CKMo

Looked into the author's history, credentials, background, and education. OpenAI couldn't find a single more qualified person to lead their AI futures?

Microsoft of all companies were the ones who had backbone here? What the heck

Reasons to not use hyperscalers, exhibit 654

There's a lot of outages this month!

There's definitely a big problem with entry-level jobs being replaced by AI. Why hire an intern or a recent college-grad when they lack both the expertise and experience to do what an AI could probably do?

Sure, the AI might require handholding and prompting too, but the AI is either cheaper or actually "smarter" than the young person. In many cases, it's both. I work with some people who I believe have the capacity and potential to one day be competent, but the time and resource investment to make that happen is too much. I often find myself choosing to just use an AI for work I would have delegated to them, because I need it fast and I need it now. If I handed it off to them I would not get it fast, and I would need to also go through it with them in several back-and-forth feedback-review loops to get it to a state that's usable.

Given they are human, this would push back delivery times by 2-3 business days. Or... I can prompt and handhold an AI to get it done in 3 hours.

Not that I'm saying AI is a god-send, but new grads and entry-level roles are kind of screwed.

Uploaded to HuggingFace (https://huggingface.co/deepseek-ai/DeepSeek-Prover-V2-671B) without any fanfare or even announcement by the DeepSeek team, but the GMI Cloud team is hosting it already!

I genuinely liked him, even as an atheist. He seemed to be trying his best to make the world a better place and I can't fault him for that.

Opening paragraphs:

"Mandiant analyzed 138 vulnerabilities that were disclosed in 2023 and that we tracked as exploited in the wild. Consistent with past analyses, the majority (97) of these vulnerabilities were exploited as zero-days (vulnerabilities exploited before patches are made available, excluding end-of-life technologies). Forty-one vulnerabilities were exploited as n-days (vulnerabilities first exploited after patches are available). While we have previously seen and continue to expect a growing use of zero-days over time, 2023 saw an even larger discrepancy grow between zero-day and n-day exploitation as zero-day exploitation outpaced n-day exploitation more heavily than we have previously observed.

While our data is based on reliable observations, we note that the numbers are conservative estimates as we rely on the first reported exploitation of a vulnerability. Frequently, first exploitation dates are not publicly disclosed or are given vague timeframes (e.g., "mid-July" or "Q2 2023"), in which case we assume the latest plausible date. It is also likely that undiscovered exploitation has occurred. Therefore, actual times to exploit are almost certainly earlier than this data suggests."

I'm pretty sure this gives credence to the monopoly argument

This is a good example of why you don't want ring0 level access for clients. Or just, you don't want client-based solutions. The provider just becomes another threat vector.