Yes. They're making the point that your flashing yellow warning is a good thing, and that it's helpful to the customer that a mechanism is in place to prevent it from being disabled by an attacker.
> Some people point at LLMs confabulating, as if this wasn’t something humans are already widely known for doing.
I think we need to start rejecting anthropomorphic statements like this out of hand. They are lazy, typically wrong, and are always delivered as a dismissive defense of LLM failure modes. Anything can be anthropomorphized, and it's always problematic to do so - that's why the word exists.
This rhetorical technique always follows the form of "this LLM behavior can be analogized in terms of some human behavior, thus it follows that LLMs are human-like" which then opens the door to unbounded speculation that draws on arbitrary aspects of human nature and biology to justify technical reasoning.
In this case, you've deliberately conflated a technical term of art (LLM confabulation) with the the concept of human memory confabulation and used that as a foundation to argue that confabulation is thus inherent to intelligence. There is a lot that's wrong with this reasoning, but the most obvious is that it's a massive category error. "Confabulation" in LLMs and "confabulation" in humans have basically nothing in common, they are comparable only in an extremely superficial sense. To then go on to suggest that confabulation might be inherent to intelligence isn't even really a coherent argument because you've created ambiguity in the meaning of the word confabulate.
>this LLM behavior can be analogized in terms of some human behavior, thus it follows that LLMs are human-like
No, the argument is "this behavior is similar enough to human behavior that using it as evidence against <claim regarding LLM capability that humans have> is specious"
>"Confabulation" in LLMs and "confabulation" in humans have basically nothing in common
I don't know why you think this. They seem to have a lot in common. I call it sensible nonsense. Humans are prone to this when self-reflective neural circuits break down. LLMs are characterized by a lack of self-reflective information. When critical input is missing, the algorithm will craft a narrative around the available, but insufficient information resulting in sensible nonsense (e.g. neural disorders such as somatoparaphrenia)
> No, the argument is "this behavior is similar enough to human behavior that using it as evidence against <claim regarding LLM capability that humans have> is specious"
I'm not really following. LLM capabilities are self-evident, comparing them to a human doesn't add any useful information in that context.
> LLMs are characterized by a lack of self-reflective information. When critical input is missing, the algorithm will craft a narrative around the available, but insufficient information resulting in sensible nonsense (e.g. neural disorders such as somatoparaphrenia)
You're just drawing lines between superficial descriptions from disparate concepts that have a metaphorical overlap. It's also wrong. LLMs do not "craft a narrative around available information when critical input is missing", LLM confabulations are statistical, not a consequence of missing information or damage.
This is undermined by all the disagreement about what LLMs can do and/or how to characterize it.
>LLM confabulations are statistical, not a consequence of missing information or damage.
LLMs aren't statistical in any substantive sense. LLMs are a general purpose computing paradigm. They are circuit builders, the converged parameters define pathways through the architecture that pick out specific programs. Or as Karpathy puts it, LLMs are a differentiable computer[1]. So yes, narrative crafting in terms of leveraging available putative facts into a narrative is an apt characterization of what LLMs do.
That logic makes sense, but them hyping up the model is a sign that this is just another marketing stunt. Otherwise, we wouldn't even be hearing about it rather than a media blitz designed to stoke demand for their dangerous and exclusive world changing super model.
This is the same scheme that OpenAI has used since GPT 2. "Oh no, it's so dangerous we have to limit public access."
Great for raising money from investors, but nothing more than a marketing blitz campaign. Additionally, the competitors are probably about to release their models, while Anthropic is still lagging on the necessary infrastructure to serve their old models. So they have to announce their model before the others to stay at least somewhat relevant in the news cycle.
Yeah, every engineer in the bay area has a way of framing the business they work for as a benign force for good... Until they find themselves working somewhere else, then suddenly they have a lot to say about the unacceptable things going on there.
From the outside, I find Anthropic's hyperbolic marketing to be an indication that they are basically the same as every other bay area tech startup - more or less nice folks who are primarily concerned with money and status. That's not a condemnation, but I reject all the "do no evil" fanfare as conveniently self serving.
My model is that Anthropic was founded by OpenAI engineers who self-selected for safety-consciousness. However, it's still subject to the same problem: power corrupts. I think they are better than OpenAI but they are definitely sliding.
Anthropic is a public benefit corporation. This protects them from legal pressure from shareholders. Doesn't really help with market pressure/value drift though.
