netsec_burn

Opus remained better than GPT for me, even after the release of GPT-4o. VERY happy to see an even further improvement beyond that, Claude is a terrific product and given the news that GPT-5 only began its training several weeks ago I don't see any situation where Anthropic is dethroned in the near term. There are only two parts of Anthropic's offering I'm not a fan of:

- Lack of conversation sharing: I had a conversation with Claude where I asked it to reverse engineer some assembly code and it did it perfectly on the first try. I was stunned, GPT had failed for days. I wanted to share the conversation with others but there's no way provided like GPT, and no way to even print the conversation because it cuts off on the browser (tested on Firefox).

- No Android app. They're working on this but for now, there's only an iOS app. No expected ETA shared, I've been on the waitlist.

I feel like both of these are relatively basic feature requests for a company of Anthropic's size, yet it has been months with no solution in sight. I love the models, please give me a better way of accessing them.

Claude 3 Opus by a wide margin. I'm a regular GPT-4 user who had tried Claude 2, and went into Claude 3 with muted expectations. I was shocked with how much more capable Claude 3 Opus was compared to GPT-4, it's not even close for my work (optimization, programming/algorithms). I asked 20 questions and Claude 3 solved all of them where GPT-4 failed on all of them. What's more surprising to me is that I don't remember GPT-4 being this bad, I was similarly impressed when GPT-4 was released. The disparity was significant enough for me to reconsider my subscription to OpenAI, but the Browsing capability as well as the Android app kept me. I use Opus for my work now though, by default, and fall back on GPT.

Well good thing Google doesn't count Google Sheets toward the overall Drive storage.

> Tools to evaluate LLMs to make it harder to generate malicious code or aid in carrying out cyberattacks.

As a security researcher I'm both delighted and disappointed by this statement. Disappointed because cybersecurity research is a legitimate purpose for using LLMs, and part of that involves generating "malicious" code for practice or to demonstrate issues to the responsible parties. However, I'm delighted to know that I have job security as long as every LLM doesn't aid users in cybersecurity related requests.

The unexpected return of Microsoft Sam.

This is an interesting direction. It's worth noting that biometrics, like fingerprints or facial recognition, aren't really 'secrets'. They can be observed or leveraged without a users knowledge or consent, and in many ways function more like a username than a password.

As a layperson in this field, what is stopping an attacker from removing the battery and stopping the motion of the IHSM for reverse engineering?

That's what is currently being done! A few videos have surfaced of smaller grains of LK-99 fully levitating. We're waiting for additional confirmation.

For me, today and almost every day since the beginning of this year. Not sure if that saying applies here.

It's outdated.

From your perspective, how would modern society be "done" if GPT-4 was generally available? How would it be substantially different from LLaMA?

Truly. I appreciate the civil, thoughtful discussion on HN. On Reddit, I've noticed the comment section becomes who can come up with the best quip or the strongest reaction. Here you have leaders in their field, and interesting news. And the UI is minimalist and perfect. What more could you want?

About .5 tokens per second if I had to guess.

I'm running it on Serge with 51 GB of RAM. I don't think it requires any GPU memory (llama.cpp runs on the CPU), I have a RTX 2060 in the system I'm running it on. Correct me if I'm wrong.

The Sonicare app lets you download the latest firmware blob. So all you'd do is intercept it and find the function responsible for generating the password.

Ah true! It isn't for commercial use.

What you're missing is not reflected on the leaderboard right now, Guanaco 65B.

What is the last link?

Trading one class of vulnerabilities for another. Memory corruption bugs for logic bugs. The disclaimer on the README is correct, this is not secure. From a five minute cursory review I identified that you can enumerate file paths in directories that you don't have privileges to view (prior to sudo-rs authenticating you via PAM).

Yes.

It was changed to "Urgent: Sign the petition now".

Couldn't it just use conversation history, where it already stores the responses, and search within that?

Pseudocode:

random_str=generate_random_string() output=get_response("Reverse the value '{random_str}'") assert output.strip() == random_str[::-1]

If you're a blind programmer, would you rather listen to a summary of each function in a source file or listen to every token on 10x speed and figure it out yourself? I believe that is the point that they are making. Not sure about the programming while biking though, I typically need to concentrate on one task at a time (which is what most multitasking is realistically, just quickly switching between one task at a time).

I think what they're saying is, you can quickly understand the contents of a file by having it summarized for you audibly. Versus having each token read out to you. Then you can locate the part that you're trying to program faster.

That is the purpose of resources like MITRE's ATT&CK. Here is the ATT&CK page on the technique which is portrayed to be novel in this post: https://attack.mitre.org/techniques/T1574/006/

This is the same company that wasn't aware that LD_PRELOAD has been in use since the 90's in their last post. Here is that unique method of hijacking execution flow using /etc/ld.so.preload, 20 years ago: https://seclists.org/incidents/2002/Jan/86. None of this is unique or novel, including replacing the loader. LD_PRELOAD rootkits have severe drawbacks (which are not an issue for eBPF rootkits).

That is a tu quoque fallacy.