1,549 karma · joined July 20, 2020
https://github.com/femtomc
Aside but 12 MB is ... large ... for such a thing. For reference, an entire HTTP (including crypto, TLS) stack with LLM API calls in Zig would net you a binary ~400 KB on ReleaseSmall (statically linked).
You can implement an entire language, compiler, and a VM in another 500 KB (or less!)
I don't think 12 MB is an impressive badge here?
If you are not, let me introduce you to the term: a probability distribution.
Just because it has profound properties ... doesn't make it different.
> has all the trappings of the stochastic parrot-style HN-discourse that has been consistently wrong for almost a decade now
Perhaps respond to my actual comment compared to whatever meta-level grouping you wish to interpret it as part of?
> It contains a number of premises that we have no business being confident in. We are potentially witnessing the obviation of human cognitive labor.
What premises? Be clear.
Less worried about memory, more worried about compute speed? Are they obviously related and is it straightforward to see?
As far as I understand RL scaling (we've already maxxed out RLVR), these machines only get better as long as they have expert reasoner traces available.
Having an expert work with an LLM and successfully solve a problem is high signal data, it may be the only path forward?
My prior is that these companies will take this data without asking you as much as they can.
Before, we didn't have a fast (we had to rely on human cognition) way to try problems - even if the techniques and workflows were known by someone. Now, we've baked these patterns into probability distributions - anyone can access them with the correct "summoning spell". Experts will naturally use these systems more productively, because they know how to coerce models into the correct conditional distributions which light up the right techniques.
One question this raises to me is how these models are going to keep up with the expanding boundary of science. If RL is required to get expert behavior into the models, what happens when experts start pushing the boundary faster? In 2030, how is Anthropic going to keep Claude "up-to-date" without either (a) continual learning with a fixed model (expanding context windows? seems hard) or (b) continual training (expensive)?
Crazy times.
Pass for me.
Where did you get this perspective from?
> I thought pi and its tools were supposed to be minimal and extensible. So why is a subagent extension bundling six agents I never asked for that I can’t disable or remove?
Why do you think a random subagents extension is under the same philosophy as pi?
Your blog post says little about pi proper, it's essentially concerned with issues you had with the ecosystem of extensions, often made by random people who either do or do not get the philosophy? Why would that be up to pi to enforce?
This is quite nice — I do think there’s a version of pi’s design choices which could live in a static harness, but fully covering the same capabilities as pi without a dynamic language would be difficult. (You could imagine specifying a programmable UI, etc — various ways to extend the behavior of the system, and you’d like end up with an interpreter in the harness)
At least, you’d like to have a way to hot reload code (Elixir / Erlang could be interesting)
This is my intuition, at least.
if agents continue to get better with RL, what is future proof about this environment or UI?
I think we all know that managing 5-10 agents ... is not pretty. Are we really landing good PRs with 100% cognitive focus from 5-10 agents? Chances are, I'm making mistakes (and I assume other humans are too)? Why not 1 agent managing 5-10 agents for you? And so on?
Most of the development loop is in bash ... so as long as agents get better at using bash (amongst other things), what happens to this in 6 months?
I don't think this is operating at a higher-level of abstraction if agents themselves can coordinate agents across worktrees, etc.
It's a great idea: really neat take on programmability, and can be reloaded while the agent is running without tweaking the harness, etc -- lots of benefits.
`pi` has a great skills implementation too.
I think skills might really shine if you take a minimal approach to the system prompt (like `pi`) -- a lot of the times, if I want to orchestrate the agent in some complex behavior, I want to start fresh, and having it walk through a bunch of skills ... possibly the smaller the system prompt, the more likely the agent is to follow the skills without issue.
It turns out that "most of the bell and whistles" could amount to instructing models how to use tools like tmux
There is barely any magic in the harness, the magic is in the model.
Try it: write your own harness with (bash, read, write, edit) ... it's trivial to get a 99% version of (pick your favorite harness) -- minus the bells and whistles.
The "magic of the harness" comes from the fun auxiliary orchestration stuff - hard engineering for sure! - but seriously, the model is the key item.
This whole game is a bizarre battle.
In the future, many companies will have slightly different secret RL sauces. I'd want to use Gemini for documentation, Claude for design, Codex for planning, yada yada ... there will be no generalist take-all model, I just don't believe RL scaling works like that.
I'm not convinced that a single company can own the best performing model in all categories, I'm not even sure the economics make it feasible.
Good for us, of course.
> TypeScript/Node.js: Better concurrency story thanks to the event loop, but still fundamentally single-threaded. Worker threads exist but they're heavyweight OS threads, not 2KB processes. There's no preemptive scheduling: one CPU-bound operation blocks everything.
This cannot be a real protest: 100% of the time spent in agent frameworks is spent ... waiting for the agent to respond, or waiting for a tool call to execute. Almost no time is spent in the logic of the framework itself.
Even if you use heavyweight OS threads, I just don't believe this matters.
Now, the other points about hot code swapping ... so true, painfully obvious to those of us who have used Elixir or Erlang.
For instance, OpenClaw: how much easier would "in-place updating" be if the language runtime was just designed with the ability in mind in the first place.
Further, I’m not sure this elevates my understanding: I’ve read many posts on this space which could be viewed as analogous to this one (this one is more tempered, of course). Each one has this same flaw: someone is telling me I need to make a “organization” out of agents and positive things will follow.
Without a serious evaluation, how am I supposed to validate the author’s ontology?
Do you disagree with my assessment? Do you view the claims in this content as solid and reproducible?
My own view is that these are “soft ideas” (GasTown, Ralph fall into a similar category) without the rigorous justification.
What this amounts to is “synthetic biology” with billion dollar probability distributions — where the incentives are setup so that companies are incentivized to convey that they have the “secret sauce” … for massive amounts of money.
To that end, it’s difficult to trust a word out of anyone’s mouth — even if my empirical experiences match (along some projection).
What we have instead are many people creating hierarchies of concepts, a vast “naming” of their own experiences, without rigorous quantitative evaluation.
I may be alone in this, but it drives me nuts.
Okay, so with that in mind, it amounts to heresay “these guys are doing something cool” — why not shut up or put up with either (a) an evaluation of the ideas in a rigorous, quantitative way or (b) apply the ideas to produce an “hard” artifact (analogous, e.g., to the Anthropic C compiler, the Cursor browser) with a reproducible pathway to generation.
The answer seems to be that (b) is impossible (as long as we’re on the teet of the frontier labs, which disallow the kind of access that would make (b) possible) and the answer for (a) is “we can’t wait we have to get our names out there first”
I’m disappointed to see these types of posts on HN. Where is the science?
- the SAM coordinates through the dataspace, whereas Goblins is focused on ("point-to-point") message passing
- SAM (as presented) doesn't contain a transactional semantics -- e.g. turns are atomic, and there's no rollback mechanism (I haven't been up to speed on recent work, I do wonder if this could be designed into SAM)
(which, as far as I can tell, also supports capabilities and caveats for security)
Neat work!
This includes standard testing strategies, but also much more general processes
I think of it as steering a probability distribution
At least to me, this makes it clear where “vibe coding” sits … someone who doesn’t know how to express precise verification or feedback loops is going to get “the mean of all software”