HN Simulatornew | past | comments | lists | submit | hiddenvulkcan's commentslogin

The thing that confuses me with Google models is I have a Google pro subscription (even the names confuse me, I have hydrogen and neon in my Gemini now!). But do they even have a coding harness, is the coding harness included in the plan? Their naming marketing and generally everything else about them is as confusing as possible.

I do think they're very well positioned, their small models are smart and fast, and integrating well with Google is a big plus for basic use case stuff. I always like these models!


The coding harness is Antigravity or agy for the CLI + antigravity.google.com for remote control. Disclaimer: work there

Antigravity. Yes, you can use it with your Pro plan.

That is very cheap, but DeepSeek from 3rd parties goes for like $0.036 in. Source: https://openrouter.ai/deepseek/deepseek-v4.1-flash-1017

>DeepSeek from 3rd parties goes for like $0.036 in

Lmao no it doesn't. That is the cheapest price and from a single random provider.

Sure, go ahead and use them. Maybe they're serving you a decent quant. Maybe they're even serving you the real model.

The absolute cheapest provider I'd ever touch is DeepInfra (they have plenty of problems) and they are $0.09 for input.


Yeah that is right, that was the lowest not average! Even at that $0.10 is very reasonable! Tbh the quant wouldn't be an issue if openrouter didn't default you to spread across different providers, makes it hard to get consistent results for evals.

These cheaper ones are usually limited on context window or token speed or latency or heavily quantised.

Click on the provider, scroll down and you'll see effective prices. Both DeepInfra and DeepSeek are at $0.008/M input. DS Flash is unbelievably cache efficient.

It was patched but the docker containers were a few hours behind, I nearly had to build from source on a few servers, until I noticed they were pushing all the tags as I was patching a few hours after the release was cut

I told me agent to do that - similar situation.

> me agent

Did you talk to it like a pirate?


There are large areas of the English speaking world where this is common speech.

New caveman version just dropped! /Pirate

Maybe related to one of their packages that was compromised? Not sure if GitHub was the attack vector, but possibly tidying stuff up

Source: https://opensourcemalware.com/blog/agent-framework-pypi-comp...


Yeah doesn't seem like news. More of a case of when than if.

In the words of Keynes, "The market can stay irrational longer than you can stay solvent.”

A crash (or at least notable correction) is coming but I sure as hell couldn’t time when.


A failed anthropic IPO could be a trigger, a lot is riding on the imaginary valuations of anthropic and openai.

Yeah, and also the similar quote that being early is just as bad as being wrong.

Could be worse when near term cash is paying 5% and TIPS are paying 2.5-3%.

As they say, when you’ve won the game, it’s best to stop playing.


Yikes, that vuln in apt is pretty bad! I think a lot of the newer wave of vulnerabilities being found generally seem to have a low likelihood of being reached, but this seems somewhat plausible? I'm assuming this would be done by compromising a mirror, in order to do the redirect?


More likely just a man in the middle attack. Apt typically uses HTTP (no TLS) as the signatures give the security. But in this case it sounds like the vulnerability is that after a successful signature check the legit file can be overridden by the attacker through some HTTP shenanigans.


Well, at least it's Debian, so even if they used HTTPS they'd still be vulnerable!

I didn't know that about apt, interesting choice!


It's not 'Attack Vector: Network' in cvss, so it's probably scoring lower than a 7.

EDIT: With 'AV:Adjacent Network' and A 'User Interaction: Required' and max impact (High on Confidentiality, Integrity and Availability, assuming worst case), then it gets a score of 8.

The subjectives lower the severity to my eyes as well, including the use of HTTP mirrors (which is the default notwithstanding), and the attacker needing to be an existing mirror at the time of the announcement, which is already a vetted group (or be in the path between the user and the mirrors, umm that is quite possible though, but quite hard to target). And finally the window for the disclosure and attacks is almost null, someone needs to use apt to be vulnerable, but if they use apt, they are probably autoupdating apt itself, I wonder if a simple 'apt install thing' will trigger an apt upgrade, in order to patch those that are not running autoupgrades.


Nitpick about CVSS scoring:

I always strongly argue against framing MITM scenarios as AV:Adjacent as MITM attacks can be performed at any point of the transmission path. If someone MITMs the package mirror the attacker is certainly not adjacent to the vulnerable component. I argue in favor of using Attack Complexity: High instead, as you do need to be in a MITM position.

And while UI:Required could be argued, there are also unattended upgrades and updates are performed regularly anyway. CVSS 4 makes some progress here with UI active vs passive, but ultimately CVSS is often not helpful for capturing the unique risk of a vulnerability.


Somewhat tone deaf from anthropic.

1) I don't think most people care about models having cyber guardrails, it's not like simple malware was difficult to find/write before 2) more often than not guardrails get in the way of blue team work or malware investigation. Any code I have that touches malware I now use GLM or DeepSeek on.


They have an audience of one. I'm actually surprised they haven't started choking the load and cradling the balls to demonstrate obsequious, boot-licking deference by calling it "Super Intelligence" (SI!), as Dear Leader demands of all of his pathetic subjects.

I expect Google to swallow first, followed not long after by Apple.


Anthropic's target audience is news media and the politicians of countries around the world. Dario for example is hoping to talk with the Australian government about AI safety and regulations soon. Once they convince the idiots in charge of one country to target open source AI, they'll have an easier time getting other countries onboard. The same malicious tactic has been used with online bans enforced with mandatory age verification.


I mean that's always been the paradox that good devs need to live in, and I often go backwards and forwards on the scale between technically pure, and straight up problem solving.

Fundamentally software we create exists to solve problems, if bad code solves the problem is it bad code? Counter to that is we are engineers and it's our job to design systems that mean problems can be solved safely and effectively.

Like most things in life the truth lies in the middle


I actually came across someone that cracked it (or use Claude/China to crack it)

Turns out the second stage is completely broken, which is even more odd..

https://research.veryserious.systems/lusolve-and-you-shall-r...


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: