HackerTrans
TopNewTrendsCommentsPastAskShowJobs

jgilias

4,162 karmajoined il y a 6 ans

Submissions

Zuckerberg says Meta made 'mistakes' in AI workforce shift

reuters.com
2 points·by jgilias·il y a 29 jours·0 comments

US Military personnel are being targeted using location data

reuters.com
14 points·by jgilias·le mois dernier·4 comments

comments

jgilias
·il y a 3 jours·discuss
I get what you’re saying and agree with the last sentence. Just wanted to touch on the “why” part.

In the world of exclusively human written software the existence of the artefact itself (code, documentation) served as the proof that there’s someone with half a brain behind it. Now that’s not the case anymore.

The conclusion stays though - it’s OSS, authors/maintainers have no obligation to anyone to do anything. Like it, use it, don’t like it, don’t use it.

As for me, I’ve found that the community and activity proxies are still good.
jgilias
·il y a 3 jours·discuss
I know it’s a typo, but I love the idea of “hood software” lol
jgilias
·il y a 5 jours·discuss
I have a degree from a German University and don’t have C2. That requirement can be interpreted as “must not be an immigrant”
jgilias
·il y a 12 jours·discuss
“Hey Claude, fix the issues with Chinese resellers and distillers. Make no mistake”
jgilias
·il y a 12 jours·discuss
Wasn’t tbe Tulip Bubble the original bubble?
jgilias
·il y a 14 jours·discuss
Can’t tell if you’re being sarcastic or not, but yes, actual value is at best tangential to success in business.
jgilias
·il y a 14 jours·discuss
It’s there, click on “words”
jgilias
·il y a 17 jours·discuss
Did it run the code to get the STDIN/OUT?

Edit: As in, actually built the binary to carry out the request?
jgilias
·il y a 17 jours·discuss
I am not my grandfather, and neither current descendants of Romans are Romans.
jgilias
·il y a 26 jours·discuss
Claude, make me rich. Make no mistakes.
jgilias
·il y a 28 jours·discuss
Ever buying a Fanta?
jgilias
·il y a 29 jours·discuss
The cost is far from linear though. Because of prompt caching and the fact that generally output tokens are a lot more expensive than input tokens.
jgilias
·il y a 29 jours·discuss
> Even if you get the right molecule, it might be the wrong way around or just clump up into a useless mess.

Sounds a lot like vibe coding lol
jgilias
·il y a 29 jours·discuss
inb4 someone calls Bessent to explain how this can be used in fentanyl production.
jgilias
·le mois dernier·discuss
“You’ll own nothing and be happy”

Thanks, but no thanks!
jgilias
·le mois dernier·discuss
Oh boy. I see this so much.
jgilias
·le mois dernier·discuss
Cool. But.

Most of the “impressive” stuff is not “the model” but “the harness”. Spinning up the subagents and teams of lower models, letting them explore, do adversarial coding. It’s all in the harness. Granted, Mythos might be better at that orchestration, but it’s still the harness.

Second is the prompting. The author is an expert in what they’re doing and prompts the system in a way that yields useful results. I see too many people believing that if an expert can achieve those results in a domain they’re familiar with, then them as non-experts will be able to as well. And that’s a fallacy that Mythos doesn’t change.
jgilias
·le mois dernier·discuss
One could argue that scalability matters more when software is expensive to make, as you need to reuse it to make the cost worthwhile.

So, for context, two data points I have that make me want to argue for the opposite side.

First, some time ago I worked for a startup that had a B2B offering that in most cases involved integration costs to align with whatever the client was already using. We tried to eat this as much as possible, but we still had to have some “integration price” we asked. More than once we had a potential client who just couldn’t lift it. They needed the software but just didn’t have the cash buffer for the initial cost (us neither). With how things are now, we’d have onboarded all of them. And much faster than it normally took. And yes, they still would have bought our solution instead of rolling their own (see the next point).

The next point that kind of ties into the previous one if you squint. I’m in a position now where I see non-technical people building stuff with AI. _Most_ can’t. As an example, the AI says they need a database. But they don’t really know what that is, and deploying one sounds scary, so they ask the AI if they can build it without a database. And the AI happily complies and makes a “CRUD” API that “persists” data in RAM. And the AI is not being dumb here. The best, most perfect model is still an LLM at the end of the day, so it completes the context window. Sure, you could make a mod that “sticks to its guns” more, but that comes across as the model being “non-compliant” and “difficult”. Now, I’ve also seen non-technical people who have succeeded. But then they have the kind of a mindset that they could’ve been engineers in the first place in different circumstances. But also, even they build fragile monstrosities that they don’t understand.

So, going back to the first point. Our clients were deeply nontechnical for the most part. Most of them wouldn’t even have attempted to build their own. But also, getting the system up and working involved more than just code - relationships with suppliers, some legal stuff, etc.

So, I can totally see how the amount of software produced might grow exponentially leading to “pre-AI” engineers being worth their weight in gold due to that. That doesn’t exclude a painful transition though.
jgilias
·le mois dernier·discuss
That was in the world where making software was _prohibitively_ expensive.
jgilias
·le mois dernier·discuss
> And we all know the demand is drying up.

I don’t think the data really supports this? Last I checked at least.