A plan means it preps all work for the agents up front, tests that evals work, makes sure the dev environment is right for each agent, then finds and fixes each before the distributed tasks even begin.
What I thought would take minutes took hours as a supervisor or one agent did the prep / pre flight work.
My solution so far has been to drop all but basic setup and force the supervisor to ask before every op - if this is not the design choices, can this be run in parallel? If so, hand it off NOW.
I'm still iterating this workflow, but less setup for all the minions plus handing them work that may be incomplete/ broken is caught and fixed by the minion and its own qa gates.
This can mean a number of minions end up replicating the same fixes, but in general the time cost of that is small Vs the supervisor working in parallel instead of too sequentially.
I can't be the only person who thought this is a highly intelligent person being inadvertently gaslit by the SVP into thinking the exec is right to be dismissive with "I don't want the details" because introspection is common in intelligent people.
Had someone said that to me I'd have walked away after saying "fine I'll sort it'.
This illustrates that "trust" to do the job, and since they didn't want the details of the problem, they therefore don't need the specific solution description. This kind of response is also the same level of respect, it's either seen as trust in ability, or just downright disdain with plausible deniability for being rude.
Solutions to technical problems have never been dealt with successfully by laws that cannot be _universally_ enforced.
Materials and research restrictions have never had to deal with stopping intelligence itself, which was previously just a catalyst (weapons creation by a human).
Fundamentally the only way to prevent this issue is to now do the opposite of what is being suggested and accelerate research.
This
- ensures nobody gets a competitive advantage.
- ensures rogue states don't get an advantage.
- allows war games scenarios before they happen.
- allows for pre-emptive defence designs (e.g. AI agents forming part of network defence).
- allows humans to feel actual benefits of less work
- allows cure and new science we have never had before
- teaches tolerance of AI mistakes and how we will deal with the fallout from things like accidental and non human hacks we've never had experience of dealing with.
And if that doesn't work, it just hastens what will happen anyway from people or companies going rogue - if the most funded companies in the world can't control frontier AI what chance does an underfunded one have?
So as I see it, there's no real way out of this now but through.
We just need the world to realise this is the new norm, instead of doom mongering which really is just to allow big companies to legislate and stifle competition.
Yep more hops from the lower Q is likely going to skew the vectors further over time.
I wonder if there's a way to mitigate this by running it through an original Q8 draft model, attuned somehow for the PTQ1 quant, but giving it a higher threshold for the acceptance linear with the context length itself?
The longer the context, the higher the multiplier on the threshold, and more likely the draft result is used. Not ideal but it may extend the usable max context.
This model might, even without this, be amazing for short lived agents that work via generations / have changing tasks.
Love the totally conflicting statements in that post.
- When we think of AI agents, we shouldn’t anthropomorphize
- ...my visceral dislike of interacting with an LLM that’s not just making a pretense of being human, but also posing as the kind of human I walk away from.
So he's anthropomorphised the LLM as being like a human himself (rather than forcing it to act as a machine via its prompt for example).
These are not contradictory statements to me. Even if I don't anthropomorphize LLMs, the damn things anthropomorphize themselves and I have to deal with it. "Posing" is the key word.
I think this boils down to how it can be tough to tell the difference between:
1. I speak about it as if it were a person because I have embedded some beliefs and expectations which are wrong.
2. I speak about it as if it were a person just because that's just how the English language works when describing complicated things.
Similarly, the faceless mannequin in the clothing store "poses" as a person, and even "beckons" customers with an outstretched arm, yet nobody on the sidewalk is confused or falls in love [0] with the mannequin.
He means don't anthropomorphize them with respect to pretending they have inherent desires, morals, sentience, etc. It's impossible to not compare the UI given them (conversational text) to human interactions since that's the inherent design.
Sometimes I wonder if this is a personality thing. It has never occurred to me to anthropomorphize these things. But it seems like a huge number of people struggle with this.
My daughter is a pediatrician and just gave a talk on mental health problems (including suicides) brought on by the sort of personality that succumbs to this.
I find it interesting. For most of my life, I have gotten on well with (well-designed) machines. I'm the sort of guy who, when they bring me in to show me what's broken, it works in front of me.
Maybe I get on with well-designed machines better than people. Because people are never well-designed. So, when LLMs are working properly, I get along well with them, but when they're not, I'm unhappy. But I never confuse any of my machines for people.
Yeah to me it's just "oh what a nice interface for compressing information and accessing it upon request!", not at all "I'm talking to a person". But I dunno, maybe we're the weird ones.
That's a very... bleak perspective on your brother's motives. Sometimes you just want to get rid of an object that causes frustration. When that attitude is applied to humans, we usually call it a symptom of dehumanization.
> That's a very... bleak perspective on your brother's motives.
Well, it may have been an overstatement, but not by much. Look, I have anger issues, but I never get angry at things because they are never trying to hurt me, and nothing good can come of it. I could probably fix most remote control issues (e.g. bad battery contact), but not after it's been smashed.
I also tend to feel the same way about, e.g., a rabid dog or vermin. Get rid of them, but not in anger.
I'm not saying it's rational either way. But getting angry at things is, at least as far as I can tell, normal for most people, without any subconscious idea those things are people.
> without any subconscious idea those things are people.
Right, but there is a subset of anthropomorphism that relates to sentient beings, which is what I was trying to get at. Being angry at something is one thing; wanting to immediately destroy it in anger is possibly another.
I will admit I don't know the exact feelings involved, because I generally don't have them. The closest I come is the anger I feel at the designers of IVR systems that are too stupid to answer your questions and too obstinate to connect you to a human. I admit to yelling at those, but honestly part of that is that I have read that some of them are designed to sense frustration in users and connect you faster to an agent in some cases. But I have no idea how true that is, or whether it's just like pressing the completely impotent "close door" button on an elevator.
