simon84

This is the reason SBOM is an attempt to force documenting dependencies.

It will not solve everything but it helps.

Other than that, it is a reponse to one's laziness to import a full library to use only one method... it is part of my code review to always question the need for imports and (try to) weight the maintenance cost.

EU companies are judged guilty of negligence because backups were not totally disconnected (even though distant site) and ransomware did destroy them.

So that is starting to dig deeper than a plain mistake. I guess we will soon-ish witness the first AI slop trial going on, this will be interesting to follow

It's not so much about velocity or quality, both of which LLM do (or will) provide.

The real question is about accountability and liability.

When a major data leak is going to happen, who will they sue or fire ? That is the value engineers provide. They understand, confirm, and take ownership.

I can understand the motivation for CLI-based clients, but the browser built-in network inspector overlaps a lot.

There is more to Oproxy with traffic shaping but would it be enough to convince ? Spawning a Docker is easy today but it would be less friction with a normal app imho.

Nice I get it now. And from your usage, a natural question is how much effort was put by the AI to find the verified answer ? Did it need to run significantly longer or less that without these constraints ?

I'm not sure if the feature is intersection or the use of AI to write it...

Polygon intersection is a well known thing. Video games and geographic information systems (topology) do that for decades.

Tell me more, what should I look at ?

Can you explain how it passes through firewall and lands on a somewhat listening socket on the remote, without being abused meanwhile ?

The most surprising for me is the network continuity. This is nice (I haven't tested it).

Now I am more into stateless things for which the server is somewhat expected to fail, so the whole feature of job continuity is not a switch incentive for me.

I think the hype does not quite cover the use cases it provides.

It is not ordinary computing, is applies to a very limited set of problems.

The buzz is that it is strong at cracking cryptography, but that's about it with what we have today. The rest is scientific exploratory statistics.

Is the cost of infrastructure and tokens accounted for ? You should add this to your dashboard to be fair.

Posting on a Saturday at Middle-of-the-Atlantic-ocean time might not be the best audience catcher

About the same feeling here. I guess not everything is about global banking scale.

I've tried clever tricks to get AI produce unsupervised stuff and came back from it. The slop and loss of cognitive knowledge about what it did was uncomfortable to me... I cannot understand how you would hand off critical job to it.

I wish :) The style could have been mine, but it diverges quickly if context is too long.

Since I didnt write the content, my brain does not have the same ownership feeling.

I think the thought process, doing and undoing makes it yours, in this case we skip that part.

I actually did this once, it was great, it actually added credibility somehow I think.

After an article, I told Claude that he can add its own note at the end as the "author" and he shall tell how I promptes him and then give a honest self-assessment.

You can check this post (regardless of the content, just check the P.S. section at the end)

https://www.linkedin.com/pulse/aeonics-simon-uyttendaele-ozt...

AI is not so good at identifying a writing style by name.

Instead I feed it a few articles I wrote and tell it to follow the same style.

This gives way better results imho

While I second that, Im not convinced that the quality of the content is altered by use (or not) of AI.

A click-bait for brainless content will not stick long on the front page anyhow.

If legit contributors feel more comfortable with AI reviewing their posts, I think that is fair. And I'd rather read a structured post instead of fuzzy ideas.

My take is that if you use it for supervized coding then max5 is more than enough.

It becomes a bottleneck when using agents that spawns to tackle larger tasks and do random trial and errors to correct itself autonomously.

As a dev, understanding the code is a vital part of my job, hence I cannot let agents write more than I can review (and not only review, but fully understand). So I'm also not benefiting from x100 coding speed, but I'm not losing control either.

Are those random people or do you have somewhat categories? I.e. I want to challenge a mechanical engineering idea, not everyone even knows what that is.

Rate limit from Cloudflare, so much for moving to EU Stack :/

I keep it open in the background, sound helps pretend I'm not alone when remote working.

As simple as can be, entirely useless, love it :)

Opinionated: I dont get the issue about people (incl. bots) visiting your website, regardless of where they originate from and what url they use.

Publishing something online means I want it to be seen.

Relying or enforcing Referer or query string basically contradicts the intent. Wanting to know where your users come from could be legit but then you basically distort the "public web" aspect to a "public trap" where you try to sneak on visitors.

Not my vibe

Actually it was sort of fun to see that the AI started writing comments to itself by gradually explaining what it was trying to do and ways it failed to do it.

Then it spent more time appending comments to its own comments rather than writing code ^^

Personally, i've taken a serious step back from 'unsupervised' vibe-coding. When the codebase is clean and you want some additional fix or small feature, Claude is quite good at mimicking your style and does a pretty good job.

When asking for a new major feature, despite hard guidelines and context (that eat half your context window), then it quickly ships bloat. The foundations are not very well organized and this is where you acknowledge it is all about random-prediction of the next word-thing.

Overall, i've wasted more time reviewing the PR and trying to steer it properly than I expected. So multi-layer agent vibe coding is no longer the way to go *for me*. Maybe with unlimited tokens and a better prompt, to be investigated...

Link this with the fact that anyone can use any name/email in commits and appear as the legitimate contributor on GitHub and it completes the chain. Love it :)