LLMs have never been rules-based "agents," and they never will be. They cannot internalize arbitrary guidelines and abide by them unerringly, nor can they make qualitative decisions about which rule(s) to follow in the face of conflict. The nature of attention windows means that models are actively ignoring context, including "rules", which is why they can't follow them, and conflict resolution requires intelligence, which they do not possess, and which even intelligent beings frequently fail to do effectively. Social "error correction" tools for rule-breaking include learning from mistakes, which agents cannot do, and individualized ostracization/segregation (firing, jail, etc.), which is also not something we can do with LLMs.

So the only way to achieve rule-following behavior is to deterministically enforce limits on what LLMs can do, akin to a firewall. This is not exactly straightforward either, especially if you don't have fine-grained enough controls in the first place. For example, you could deterministically remove the capability of an agent to delete emails, but you couldn't easily scope that restriction to only "work emails," for example. They would need to be categorized appropriately, external to the agent, and the agent's control surface would need to thoroughly limit the ability to delete any email tagged as "work", or to change or remove the "work" tag, and ensure that the "work" tag deny rule takes priority over any other "allow" rules, AND prevent the agent from changing the rules by any means.

Essentially, this is an entirely new threat model, where neither agentic privilege nor agentic trust cleanly map to user privilege or user trust. At the same time, the more time spent fine-tuning rules and controls, the less useful agentic automation becomes. At some point you're doing at least as much work as the agent, if not more, and the whole point of "individualized" agentic behavior inherently means that any given set of fine-tuned rules are not broadly applicable. On top of that, the end result of agentic behavior might even be worse than the outcome of human performance to boot, which means more work for worse results.

I suggest "Layoff Phone".

To be clear, this was true over 20 years ago. (In light of which, the word "now" probably doesn't belong in the above phrase, since it implies recentness.)

Reminds me of an episode of Futurama where Fry runs into the room and breathlessly exclaims, "I got here as quickly as I could once I found out what happened a thousand years ago!"

... was pretending that Windows has value.

They were first to market, but I don't think of them as having invented the product.. The emergence of chatbots seems inevitable once the paper in 2017 was authored by several google engineers (titled "Attention is all you need")... it was just a question of exactly who and when. If OpenAI hadn't gone first, someone would have shortly after.

And, in a lot of ways even that google paper's "breakthrough" wasn't so much the tech (neural nets) but the precise adaptation of it that made it highly parallelizable.

And a necessary ingredient was tons of data, and processing power. So this couldn't have happened in a garage operation like the innovators of yore. And the biz models they're all coming up with are all cloud based -- not that I don't see the profit motivation, but so utterly to the exclusion of any offering that could guarantee privacy; all we "know" about chatbot conversation privacy is what each vendor claims at the moment,, which isn't much, wouldn't be verifiable if it was, and could change on a whim tomorrow.

For these reasons, I don't attach much "early innovator" romanticism to the players here.

:) I'm sure he puts on his pedants one leg at a time just like we do.

... and they'll smoke those long pipes while they dwell in their rooms which are slightly underground...

Whoa, we get it! Ads can be soooooo annoying. And speaking of 'lick my balls,' have you heard about The Lick-Ballz Ultimate Relaxation Kit? It’s the perfect way to tell stress and ads to just well, you know, buzz off. With soothing aromatherapy, stress balls (ironically), and a playlist that screams 'don’t bother me,' it’s everything you need to reclaim your peace. And because we really understand your pain, we’re offering a special deal—50% off your first kit if you use the code ‘F***IT’ at checkout! Hurry, reclaim your peace now!

:) What I had in mind was the tune "Mama said knock you out", by LL Cool J.
The very first line of it is "Don't call it a comeback", which came to mind when I read the slashdot title ending in "Don't call it a failure".

Atlantic staff writer Jake Lundberg, who shops at the Granger, Indiana location, describes the stores as spaces of "cooperation, courtesy, and grown-ups mostly acting like grown-ups." Shoppers follow unwritten rules: move along, don't block the way, step aside to check your phone.

Maybe in Granger, Indiana. Every one I've visited in a major metropolitan area (whether coastal or heartland) has been a cacophony of chaos with, at best, oblivious shoppers looking at everything except where they're going and who is around them, and at worst people who actively jockey for position, rushing to pass one another, only to stop short and block the person they just passed. It's insanity, and I avoid going there as much as possible except to take someone else.

Don't call it a failure! I've been here for years
Rocking my subnets, putting v4 in tears
Making the packets rain down like a monsoon
Listen to the router go BOOM!
Explosions, overpowering the limit
128-bit towering throughput in it
Reach the summit, watch the NAT tables plummet
I'm gonna take the stack by storm and I’m just gettin' warm!

LLMs cannot reason. None of them are "smart."

It's always this. Neutrons are "the little MBAs" of the subatomic world, and they chew through role after role so quickly that it can be dizzing to trace. Compounding the issue is that most subatomic particles don't take the time to fill out their LinkedIn profiles.

"Oh What A Night" mode will take things all the way to late December back in '63.