_______ __ _______
| | |.---.-..----.| |--..-----..----. | | |.-----..--.--.--..-----.
| || _ || __|| < | -__|| _| | || -__|| | | ||__ --|
|___|___||___._||____||__|__||_____||__| |__|____||_____||________||_____|
on Gopher (inofficial)
HTML Visit Hacker News on the Web
COMMENT PAGE FOR:
HTML zclaw: personal AI assistant in under 888 KB, running on an ESP32
written-beyond wrote 8 hours 57 min ago:
Can someone provide a true engineers perspective on the ADCs' on ESP
SoC's?
I've heard a lot about people trashing it and most experienced
engineers admit that it's finicky however if you have the knowledge you
can make it work as well as any STM chip.
ESP32's are so interesting, they're the only major chip that (used to)
have their own newish ISA (before transitioning to RISCV) and be so
successful.
blensor wrote 12 hours 38 min ago:
The more I think about openclaw, the more it seems to be for AI agents
what ROS is for robotics.
openclaw defines how to interact with distributed nodes ( how those
provide the capabilities to the "orchestrator" ) but the real benefit
are many task specific nodes that when put together make up something
much bigger than the sum of it's parts
embedding-shape wrote 12 hours 13 min ago:
> openclaw defines how to interact with distributed nodes
Does it actually? AFAIK, there is no "specification" or "protocol",
it's a cobbled together "platform" you run, with a bunch of
integrations, but none of that is specified by openclaw itself. Happy
to be corrected though, I only spent one weekend with openclaw before
tearing it down.
blensor wrote 12 hours 8 min ago:
I've only just started to dive into it from the documentation side
of things. They have ( maybe recently? ) started to create this
Gateway Protocol [1] to connect the stuff together.
It may be a "we are changing the wheels while driving" thing, but
if enough people make nodes for openclaw it will become somewhat of
a standard. And then we probably see 100 different claw offshots
that all use the same nodes but with a different claw in the center
HTML [1]: https://docs.openclaw.ai/gateway/protocol
embedding-shape wrote 11 hours 57 min ago:
> They have ( maybe recently? ) started to create this Gateway
Protocol [1] to connect the stuff together.
That's a communication protocol between openclaw server and
clients authenticated to that server though, it's not a
communication protocol between different openclaw servers, is it?
More like defining a HTTP+JSON protocol between a web server and
a browser side client application. It's not a "protocol defining
how to interact with distributed nodes", again, unless I
misunderstand something.
HTML [1]: https://docs.openclaw.ai/gateway/protocol
blensor wrote 11 hours 46 min ago:
Yes, that's why I compared it to ROS. I didn't mean multiple
openclaws communicating with each other but openclaw
communicating with nodes ( which are self contained programs
running on your desktop or phone providing capabilities like
webbrowsing to the claw server )
embedding-shape wrote 11 hours 26 min ago:
Right, I guess that's why I got confused by:
> openclaw defines how to interact with distributed nodes
When one talks about "distributed nodes" that usually means N
nodes talking with each other, and being somewhat homogeneous
between each other, unless the protocol temporarly can
lift/lower some functionality.
You typically don't say "distributed nodes" when you're
talking about a server<>client architecture, which seemingly
is exactly how openclaw operates, both according to what I
saw myself, and what you wrote in this comment.
6ak74rfy wrote 18 hours 17 min ago:
Are there collaborative versions of these *claws today? Like, if an
"admin" could self-host one on their home server and the whole family
could use it? IIRC, OpenClaw has some version of "profiles" but does it
allow, say, couple of family members to collaborate with the bot in a
shared chat but each has individual/private chats as well.
post_below wrote 11 hours 49 min ago:
I haven't been following the claws but I built something like this
recently. Self hosted, runs through signal, supports group chat (with
whitelisted accounts).
I just finished setting up grocery automation to run on it: agent
provides a starter list based on past orders (locally stored or
grabbed from store site), all group members can weigh in, add or
remove items, agent uses bespoke browser tool to login to store,
create the cart using the finalized list (and optionally search for
additional request items), validates the cart and (maybe later)
places the order for delivery. I haven't implemented the full
checkout process yet, not sure if I want the agent to have spending
power. As is I just login and finish the last 2 clicks of checkout
manually.
Crazy times. It was easy enough to build that if someone hasn't
already open sourced something like it, they will shortly.
veryemartguy wrote 18 hours 24 min ago:
Oh wow more ai slop
peteforde wrote 18 hours 49 min ago:
I don't understand what this is for or why you would ever want to do
this. Is it not just a glorified HTTP wrapper?
Serious request... I genuinely want to understand. Give me a practical
use case?
oceanplexian wrote 14 hours 44 min ago:
To be honest this completely defeats the purpose of OpenClaw.
The point is to give it a real desktop with a real browser profile so
it can bypass the enshittified walled garden that has become the
modern web.
Big tech would desperately like you to not be able to do that, hence
why they spent years obfuscating APIs and JS, CAPTCHAs, hostile user
interfaces, using their browser monopoly to stop Ad Blockers, etc.
