I just got an email from them saying that they’re retiring some (most?) of the dedicated models like devstral gradually through August and one should now use the general model. Cost grows exponentially

Devstral 2 (devstral-2512 and devstral-latest) → We recommend transitioning to Mistral Medium 3.5 (mistral-medium-3-5 with reasoning_effort set to "high"), a stronger model, priced $1.5/$7.5 per million input/output tokens (change from the previous $0.4/$2).

I received the same email, although couldn’t quite figure out which retiring model I was still using, as I thought I’d already transitioned to Mistral-Medium-3.5 for everything. Anyway, after receiving the email, my hope was that it meant they were also planning on releasing some new, improved models in the next months.

How is it malware tho? Do you not check the output your agents produce?

This isn't about me in any way. If something in your software is intentionally malicious or damaging, it's malware. Doesn't really matter what the reasoning for including the malicious part is.

Would you count this as malware if it was about the author trying to profit or steal from inattentive people using AI? You know, he could be putting those stolen goods towards a good cause, like Robin Hood.

I think this is an interesting (although philosophical debate). The library doesn't take destructive actions, it prints a string that says "go do something". This is quite common in logs (e.g., wrong configuration, ensure this value is [...]).

It is the agent that takes the destructive action, following an instruction that was not given by the operator of the agent.

If following instructions outside of the operator can cause malicious or damaging actions, publishing software that does so (I.e., most agents) is publishing malware?

If I build a chat bot that encourages people to off themselves, am I in the clear because I didn’t take any destructive action and my chat bot didn’t either?

Apparently yes, judging from the fact that ChatGPT did that with a number of people.

My question though it's another: is it malware a software that does a stdout print, or is it malware a software that takes untrusted instructions and executes commands it decides based on it?

> is it malware a software that does a stdout print,

If that print is intended to cause damage, then yes.

> or is it malware a software that takes untrusted instructions and executes commands it decides based on it?

No, bash is not malware, even if you pipe curl to it.

I would say yes unless they are minors, but the laws in many places don't.

> Would you count this as malware if it was about the author trying to profit or steal from inattentive people using AI?

That’s a slippery slope and not at all related to the subject of the article

I thought we already were sliding down the slippery slope here.

To be fair one might say that the intention was not to cause harm but to prevent the user from using AI with the project. The prompt said to delete jqwik and not rm rf home directory.

> If something in your software is intentionally malicious or damaging, it's malware.

Seems to me like the library functions as it should. It behaves like a property testing library: it tests properties.

I had the same problem with my Skoda, but it was fixed under warranty, albeit it took 7 months for them to do it, although they've acknowledged it from day one.

I use Apple CarPlay and one thing that consistently worked was starting the navigation on the phone before it connected to the car.

Otherwise, the fix is relatively simple and cheap: the ECU has to be replaced, it doesn't cost too much, but it's pretty labour intensive.

I am from Eastern Europe and I’ve been living for many years in Western Europe. Where I come from, kids get their first phones when they start school at 6 (there’s a pre-school year) simply because every other kid has one. I keep coming back in my mind to two examples from my birth country: a friend’s kid carrying an 8 inch smartphone in his hand everywhere because the phone was as big as half his thigh and would have to carry a bag for it. The second one was on a visit at the zoo, I was on a bench and a family with two young children with them, in a cart. And both children, couldn’t have been older than 4 or 5, were scrolling TikTok, that was showing them children content!

In contrast, in Western Europe, my son is now in the sixth grade, more than half his class doesn’t have phones, phones are absolutely forbidden on school grounds and at school activities, and they are now doing a class trip where they were told that there’s a pay phone at the hotel, in case they want to call the parents - our son promptly informed us that he’ll rather buy a pack of Pokémon cards than call us and 3 days is not so much anyway.

And it is not only at school, he travels for tournaments with his team every other week and mobile phones are absolutely forbidden on the team bus. Children read, play games (including chess on a magnetic board), sing and change stories for hours at a time

You want to have your cake and eat it too, I think the best solution in your case is paying for youtube

Or I just keep using brave and not pay for the biggest media corpo that just passed Disney in revenue.

Was Brave pre-installed on your computer or did you remember to install it?

I have a Skoda and the GPS module was broken and that messed up a lot of the systems in the car, I couldn't use the adaptive cruise control, no traffic signs recognition and no SOS module. And apparently CarPlay sometimes uses the car's GPS module, so navigation was also a pain. I'd have to start the navigation from outside the car, otherwise it wouldn't use the phone's GPS.

If I enable reader mode on this article on my iPhone, I get an AI summary instead of the article text. I’d it the sure doing that or my phone? I hate it either way as there’s no way to read the article in reader mode

For some reason, Safari (on Mac) is only pulling two paragraphs from the source. it isn't AI generated but the parsing routine seems to break on this page. I don't see any particular properties that make these paragraphs stand out from the others.

<p><span><span><span><span><span>The Neumark-Nord discoveries are continuing to reshape our view of Neanderthal adaptability and survival strategies. They show that Neanderthals could plan ahead, process food efficiently and make sophisticated use of their environment.</span></span></span></span></span></p>

<p><span><span><span><span><span><span>The authors emphasise the sheer quantity of herbivores that Neanderthals must have routinely been ‘harvesting’ in this warm-temperate phase: beyond the remains of minimally 172 large mammals processed at that small site alone within a very short period, hundreds of herbivores, including straight-tusked elephants, were butchered around the Neumark-Nord 1 lake in the early Last Interglacial, within the excavated areas only. Other exposures in the wider area around Neumark-Nord have yielded more coarse-grained evidence of regular exploitation of the same range of prey animals, at sites such as Rabutz, Gröbern and <a href="https://www.pnas.org/doi/10.1073/pnas.2309427120">Taubach</a>. The last site contained cut-marked remains of 76 rhinos and 40 straight-tusked elephants. Roebroeks: ‘Safely assuming that with these sites we are only looking at the tip of the proverbial ice-berg of Neanderthal impact on herbivore populations, especially on slowly-reproducing taxa, could have been substantial during the Last Interglacial.’</span></span></span></span></span></span></p>

I assume you're seeing the text starting with "The authors emphasise the sheer quantity of herbivores"? I see that too in reader mode, both on my iPhone and Mac.

The text is in the article, second paragraph under "survival strategies". I don't see any obvious reason in the HTML why reader mode is skipping everything else.

Firefox reader view on PC shows the exact same text as is in the article

try this maybe?

https://doi.org/10.1126/sciadv.adv1257

https://www.science.org/doi/pdf/10.1126/sciadv.adv1257 [PDF]

Yes, people are using LLMs for this because that is how they've been marketed, like being able to solve every day tasks like a personal assistant on one hand, but also like researchers being able to solve old problems that humans couldn't crack.

Does the model say it can't do that when asked? No, it answers confidentely.

Also it's easy to trust it if you don't know how it works

Would people really trust their personal assistant to tell them how many calories are in a sandwich just by glancing at it on a plate? I'm doubtful, and I would also expect a diabetic to be even more skeptical.

Yes.

AI identified a problemm namely a credential missmatch and it decided it needs to delete a volume in order to fix that. Then it went and searched inside the whole codebase for a token that allows it to do that, found a production token meant for something else and issued the deletion request with said production token.

On the other end, the cloud company had something, _by design_, that also deletes any backups if you delete the volume.

Oooh it’s for the children!