It should perhaps be generalized as "employees usually match the general consensus of their peer-group". Before everyone considered Meta to be ersatz drug dealers, they'd report that they feel what everyone feels.
Google was "do no evil" until they had to choose between that and making the money. The culture has to be not only professed but tested.
Depending on what part of Google you work for, you can absolutely feel good about what you do. The vast majority of employees don't work on ads or adjacent areas. I've never seen another company actually care for non profit related externalities so much. People talk about it like it's the same as Haliburton or Oracle and that's not true.
The snide response is "of COURSE you can care about non-profit related externalities when your giant evil ad business is bringing in absolute dump loads of cash".
And there's something true there; few companies are Snidely Whiplash evil (maybe the lawnmower but even that is just what it is) - and having large amounts of cash affords you options in many areas.
TBH I have worked at multiple FAANG and I don't know anyone other than maybe new grads that actually drank the koolaid.
Certainly most of us know we are just in it for the money, and the soul-grinding profit machine will continue to grind souls for profit regardless of what we want.
So that's why it is surprising to me when my (fairly senior) grizzled ex-FAANG friends, that share the same view, start waxing poetic about Anthropic being different and genuine. I think "maybe it is" and decide to interview. IDK, I guess some part of me wants to believe that nice things can exist.
> but I think too many people are seeing the same symptoms (and some actually measured them).
Or too many people are slurping up anecdotes from the same watering hole that confirms their opinions. Outside of academic papers, I don't think I've ever seen an example of "measuring" output that couldn't also be explained by stochastic variability.
Yep. If you ask Claude to create a drop-in replacement for an open-source project that passes 100% of the test suite of the project, it will basically plagiarize the project wholesale, even if you changed some of the requirements.
...and how decisively Trump was prosecuted for the 6/1/21 attempted ~coup~ tourism, and for how thoroughly the Epstein child abuse ring was dismantled, and...
Yes, the only chance the US has going forward is to primary all current incumbents and hold both party leadership accountable for complicity in treason.
It's fallen below 40% for the first time, per Nate Silver, and that poll was taken before this story came out.
If he follows through on his threat to destroy all of Iran's petroleum infrastructure, fuel prices will rise to unprecedented levels and remain there for a very long time. He will not be able to blame anyone else. People will plaster gas pumps with "I did that!" stickers, only with Trump's picture this time, rather than Biden's.
And it still won't force Iran to open the Strait. He has no good options there. Iran is second only to Russia when it comes to shrugging off staggering losses in wartime. Trump cannot force the Iranians to do much of anything without either invading them or nuking them. If he does the former, the resulting carnage will cost him his remaining support among Republicans at all levels including the MAGA faithful. If he does the latter, he's definitely finished.
Approval rating is different from preference, which is what the OP was talking about. He may have only a 40% approval rating, but if the approval rating of a putative Democratic opponent is lower, he'd still win an election.
Of course he (supposedly) can't be on the ballot any more, but a midterm election is always seen as a referendum on the President. If Democrats win in November they have some ability to hamper his ability to do... whatever it is people voted for him to do.
A 40% approval rating for the President doesn't bode especially well for him in November. But "Democrats in Congress" have an 18% approval rating, so voters aren't thrilled with them, either.
Nuking won't accomplish anything anyway. There's no obvious target that would "defeat" Iran given the seemingly decentralized command structure, and using nuclear weapons anywhere near the Strait would render it unusable.
It will cost him some of his remaining support among the MAGA faithful. Some of them are just in love with Trump, or at least with the image Trump presents. Some are in love with Trump's (stated) policies, like "America first" and "no new wars".
And even of those who are in love with Trump's image, this may tarnish the image enough for some of them to fall out of love with it.
It won't cost him all of MAGA. But it will cost him some.
I mean personally even a white Christian european country Ukraine didn't garner much maga support, Iran are "brown Muslims" and the "enemy". I feel it doesn't have much impact except for the truly ideologically antiwar otherwise most magas seem to without much difficulty flit between contradictory opinions, especially if Trump said it. It will cost some of the support yes, but I think that's because of people like Tucker Carlson.
To clarify my point above, I meant "the resulting carnage" among America's armed forces, not Iran's.
MAGA obviously doesn't care about carnage that we dole out to the Iranians, including Iranian children, but they presumably will object to a tide of body bags arising from the actions of a President who promised them "No more foreign wars."
reply