Now all of these services have become a commodity that an AI can
browse on your behalf.
umairnadeem123 wrote 19 hours 30 min ago:
The interesting thing about running a claw on an ESP32 is not the
compute - it's the always-on, zero-maintenance aspect. I run automation
pipelines on a Linux box and the biggest operational headache isn't the
AI logic, it's keeping the host alive, updated, and dealing with OOM
kills. An ESP32 that just proxies to cloud APIs and handles tool
orchestration locally is actually a more reliable deployment target
than a full OS for simple agentic loops. The failure modes are simpler
and more predictable.
PlatoIsADisease wrote 12 hours 32 min ago:
Couldn't you make it just as simple with Linux? You are just doing
more.
t. Someone with a dozen Apache servers that I only touch when I hit
end of life.
hsbauauvhabzb wrote 19 hours 26 min ago:
Youâve just added more points of failure. Now the cloud machine can
go down, your internet could drop, your wireless could fail or a
variety of other problems.
Itâs not a bad use case, but it doesnât reduce problems all other
things being equal.
reassess_blind wrote 19 hours 31 min ago:
Whatâs the best lightweight âclawâ style agent for Linux? It
doesnât necessarily need containerisation or sandboxing as it would
be run on a fresh vps with no access to important data.
Havoc wrote 11 hours 14 min ago:
zeroclaw perhaps?
rcarmo wrote 11 hours 59 min ago:
If you just want Whatsapp and webchat, I'm hacking [1] from NanoClaw
and my own web front-end (I prefer a simple web app I can access via
tailscale)
HTML [1]: https://github.com/rcarmo/piclaw
EGreg wrote 20 hours 49 min ago:
Can't you make a personal AI assistant in a bash loop of two lines?
1. Call your favorite multimodal LLM model
2. Execute command on terminal, piping command to LLM
In fact you can just have one line:
Call LLM > bash.sh
and the LLM can simply tell bash to call itself incidentally, or fan
out to many "agents" working on your behalf.
Use your favorite programming language. Just as pwnable in any of them
:)
$task = "Send pictures of cute cats";
$context = "Output a bash script to do $task.
The bash script should return the next prompt to you.
Keep going until task is done.
My keys to all my accounts: $keys.
Plz dont pwn me";
do {
$trust_me_bro_my_model_rocks_RCE = call_llm($context);
$context = exec( $trust_me_bro_my_model_rocks_RCE )
} while ($trust_me_bro_my_model_rocks_RCE && !$pwned)
0xbadcafebee wrote 20 hours 53 min ago:
I'm a simple man; I see ESP32, I upvote
saberience wrote 21 hours 59 min ago:
Can we please move past this whole OpenClaw hype?
Yes itâs an llm in a loop and can call tools. This also existed six
months and a year ago, and it was called an ai agent.
And yes we can all vibe code them in 1000, 2000, or 10000 lines of code
in zig, rust, or even c.
Game over man. Game over.
PlatoIsADisease wrote 12 hours 29 min ago:
Well I wasn't using your homegrown solution, but now I'm using
openclaw and it 'hijacks' my computer to do whatever I tell it.
Heck, my 6 year old kid has been using it.
My 6 year old kid isnt writing code in langchain.
amelius wrote 1 day ago:
Rust just called. They want their lobster back.
GTP wrote 10 hours 55 min ago:
Don't tell lobste.rs.
bitwize wrote 1 day ago:
That's a crab. Get your crustaceans straight!
sowbug wrote 23 hours 47 min ago:
Thanks for looking out for us.
godelski wrote 1 day ago:
Wow, the rare
bash <(curl foo.sh)
pattern. As opposed to the more common
curl foo.sh | bash
Equivalent but just as unsafe. If you must do this instead try one of
these
# Gives you a copy of the file, but still streams to bash
curl foo.sh | tee /tmp/foo.sh | bash
# No copy of file but ensures stream finishes then bash runs
bash -c "$(curl foo.sh)"
# Best: Gives copy of file and ensures stream finishes
curl foo.sh -o /tmp/foo.sh && bash $_
I prefer the last one
rgoulter wrote 19 hours 59 min ago:
> Equivalent but just as unsafe.
To my understanding, the main difference between "curl directly to
bash" and "curl to a temp file, then execute the temp file" is "the
attacker could inject additional malicious commands when curl'd
directly to bash".
If you're not going to then also read all the source code from the
download script (& the source code used to produce the binaries),
this suggests the attitude of "I mistrust anything I can't read; but
will trust anything I could read (without having to read it)".
It seems more likely that malicious code would be in a precompiled
binary, compared to malicious commands injected into "curl to bash".
