates, weren’t even necessary in the first place. Which suggests that this heroic agentic quick fix was the proverbial pound of cure.
I’d even go so far as to speculate that perhaps agentic coding also steered people away from the ounce of prevention in the first place. Humans aren’t perfect either, but over the past couple years I’ve had to spend increasingly much time asking, “Why again are we creating all this unnecessary complexity?”
This is partially driven by the good old enterprise game of not being at fault when things go wrong and being able to communicate that to the stakeholders
It's not fancy but it's necessary in a lot of enterprises
I don't want to deny that. I mean I am using it. And if works well it's great. If it doesn't though it's exhausting. And if it produces bad code, it can do that at a speed that is overwhelming.
Ah yes complex global rollout being forced to production on an arbitrary rapid deadline is exactly the time when more code of unknown quality should be added to the system at the last minute.
reply