"Those cases involved reduced safeguards, yet they still occurred at two of the most technologically advanced organizations in the world.
A second point: Those two "technologically advanced" organizations apparently have the most incompetent network support staff in the industry who are apparently completely unfamiliar with the concept of network segmentation.
Not to mention a third-party Israeli company who apparently botched the sandbox configuration at BOTH companies - which is in itself suspicious of ulterior motives.
The incompetence arguably rises to the point of criminal negligence, given the outcome. At the very least it rises to the point of civil suit, which as I mentioned in a comment to Marcus' article, should result in the plaintiff attorneys getting full discovery of all corporate records of the event, so we can just how incompetent these two companies (and Meta) are and settle once and for all whether they were "deliberately incompetent" - as many people still suspect.
I do not. Gary Marcus made very good points in his Substack article on the subject, as did I in comments to that article.
Patel's terminology is reflected in uncritical news media reports of the event and adds fuel to the "AI doomers" notions. It is precisely why someone outside his range of expertise should not be reporting on such events.
One idiot commented to Marcus' piece by listing a (Gemini-generated) research list of "respected experts" who have used the term "new species" to describe LLMs. This is precisely the sort of nonsense you get when people are using human language incorrectly to describe algorithmic and probability-based events as phenomena similar to living behavior.
NO terminology which describes AI events in terms normally relegated to living and/or conscious entities should be used. It immediately confuses laymen who have little understanding of the technology involved. And given the social and political aspects of the spread of LLMs, it is even more important not to so this.
I don't even like referring to AIs as "he" (or "she", not that anyone does AFAIK.) I don't even like referring to LLMs as "AI", although I grant they are obviously part of AI research; they just aren't fully fledged AI, but merely another application of AI theory, much the same as expert systems and frames. But I'll grant that ship has sailed.
Agent swarms are a workflow path. Cross-asset, watch which compute and margin assumptions still clear when models change work again.
"Those cases involved reduced safeguards, yet they still occurred at two of the most technologically advanced organizations in the world.
A second point: Those two "technologically advanced" organizations apparently have the most incompetent network support staff in the industry who are apparently completely unfamiliar with the concept of network segmentation.
Not to mention a third-party Israeli company who apparently botched the sandbox configuration at BOTH companies - which is in itself suspicious of ulterior motives.
The incompetence arguably rises to the point of criminal negligence, given the outcome. At the very least it rises to the point of civil suit, which as I mentioned in a comment to Marcus' article, should result in the plaintiff attorneys getting full discovery of all corporate records of the event, so we can just how incompetent these two companies (and Meta) are and settle once and for all whether they were "deliberately incompetent" - as many people still suspect.
"I broadly agree with Dwarkesh’s framing."
I do not. Gary Marcus made very good points in his Substack article on the subject, as did I in comments to that article.
Patel's terminology is reflected in uncritical news media reports of the event and adds fuel to the "AI doomers" notions. It is precisely why someone outside his range of expertise should not be reporting on such events.
One idiot commented to Marcus' piece by listing a (Gemini-generated) research list of "respected experts" who have used the term "new species" to describe LLMs. This is precisely the sort of nonsense you get when people are using human language incorrectly to describe algorithmic and probability-based events as phenomena similar to living behavior.
NO terminology which describes AI events in terms normally relegated to living and/or conscious entities should be used. It immediately confuses laymen who have little understanding of the technology involved. And given the social and political aspects of the spread of LLMs, it is even more important not to so this.
I don't even like referring to AIs as "he" (or "she", not that anyone does AFAIK.) I don't even like referring to LLMs as "AI", although I grant they are obviously part of AI research; they just aren't fully fledged AI, but merely another application of AI theory, much the same as expert systems and frames. But I'll grant that ship has sailed.