asdewqqwer

Evidently Fable is so powerful that it already allow Anthropic to break Shannon's theory.

>We will require 30-day retention for all traffic on Mythos-class models, on both first- and third-party surfaces. We won’t use this data to train new Claude models

>The data will help us defend against complex and novel attacks (including new jailbreaks and attacks that operate across many requests) as well as help us identify and reduce false positives.

As if the dishonesty of human who are good at business has not been criticized since business ever exists

So you're trying to convince a community of mostly engineers — using the example of a terribly outdated technology that stays overpriced purely through symbolization because the luxury industry's bubble keeps holding — that flashy looks, advertising, and fancy concepts aren't really beloved and worthy? Fascinating.

I keep noticing a wrong narrative when it comes to FOSS software. Comparing software to traditional labouring jobs is quite pointless: you designed and built a car for yourself, building a second one for someone else has costs; you cleaned your own private garden, cleaning someone else's garden has costs; repairing every bike, however sophisticated you are, has costs (Just listing the narratives I see on this page). Giving away a software you already have wrote for yourself has almost 0 cost, and I think this is the main motivation (at least at the beginning) of most FOSS software: the developer themselves needs to use it anyway. This uniqueness is what I consider one of the main reason why FOSS can succeed in the first place unlike in almost any other engineering fields. In a sense, you can say it is exactly what Marx call "all sources of collective wealth flowing abundantly".

This is very different from: I see an ongoing demand for X, and thus I work to provide a product that supplies X. In this case, naturally, I will not want to give it away for free, even if it is software.

I think there are plenty of such reply on github. For example the one to AMD AI director's issue.

yet the poc exploit itself take $2000 and one day, I don't know how the math works, maybe there is some extremely clever way to figure out runs that are not worthy to attempt exploit.

There is a huge gap between the shining examples and actual use case: What is the false positive rate? How to judge false positive?

If you need 1000 run that cost 20000 USD to find a vulnerability, and you need 2000 USD to generate a exploit (which makes it self-verifiable to be not false positive), than your cost is not 22000 USD but 1000x2000+2000 which is 2 million USD: you have to try generating exploit for every trial before you know it is true, or you need to hire one (or several) senior security people to audit every single of them.

A broken clock being correct twice a day is not impressive.

> assume it is the single client that will operate that account

You are still making wild assumption without actually thinking about what means to writing an email client.

If you ever try to write a email client, you will immediately realize how difficult, if not impossible, to fix all the bugs for a email client. It is a multi-different-protocol-version-async-client-handling-same-database-with-thousands-of-race-condition backwards compatibility nightmare.

Writing a email client with support of just up-to-date protocols and assume it is the single client that will operate that account is trivial, write one that covers all corner cases is a totally different story.

Sure, then governments are also just people. How about we restore Monarchy now so someone can actually be held responsible? Also we should completely abandon Nulla poena sine lege since evidently the imbalanced power does not exist between people and also people (government).