apstls

What you claim contradicts Anthropic’s statements. I assume that is the contention.

It is pretty clear, no? Anthropic claims that the jailbreaks they were made aware of did not access the model’s raw capability, explained that there are protections to mitigate the impact of successful jailbreaks, etc. Coming here and stating something to the contrary with zero explanation or actual evidence is the assumption.

That's why you discard ICMP errors.

There are certainly things to be learned from analysis of the dataset. Keep your friends close but your enemies as JSON, or something...

Sit down and re-read your comment one night with your "I am an engineer and will solve this as an engineering problem" hat firmly on. If you stop thinking of LLMs as lobotimized coworkers trapped inside an API wrapper and instead as computational primitives then things become much more interesting and the future becomes clearer to see.

they are working hard on making it impossible for other models to support their every increasing agent feature set (skills, teleport and remote sessions, LSP, Chrome integration, etc). The move totally makes sense, like it or not.

I don't understand, why would other models not be able to support any, or some, or even a particular single one of these? I don't even see most of these as relevant to the model itself, but rather the harness/agentic framework around it. You could argue these require a base degree of model competence for following instructions, tool calling, etc, but these things are assumed for any SOTA model today, we are well past this. Almost all of these things, if not all, are already available in other CLI + IDE-based agentic coding tools.

Is it wrong that I take the prolonged lack of Linux support as a strong and direct negative signal for the capabilities of Anthropic models to autonomously or semi-autonomously work on moderately-sized codebases? I say this not as an LLM antagonist but as someone with a habit of mitigating disappointment by casting it to aggravation.

grep --include=*.{c,h} -rnw -B3 -A15 'XXX' ./ | claude -p 'Analyze each code snippet and pick the five most concerning, from a security perspective.'

Sounds very cool.

I wanted to try this out, so I opened Windsurf for the first time in ages and clicked the "Upgrade Available" button, which sent me to: https://windsurf.com/editor/update-linux

  Did you install using apt or apt-get? If so...
  
  1. Update package lists
  
  sudo apt-get update
  
  2. Upgrade Windsurf
  
  sudo apt-get upgrade windsurf

EDIT - I don't want to detract from the topic at hand, however - after upgrading (with `sudo apt-get install --only-upgrade windsurf` :)) and playing around a bit, the Codemaps feature indeed seems very nifty and worth checking out. Good job!

Well, slam _dump_ does sound like a potentially apt description.

Genuinely curious: what were the abuses by the Biden and Obama administrations?

There is likely a small number of people who could collectively list out the events it _did_ help Iran avoid.

The fact 2nd amendment crowd had a chance to prove their principles a few weeks ago and they did it with flying colors when the NRA came out against any restrictions for trans people.

This is effectively a strawman argument.

We are not discussing their principles that pertain to whom should be allowed to own guns, nor is that question even relevant to the heart of this topic. We are discussing the core, underlying principle that has been used as the primary justification for WHY people should have a constitutionally-protected right to own mechanized, efficient killing machines. You of course already know: the rhetoric (for literal centuries) has been that the preservation of our system of democracy from hostile internal actors requires a citizenry that has the means to effectively fight back.

Without a noble-sounding pretense of existential importance for 2A defenders to shield themselves with, things start to look a lot more like weighing "guns are fun, plus think about hunters" vs "we should reduce our yearly slaughter rate for schoolchildren to match those of other 1st world countries" and choosing the former. So, when people see what they believe to be the increasingly-obvious beginnings of authoritarian overreach in this country, and at the same time seeing most of the 2A crowd saying absolutely nothing about it, it looks like a genuinely remarkable, thoroughly nauseating display of hypocrisy and selfishness.

But as a 2A supporter I don't feel any obligation to rage against ICE assembling a social media team. These seem to be completely disconnected concepts.

Dishonest framing aside ("assembling a social media team", like how drug cartels simply assemble a team of chemists, semantics be damned!) and with full context considered, I assume you are stating here that you see no problem with the current actions of this administration, both ICE-related and not, and have caught no whiffs of authoritarian overreach. Otherwise, social media monitoring and tracking teams within an organization that does not have the rules, oversight, requirements, or legal vulnerability of existing (previously) non-partisan agencies with similar teams, would be extremely concerning.

Though after writing this out, I'm not even clear on how much of this is a matter of opinion versus a matter of awareness and understanding of current events. One needs to only look at the events of the last 7 days to find egregious evidence of authoritarian movement, things that would've sent recent Democratic presidents into political exile and impeachment. I just learned that my neighbors and I are enemies of the US last Tuesday - literally, verbatim, "enemies from within" with no additional qualification beyond living in a blue city - for example.