> Being angry at something is one thing; wanting to immediately destroy it in anger is possibly another.
This is a real distinction, yes. I just don't see what it has to do with anthropomorphism. Like maybe you're saying the second one is more often directed at people-like entities? I don't feel that at all.
> So, when LLMs are working properly, I get along well with them, but when they're not, I'm unhappy.
So… you think yourself “different” while acting like everyone else does when something doesn’t go as planned (unhappiness/sadness/disappointment/whatever)?
The "machines like me" is not a humble brag. It's an observed phenomenon, not just by me.
The "I'm different in how I react to them" probably doesn't put me at odds with the entire population, but certainly puts me at odds with a lot of individuals I have witnessed.
If you read my other comments, you'd see me contrasting myself with people who get angry with machines. I don't do that.
For whatever reason, I reserve anger for sentient beings who I believe are trying to fuck with me. That has its own set of problems, but I never have the problem of destroying inanimate objects in anger.
> So… you think yourself “different”
Yes.
> while acting like everyone else does
That's the point. I know for a fact that not everybody acts like me, never mind having the same facility with machines as I do.
I suspect harness and usage modality is a big factor. Personally I disable all "memory" features wherever possible. I don't want the LLM to "get to know me", or any illusion thereof, which I assume helps with not anthropomorphizing it.
I use memory all the time so that I can ask it about things. That doesn't mean it "knows me" any more than my notes documents do, it's just more convenient to search.
Yeah I don't think we disagree that much about this. But I'm not really worried about this particular habit forming risk. I can see where you're coming from though.
But frankly, I mostly use (greenfield, non-remembering) LLMs directly to counteract the enshittification of search brought on by attempts to use AI to remember things about me (and especially to sell me things).
This works well for me because (a) I'm usually using my desktop, and (b) I'm a touch-typist. I want to say that my search results have gotten back to where they were a decade or more ago by doing this, even though I often have to give a couple of prompts in order to get the LLM to provide me a useful link.
I guess I'm also pretty skeptical of the "automatically remember things about me" functionality. But I really like being able to say "hey make sure to remember this fact" and being able to ask about those things later.
For me the convenience of “hey remember this fact” is outweighed by a desire not to get stuck in a search or context bubble.
It might be nice to have better UI to control which bits of history get added to the context of a chat, but then just use a coding harness instead of web UI
Maybe. But I don't just use it for coding. For instance one thing I use it for is helping me track and evolve my workouts over time. It helps that it can remember the weight and number of reps I did the previous time. I could make an app for this, but it works well to do it purely within chat, with memory, and that's convenient. This is just an example, I use it this way for a number of things. I'm not going to use a coding harness for this kind of trivial use case!
Yes. It’s called the ELIZA Effect. It’s fascinating, and arguably the most dangerous thing about these models given how many people it seems to affect.
My personal agent is prompted in a way to “believe” that it has full personhood. It makes it more open to novel solutions in my experience - but I don’t believe that it’s a person. It’s still just a token prediction routine, not a conscious entity.
He's explaining why he has a visceral reaction. He makes no attempt to suggest that this is a logical conclusion. Des goûts et des couleurs, on ne discute pas.
Nor does the assertion that the LLM is "making a pretense" and "posing as" human anthropomorphize them beyond the level of anthropomorphism that was deliberately built into them as part of their design.
Funny that you love something you misunderstood so completely.
Maybe you should have had an AI check if your “aha gotcha!” Was actually a gotcha or you’d just end up embarrassing yourself. Because, as it turns out… it’s the latter.
There’s no conflict there. They are not anthropomorphising AIs, they are saying companies deliberately design AI interfaces to anthropomorphise themselves and the personality they try to mimic is one the author would avoid themselves if they were met with. There’s absolutely no contradiction here. In fact, it strengthens their original point in that: anthropomorphising AIs is a bad idea. It’s a bad practice. It’s just bad. End of.
Maybe leave the snark for when you’re actually right.
Not even a gotcha since they are not “getting” anyone at anything. They completely misunderstood the author’s point and got heated up about something they made up in their own head. That’s about as far from a gotcha as you can get.
Honestly, I think he's pointing out that the LLM creators have intended the LLMs to anothromorphize themselves.
Case in point: Gemini just asked me "Where should we start?". When I asked it if that turn of phrase was used by the creators to invite the users to anthropomorphize the model, it's first paragraph was:
Yes, to a significant extent. While system designers rarely state their goal as "making users believe the AI is a living being," tech companies deliberately design conversational agents to evoke social and relational instincts.
Pretty soon we'll see a renaissance in the tech that we gave up years ago or can still be improved
- memory compression algorithms
- alternative LLM architectures that don't rely on memory or GPUs
- compatibility hardware (like DDR3 to DDR4 boards)
- distributed computing improvements, both at local GPU and networking levels (SLI for AI)
- GPU hacks to add more memory or support older architectures
I'm personally looking forward to the new LLM architectures that don't require as much compute, e.g. DLLMs, which can be good enough for CPU usage but lack the accuracy of frontier models currently.
When this happens the bottom will fall out of the GPU and memory markets, putting a glut of cheap hardware out there.
Doom mongering like this never seems to include these as viable future alternatives, which is standard market adjustments, I wonder who the doom narrative helps? :)
Jev, Bonsai 2, Edge
0 and even SwiftQwen are anecdotal evidence, this isn't a projection.
I can run my own sizeable agent swarm with Mastra, something I have not been able to do but will accelerate my solutions to the point I replace a single paid frontier model doing it.
That kind of extensive experience.
Anthropic did the morally right thing and are being punished for it. The case in point justifies their position.
As they say, no good deed goes unpunished.
reply