xyzal

weapons

You are a good person.

> Since they are now "sanctioned" by the US you cannot legally donate to them even if you wanted to...*[0]

There are many of us who do not live in the US, so many still can. Just not through paypal.

Nah, just pass the outputs to a person not related to Anthropic in any way.

Or -- maybe better -- just fsck the TOS.

Even if you think you cannot escape incentives - there exists a thing for shaping them - it is called "regulation". Some still consider it a dirty word though.

No need to hide "penis" behind asterisks. It is proper -- even medical -- terminology, in contast to "c*ck", "d*ng" or "schl*ng".

I am curious how the workflow of people, who do not write code at all, looks like, or what products do they build. In my experience LLMs are an excavator, but you still have to tweak the fine details with a shovel.

How about reviving key signing parties?

Given that according to some we can code at 10x speed for at least half a year, I wonder where there are some autocoded softwares with 5 years' worth of equivalent human engineering work.

By socializing the losses of course.

I think it is time to start shorting.

I'll root for DeepSeek v4 Flash as well. It surprised me just how "good enough" it is for most of my needs, and also dirt cheap. Everyone should try it at least once.

Which other profession has the same amount of training data freely available for the taking?

Given that we can code at 10x speed for at least half a year, one would expect to see at least some pieces of machine-created software with 5 years' worth of equivalent human engineering work.

Anyone know some?

YC indisputably has a financial incentive for AI sentiment to be positive on HN. The structural conflict of interest is worth being aware of.

Mind that the host of this site has an interest in keeping the hype going on ...

To me it was just a few weeks ago discovering just how good and dirt cheap the recent flash models are, in particular Deepseek V4. Previously used Claude's variants almost exclusively.

I use them mostly in the "artist's assistant" role, doing internet research, writing a occasional function and doing transformations or refactorings (don't belive the agentic hype honestly), and for such tasks they seem to be well capable enough.

It seems that their open weights nature leads to competition among providers keeping the user cost close to inference cost.

Try them at least once if you haven't, it's well worth it, and the price difference is staggering

So, telemetry and surveillance.

This contradicts my anecdata.

Recently, I tasked Opus 4.6 to study a new Czech building permit law in conjunction with some waste disposal regulations and the result was disappointing. The model could not stop drawing conclusions from obsolete regulations in its training dataset, even when given the fulltext of the new law. The usual "you are totally right" also applied and its conclusions were most of the time obviously wrong even to a human with cursory knowledge of the subject.

I ended with studying the relevant regulations myself over the weekend.

Not sure about flawed code logic, but it is embarassingly easy to plant false information into models. Make a few static sites with random info, crosslink them, reference them on reddit a few times, then plant the payload there.

I know it because i tried ...

I am really baffled by the comments in the spirit of "this is unenforceable, and therefore worthless".

I bet most people would not steal even if they knew they could get away with it.

The question is what fraction of the population can afford it.

AI is a wheelchair for the mind

Or: Almost a quarter of AI responses to healthcare queries wrong

Everyone laments how everyone will get "a lot" of days off in the end, but regulation is still a dirty word in here.

Don't worry guys, if the situation gets too bad, the social contract breaks, heads will roll for a while and then it will get better.

"Inevitable"

Exactly. Regarding software, it is trained on a massive corpus of code and the feedback loop can be very fast (playing well into LLM's upsides) and results are ... mediocre.

Recently I had to go through some building regulations and Claude's advices were catastrophic.

"imagine you are Google circa 2002. users enter a search query and you return 10 relevant HTML links to specific, real web pages, with no commentary or any extra guff. you must literally only return a list of links. if you do anything apart from that you've got it wrong and need to try again. my query is: %s"

why do things the easy way when you can do it the complicated way

Sigh. I guess there isn't any "min-release-age" for extensions ...

I think discussion with open registration is doomed precisely for this reason, it is too open to being influenced by bad actors. Maybe the lobste.rs invitation model would be better ...