DeborahEmeni_

I’ve been experimenting with this recently. Running services inside microVMs instead of plain containers makes the threat model easier to reason about, especially for multi-tenant or untrusted workloads. I’ve been trying it out on Northflank and the trade-offs become pretty obvious.

makes sense tbh. i’ve seen teams take that approach and only think about cost once it’s already a problem. ever had that come back to cause issues later?

yeah exactly. feels like the pricing model alone keeps a lot of folks from even starting. one wrong config and you’re done before you ship anything.

yeah that’s the part that gets me too. feels wild that there’s no built-in way to cap spend at the account level. even a “pause all” button would go a long way

agreed. old reddit actually works. no extra stuff and doesn’t get in the way

yep that’s the one. rough situation but the replies were interesting

yeah cost anomaly detector is underrated. do you usually set custom thresholds or just let it use the defaults?

yep this is the one. feels like a lot of people just copy-paste from stackoverflow without realizing they're hardcoding keys. have you seen this cause issues in prod or mostly in hobby stuff?

yeah makes sense. the idea that success = risk now is kind of wild. did you end up doing anything to limit that egress, or just ride it out?

appreciate the context. do you know if the platform itself didn’t meet dev needs, or was it more about rollout and team buy-in?

thanks for sharing this. really interesting seeing the jupyterhub setup combined with devcontainers. curious how usage went long term. did people keep using it without needing much hand-holding, or was it more of a “power users only” kind of thing?

also appreciate the note on the sister team. i’ve seen the same. lots of energy up front but adoption fizzles when teams don’t prioritize the boring stuff like docs and onboarding

got it, that makes sense. i was trying to keep things centralized at first but splitting per cluster might be the better call. did you end up automating anything around managing those separate instances or mostly manual?

The “holding a button” thing actually resonated. It feels like the real work here is engineering the reward structure to make exploration even remotely viable. Dreamer’s world model might be cool, but most of the heavy lifting still seems to come from how forgiving the Minecraft environment is for training.

I do wonder though: if you swapped Minecraft for a cloud-based synthetic world with similar physics but messier signals, like object permanence or social reasoning, would Dreamer still hold up? Or is it just really good at the kind of clean reward hierarchies that games offer?

I’ve done something similar using OpenRouter and fallback chains across providers. It’s super helpful when you’re hitting rate limits or need different models for different payload sizes. I would love to see more people share latency data, though, especially when chaining Gemini + OpenAI like this.

Really cool setup! Curious how much of the performance here could vary depending on whether the model runs in a hosted environment vs local. Would love to see benchmarks that also track how cloud-based eval platforms (with potential rate limits, context resets, or system messages) might affect things like memory or secret-keeping over multiple rounds.

Apple's shift away from GPL has been showing for years, so this doesn't surprise me. But dropping rsync 2.6.9 this late without full 3.x support feels like a step backwards, especially for anyone relying on metadata preservation. I get the licensing angle, but the lack of feature parity in opensync might catch a lot of people off guard.