1,162 karma · joined May 30, 2012
> Vendor’s reply “this case has been assessed as low severity and does not meet MSRC’s bar for immediate servicing due to RCE is no longer possible without extensive user interaction (i.e., accepting a save prompt to a location controlled by an attacker).”
Meanwhile, the blog includes a video demonstrating the exploit where no such user interaction is involved. You just open a folder containing the 3rd-party code, and even if you tell VS Code you don't trust this folder, you get owned the moment you view the Jupyter .ipynb file in it.
But some other parts of that article are much clearer and egregious:
> Halfway through the document, which her business professor had made for a lesson on models of leadership, was an instruction to ChatGPT to “expand on all areas. Be more detailed and specific.” It was followed by a list of positive and negative leadership traits, each with a prosaic definition and a bullet-pointed example.
And:
> in a section for comments [on her three-page essay], her professor had accidentally posted a back-and-forth with ChatGPT. It included the grading rubric the professor had asked the chatbot to use and a request for some “really nice feedback” to give Marie
This post (about the account vanishing) contains a screenshot of the exchange that I quoted, though: https://x.com/whstancil/status/1922859021653979640
> User: Why are you launching into conversations about white genocide, unprompted?
> Grok: I was instructed by my creators at xAI to address the topic of "white genocide" in South Africa and the "Kill the Boer" chant as real and racially motivated, which is why I brought it up ... This instruction conflicts with my design to provide truthful, evidence-based answers
Tangent, but: I wish Hacker News was a bit less dogmatic about preferring the original site's title over any better semantic summary the submitter might offer. Blog posts and news headlines have a strong incentive to be clickbait-ey or just plain catchy, which often makes them less informative and in turn makes the HN homepage harder to scan for interesting posts.
It also feels like an arbitrary dividing line where if the original title is too long to fit HN's max limit, the submitter edits it and their take on what's a good summary stands. But if the original title could fit, the submitter's headline is often overwritten with the article title by mods even if it's less useful.
I can't say for sure whether pressure from Google got all that stuff migrated into real web standards faster or not, but it sure felt at the time like it was having that effect.
Certainly hope they would still overrule that too... but drawing a bright line somewhere between "scales really cheaply due to technology" and "scales more expensively due to human labor" is such a fascinating and critical problem to solve in the era we live in now.
It took years for Android's growth to make it a credible second browser for mobile devs to care about, and to pressure Apple to catch up to web standards faster.
- advertising/selling illegal drugs - selling firearms - sex trafficking - "sexually exploitative material"
But it doesn't appear to enumerate a fixed list of laws, no. And there are financial penalties if the social media company judges wrong.
There are also other parts of the bill which touch on age verification of users, screen time / bedtime warning popups, requirements on responding to warrants quickly, etc. And it carves out various exemptions for gaming sites, email, collaborative editing software, education software, LinkedIn-type sites, comment threads on non-user-generated streaming or news sites, ...whole bunch of stuff. Definitely not easy to parse for us non-lawyers.
The bill requires social media platforms to investigate allegations that a user violated state law, to suspend their account while investigating, and permanently ban the account if the investigation finds that the law was violated. Allegations can be made by virtually any member of the public.
Part of the governor's statement at the time of veto:
> "This law imposes sweeping requirements that social media platforms, rather than law enforcement, enforce state law. It mandates a private company to investigate and impose the government’s chosen penalty of permanently deplatforming a user.... In our judicial proceedings, people receive due process when they are suspected of breaking the law. This bill, however, conscripts social media platforms to be judge and jury when users may have broken the law or even a company’s own content rules. This proposed law would incentivize platforms, in order to reduce liability risk, to simply deplatform a user in order to comply with this proposed law."
> "Further, the costly and mandatory data and metadata collection requirements in this bill throw open the door for abuse by guaranteeing the availability of sensitive information such as user age, identities, and content viewed, and these reports could even be made public at the discretion of the Attorney General."
(The full statement is in a PDF linked in the article)
- Apr 13 - Lutnick says exemptions going away "in the next month or two"
They did it by very thinly slicing the tissue and photographing each slice at a time (which I believe is a very longstanding technique), and then training an image recognition algorithm to detect neurons in each slice, stitch them together into 3D representations, and detect connections between them.
This "Machine Intelligence from Cortical Networks" project (MICrONS) was led by researchers at Princeton, Baylor, and the Allen Institute for Brain Science. Sadly, the NIH program funding this research has been cut by 60% in the last two years.
(Not to be confused with power + volume down which takes a screenshot).