jas8425

I really want to like these "agentic assistant" tools, because I feel like the problem they claim to solve is real: give me an interface across desktop and mobile to a persistent backend where I can set up agents (using natural language) to do... whatever I want. Deep market research? Building + hosting a browser game? Checking my email? etc.

But after trying both Hermes and OpenClaw, it feels like they both... miss the point? Last time I tried OpenClaw it wanted to download something like 11 GB of local models to do... something (embeddings for memory indexing or chat labeling/classification maybe?) which my sorry old 16gb M1 is certainly not capable of running.

Hermes seems to suffer from the same problem: why do I need to download (and then immediately disable to avoid confusing my poor "agents"... a concept which I also feel like way too many tools fundamentally misunderstand) skills for managing Spotify playlists or pokemon or minecraft in order to run the thing? (I acknowledge that they cleaned some of this up in a recent release, so maybe this isn't as bad as it was when I last tried it)

WRT "agents"... can someone explain to me why there's so much effort put into naming agents and giving them personalities? An "agent" is simply a separate context window with different prompting (itself written by the spawning/parent "agent") that's specific to a partial slice of the task you're trying to solve. If you have to write their prompt ahead of time that defeats the whole purpose of a programmable, autonomous subagent, doesn't it?

As an outsider to the fascinating world of terminal emulators... can you explain why this might be? Rather, what about `libghostty` would be off-putting vs `libtermengine`?

Just that it's a specific "product"-y sounding name? Would you also be concerned about "libwayland" vs "libcompositor"? Genuinely curious: this seems like an insightful question, I just don't follow the reasoning.

Is anyone else put off by the AI-sounding text? Two things that give it away for me are the excessive use of punctuation-emphasized sentence fragments ex:

> No clicking. No undo. Just recompile.

> That's our mascot. Entirely CSG.

> No garbage collection pauses. No floating point surprises from a scripting layer.

And worst of all, the dreaded "and/but honestly":

> But honestly, the main reason is the toolchain.

Am I misreading things?

So you're saying that instead of receiving $200/month worth of food, poor citizens should go into debt to the central bank by $200 every month? How would that be a better approach? Personal debt is already a huge burden, this seems predatory.

I'm 95% sure that it does not, at least as far as the the essence of any arbitrary concept does or doesn't relate to the "length of text". Theoretically you should just as easily be able to add or subtract embeddings from a book as a tweet, though of course the former would require more computation than the latter.

If embeddings are roughly the equivalent of a hash at least insofar as they transform a large input into some kind of "content-addressed distillation" (ignoring the major difference that a hash is opaque whereas an embedding has intrinsic meaning), has there been any research done on "cracking" them? That is, starting from an embedding and working backwards to generate a piece of text that is semantically close by?

I could imagine an LLM inference pipeline where the next token ranking includes its similarity to the target embedding, or perhaps instead the change in direction towards/away from the desired embedding that adding it would introduce.

Put another way, the author gives the example:

> embedding("king") - embedding("man") + embedding("woman") ≈ embedding("queen")

What if you could do that but for whole bodies of text?

I'm imagining being able to do "semantic algebra" with whole paragraphs/articles/books. Instead of just prompting an LLM to "adjust the tone to be more friendly", you could have the core concept of "friendly" (or some more nuanced variant thereof) and "add" it to your existing text, etc.

This poses an interesting question: how to get rid of that energy? Once the battery is full, you'd need a pretty substantial resistor bank to dissipate all of that excess energy. I wonder if they intentionally start the day at a low state of charge and end fully charged? The idea of a "reverse charger" to discharge the battery into the grid every night would be hilarious and not entirely unfeasible.