Author here. The search algorithm was the easy part. The LLM already encodes domain knowledge from ML papers; it knows learning rate warmup helps with transformers, that batch size and learning rate are coupled. It converged on the winning GRPO config by iteration 1. Grid search needed 8 iterations.
The hard part was per-iteration GPU isolation. A botched run that leaves stale optimizer state or corrupted weights in memory will poison the next iteration. Each iteration needs a fresh CUDA runtime, fresh filesystem, fresh memory. No state leaks. That's where most of the engineering went; ephemeral containers with TTL-based cleanup, one A100 per iteration, torn down after metrics are emitted.
How long does it take to have my domain whitelisted? and should I be able to send emails under one domain? Do you support react-native? Can I send it from the backend (eg next.js server-side backend)? Is it for only transactional emails?
> I suspect a Web3 startup will be the next thing VCs will hype up.
You mean, the whole NFT hype? Just happened. What's next? I'm betting on some VR/AR/XR thing (like pokemon Go, but different, same-same, but different).
that's fine. Maybe then for a better user experience have grey-out or removed the IPFS download button for the articles that haven't been uploaded yet.
TL;DR; My personal advice is to keep talking to your loved ones with an open-minded dialogue.
I'm on a similar situation myself too; in my case my loved ones are well educated (physicists and biologists), and aged (over 60s & 70s).
You need to change your perspective and realise that for them, you are the person that believes on "fake news". During my talks with them, I've tried lots of different approaches to demystify some of their arguments; I made my own plots based on open data for COVID-19, I used well-sourced articles from newspapers or even published white papers, until I realised that the matter is based on opinionated believes rather than a lack of fact-checked information.
Now, I'm treating these conversations similarly to having a political discussion; there are some parts that we both disagree and this is fine, but we also have some grey areas that we can discuss further about. The fact that I'm also "challenging" myself with alternative questions and perspectives, I think it makes me better to do a better data-search for the next time.
Having said that, what I do respect the most on them is the fact that they try to seek their own personal opinion. But "popularised science" in ongoing matters is hard to find, thus the only easy-to-digest content is through groups and videos that carry-on with promoting misinformation; news that are not fact-check but only based on opinions (usually political driven).
So as you mentioned, the main problem is the lack of fact-checking popularised sources & channels. One example that does exist is "TL;DR; News" [1].
I will only partially agree with you; mainly because indeed, that's what they're pitching crypto as - To be your own bank. But practically, most blockchains are replacing the inter-banking networks (like SWIFT) rather than the banks themselves, as you still need centralised entrypoints.
To save some comments; Yes, they do have the potentials to create their own closed-circle economic ecosystems, but for the retail banking services to work (eg overdrafts, mortgages etc) you will end-up with centralised players as these kind of services need to be backed by wealth (it's not a tech-issue).
Personally I find it interesting as tools like these have been helping me to understand the team's and delivery dynamics when I'm joining new dev-teams.
This particular is great, as it reminds me of a book that I had read a while back; Your Code as a Crime Scene [1] by Adam Tornhill.
Adam is trying to explain something similar, but takes the whole concept onto the next level by explaining how tech debt and hidden coupling can be discovered using the git history and similar file structure visualisations.
I use chef and test-kitchen to bootstrap my dev-VM (vagrant is used by test kitchen). So I have written my cookbooks, and depending on the project, I converge (spinning-up) VMs depending only on cookbooks that I need, eg no java will be installed if the project requires node.js only. In addition the main gain with that, is my dev VMs are totally disposable and they own nothing. Everything is being synced into the VM from my host machine, same with data.
Lately I am trying to "dockerize" my backends, so in the case my workspace project needs a mongoDB or an other backend from my architecture, I should be pulling those containers up on converging. That will make my life easier when writting cookbooks for the backend dependencies.
I believe you can achieve the same using ansible, chef was a personal taste.