This statement is one of those useless exercises in pedantry like when people say "well technically coffee is a drug too, so..."

Code with publicly-known weaknesses poses exponentially more danger than code with unknown weaknesses.

It's like telling sysadmins to not waste time installing security patches because there are likely still vulnerabilities in the application. Great way to get n-day'd into a ransomware payment.

This metaphor drops some pretty key definitional context. If the common belief prior to this race was that cars could not beat horses, maybe someday but not today, then the article is completely reasonable, even warranted.

Should Fox, Newsmax, OANN, Alex Jones, Tucker, Bannon, the deputy director the the FBI (in a prior gig, to be fair), the president of the United States (current & prior gigs), members of congress, MAGA influencers like Tim Pool, the company paying Tim Pool, the people paying the company that pays Tim Pool, etc, etc, etc, and etc, be allowed to?

How could they if you don’t provide references to them?

Could you explain further? Specifically in the context of this discussion/today’s meeting and conduct?

Consider the space of possible paths towards peace in UA that the current administration can choose from. Do you believe the paths with the highest likelihood of success for reaching a mutual peace agreement involve making one of the sides unhappy, uncomfortable, angry, or by publicly berating them?

It is true that he doesn’t care if people are unhappy or uncomfortable, it’s true he doesn’t care about optics, it is true that he has a set of things he wants to get done, and it’s true the first two points influence his approach to achieving the latter. However the question - the only question that matters - still remains: is this approach the most effective course of action to choose? Maybe it’s not reasonable to assert with complete confidence that it isn’t, but it is certainly unreasonable to assert that it is.

Out of curiosity, what are the other similar public meetings/negotiations you’re aware of that were conducted in a similar way?

Or generally, since you say this did not deviate from your expectations, what are the past events that influenced your baseline expectation for conduct on the geopolitical stage?

I think these types of arguments need to at the very least acknowledge the distribution of cost between training and inference.

LLM context windows are quite large now. This is very likely a simple matter of including NRQL docs or specification in the prompt.

What is the current state of DSPy optimizers? When I originally checked it out it appeared to just be optimizing the set of examples used for n-shot prompting.

Welcome to LLM-related threads on HN.

It hinges on what the word “that” in “release that” is referring to. If it’s referring to releasing his name, then he’s implying not releasing his name is a choice which implies they have the option to release it, so they must know it. If it’s referring to releasing whether or not they know his name, then it’s not implying anything. If this was said by someone with a history of well-spoken and thoughtful public statements, then it’d most likely be the latter interpretation. Given it’s Eric Adams, either is plausible. In fact, the bullshit-ness of the former may make it even more probable here.

The endpoint for upgrading for the normal web interface was returning 500s for me. Upgrading through the iOS app worked though.

when building Crumb (https://github.com/liam-ilan/crumb)

As an aspiring engineer

I’ve got some good news for you.

It could be key spraying, maybe targeting a particular organization with distributed infrastructure for which the attacker already has some keys, but more likely groups blasting default keys (i.e. for some crappy IoT devices that included them in the firmware etc) for a nice & quick botnet.

The term 1day is not uncommon, not only among exploit vendors on either side of the law (in other words, selling to crime groups or selling to governments) but also among the general threat intelligence and broader cybersecurity community. It doesn’t stand out as strange at all, really. However, I don’t believe their particular definition aligns with the typical/colloquial usage of the term. It’s usually used in the more direct sense, i.e. a very new vulnerability that is unlikely to have been broadly patched.

As far as the term 0day, I don’t think there’s much debate or contrarian opinions to be had. The only room for argument I see is between defining it as an unreleased vulnerability unknown to the vendor versus a vulnerability known to the vendor but not yet patched, basically what this article defines a 1day as.

Either way, it comes down to nitpicking nuances of definitions of commonly-used terms. I don’t think there’s much meaningful discussion to be had.

I would love if (_consistently available_) $15/day street parking was a thing in Manhattan, it'd be a good deal cheaper than garages and obviously a lot more convenient than keeping your car elsewhere. There isn't much benefit to having a car in Manhattan for day-to-day life, but it would be nice to have for things like day trips. Right now I park my car about 45 minutes away in another borough (at my family's house) so when I do need to drive I have a +90min fixed cost added to my commute time.