HN user

madarco

139 karma

Maker, Developer, Founder - 3DRacers.com

Posts2
Comments77
View on HN

It had already been like this long before widespread LLM adoption: quality hiring was only really possible through manual candidate scouting on LinkedIn, at conferences, through word of mouth, and so on.

Sending a CV had already become mostly useless 3–4 years ago because of the huge amount of noise: candidates applying from all over the world, often even spoofing their actual location, FAANGs firing, flooding the market with (in theory) great candidates with great resumes.

The solution is the same one that has been successfully used forever: a trial period.

Luckily, a video interview with a senior developer is still enough to spot a good candidate.

Go through real code: add a bug to a branch of your codebase, have the candidate share their screen on TeamViewer, and let them debug and fix the issue. Ask questions live to understand how they reason about the system, how they would test whether the change works, and so on.

This will filter out 99% of candidates. But it is still possible to get lucky, which is why the trial period matters.

I’ve never had major issues and have always hired very strong engineers. I only had to terminate someone after the trial period twice.

I do the same, and faced the issue that claude/codex loose context when doing subtasks (and subagent don't have plan mode).

So I've built Agentbox to be able to launch from claude/codex multiple VMs with claude/codex (can also mix). The parent agent watch for prompts and questions, enforce /review, /simplify, that the sub agents file a PR and wait for bugbot comments etc.

This way the parent agent running in a /goal don't loose context, enforce a good workflow, manage the backlog and parallelize/merge back the work on the main repo

https://github.com/madarco/agentbox MIT license

you are right, I've experimented with cp -a on macOS as well for https://github.com/madarco/agentbox and in the end found it's actually faster to use worktrees inside docker containers while mounting your .git repo inside them.

Then after the node_modules (or apt packages) are installed, take a docker commit snapshot.

Now I have truly isolated parallel workspaces in <10s.

Also the system was easy to adapt to cloud environments as well so now I have Hetzner, Vercel, Daytona as well (using their native snapshotting systems for fast boot after the initial setup)

I think it's because the library calls directly the LLM AND saves any debug/trace info. While other tools let you use the standard (eg OpenAI) sdk and uses an https proxy to intercept the requests

GPT-4o 2 years ago

Whisper, the STT model created by OpenAI can understand you whispering. You can try it with the ChatGPT mobile app to see it yourself

On the contrary. Unless some AI armageddon happens, humans will be nurtured by AIs like in the movie Wall-E. Totally dependent by it, but a true utopia.

oh god, of all the amazing problems to solve with this wonderful technology, you surely did pick the most useless. What's this obsession with interviews on HN? Hired a bunch of people in my career and 1 call was enough with a success ration of 98%

Nice demo but its lacks one feature to make it practical: splitting the input documents in chunks. Without it the embeddings will be too broad, and when retrieved the docs will consume a lot of input tokens, making the request slower/expensive

Sure, because this implementation won't work in practice. He's embedding the whole documents at once, without splitting, summarization, Q&A extraction etc.

Actually in this case I don't think he's wrong. Even in Italy (highly regulated job market), Companies can restructure like this.

The only obligation is that if they need to hire a similar role in the next 1 year, they'll have to re-hire the former employees if still available.

Bambu Lab A1 Mini 3 years ago

copy on that! I often use the "print by object" feature, and it's super useful being able to print each object in a different color with no color changing

Totally agree on the fancy mics: even wireless hi-end bose headphones sucks.

Distance from mouth is the most important factor in voice quality, I've found that a cheap pair of headphones with buil-in stick mic sounded much better than a Yeti mic on the table.

It's quite easy to test yourself, but nobody does that: just record a test audio snippet using the OSX Notes app and listen to it yourself.

PS: swc and esbuild aren't good example, because most of the speed improvements comes from the fact that they are just stripping TS-specific syntaxes to generate JS code.

Also tsc is slow, sure, but only for the first run. Enabling `incremental` flag or using watch mode with `--transpile-only` usually brings compile time under 100ms, Making it practically indistinguishable from SWC or ESBuild.

Wing Commander IV 3 years ago

If you enjoyed the original Privateer game, I strongly recommend that you try Rebel Galaxy Outlaw. https://rebel-galaxy.com/

It's a nice homage to the original with improved graphics, more variety of ships and engaging space combat.

I'm evaluating Budibase for adoption at work, what is the main business model?

We'll see more and more features behind a paywall?

I really love the breadth of Internal Admin tools that are available today, however, I feel like none has an even minimal support to proper dev flows like dev/staging/production or a way to synchronize releases between our own Backend API and the admin pages.

Always loved Admin tools provided by modern frameworks, like Sonata-Admin for Symfony or Django-admin.

But still can't find a generic Admin generator/Crud tool for standard REST APIs or GraphQL

Even Retool/Budibase force you to build countless connectors/queries for each resource.

I'd pay $$$ for something that could parse my Graphql API SDL and generate an admin UI for me!