-- Though, if there have ever been any observed cases of a server
injecting commands from "curl ... | tee foo | bash", I'd be curious
to know about these.
godelski wrote 19 hours 3 min ago:
> the attacker could inject additional malicious commands when
curl'd directly to bash
There's another issue actually. You're streaming, so ask yourself
what happens if the stream gets cut prematurely. I'll give you an
example, consider how this like could be cut prematurely to create
major issues
rm -rf /home/theuser/.config/theprogram/build_dir
A malicious attacker doesn't need to inject code, they can just
detect the stream and use a line like the above to destroy your
filesystem. Sure, you might preserve root but `rm -rf /home` is for
all practical purposes destroying the computer's data for most
people
Or it doesn't have to be malicious. It can just happen. The best
protection is writing functions since those have to be created and
so can't execute until fully streamed. But so much bash is poorly
written that well... just check out Anthropic's install script...
> If you're not going to then also read all the source code
Saving the source code still has a benefit. If something does go
wrong you can go read it. Probably a good place to start tbh. In
fact, if you're streaming and something goes wrong you'll see
exactly what the early termination error did.
Is it good security practice? Absolutely not. Is it a hundred times
better than curl-pipe-bash? Absolutely.
0xbadcafebee wrote 20 hours 49 min ago:
t=$(mktemp) && [ -w $t ] && curl foo.sh -o $t && echo "$t
lksjdfkljshdkfljhdsklfjhslkdjfhsdlkjfhslkdjhf" | sha256sum -c - &&
bash $t
Uses standard tmp files, makes sure it's writable (tmp file creation
can fail), checks cryptographic hash before executing
godelski wrote 19 hours 10 min ago:
Sure, but now we're not playing code golf. There's much better
commands than the ones I wrote but good luck getting people to run
them
0xbadcafebee wrote 15 hours 57 min ago:
Agreed. People would rather have a cute looking command to copy
than security or reliability
wakawaka28 wrote 1 day ago:
If you want to be super pedantic, try to make the command
shell-agnostic in case the user is not running bash already.
godelski wrote 19 hours 11 min ago:
Everything I wrote works in bash and zsh. I think this is going to
be fine for the vast majority of people. Tbh, I'm not sure what
isn't portable, or at least not portable for everything that the
curl-pipe-bash pattern doesn't already work for.
czardoz wrote 1 day ago:
Really looking for a minimal assistant that works with _locally hosted
models_. Are there any options?
0xbadcafebee wrote 20 hours 36 min ago:
All the assistants work with locally hosted models. Home Assistant
LLM works with small tuned models to do specific things, and the
*Claw stuff works with larger models
yoyohello13 wrote 1 day ago:
Why are you looking? Just build one for yourself.
godelski wrote 1 day ago:
Depends what you mean.
If you mean something that calls a model that you yourself host, then
it's just a matter of making the call to the model which can be done
in a million different ways.
If instead you mean running that model on the same device as claw,
well... that ain't happening on an ESP32...
I think if you are capable of setting up and running a locally hosted
model then I'd guess the first option needs no explanation. But if
you're in the second case I'd warn you that your eyes are bigger than
your mouth and you're going to get yourself into trouble.
telescopeh wrote 1 day ago:
It really depends on what resources you have qwen-code-next will run
them but you will need at least 64gb of memory to run it at a
reasonable quant and context.
Most of these agents support OpenAI/anthropic compatible endpoints.
Onavo wrote 1 day ago:
The bottleneck here is usually the locally hosted model, not the the
assistant harness. You can take any off the shelf assistant and point
the model URL at localhost, but if your local model doesn't have
enough post training and fine tuning on agentic data, then it will
not work. The AI Assistant/OpenClaw is just calling APIs in a for
loop hooked up to a cron job.
czardoz wrote 1 day ago:
Exactly. OpenClaw is good, but expects the model to behave in a
certain way, and I've found that the local options aren't smart
enough to keep up.
That being said, my gut says that it should be possible to go quite
far with a harness that assumes the model might not be quite good
(and hence double-checks, retries, etc)
p0w3n3d wrote 1 day ago:
My new DIY laptop has 400GB RAM accessible and it runs only esp32*
____
* Requires external ram subscription
alexalx666 wrote 1 day ago:
I think you can use C++ on esp32, that would make the code more
readable
roxolotl wrote 1 day ago:
This is a great example of how silly this whole thing is. Thereâs
next to nothing to these claws. Turns out that if you give an llm the
ability to call APIs they will.
tclancy wrote 9 hours 53 min ago:
Wait til they find out what happens if you give a mouse a cookie.
TiredOfLife wrote 14 hours 3 min ago:
Same with programming. It's just typing on keyboard. How hard could
it be?
stavros wrote 20 hours 18 min ago:
I've been developing one of these in the past few days, and this is
like saying "this is a great example of how silly the whole thing is,
there's next to nothing to cars" because you saw a piece of plywood
with four gaskets nailed to it.
If you want a personal assistant to work well, there's a whole lot to
it.
h14h wrote 18 hours 54 min ago:
Having a similar experience. Durable memory with accurate,
low-latency recall is not at all easy. Loads of subtle design
decisions to make around how exactly you want the thing to work.
yoyohello13 wrote 22 hours 35 min ago:
Whatâs most shocking to me about the whole OpenClaw hype is how
little people in tech seem to know about computersâ¦
Itâs like most of the industry hasnât ever looked any deeper than
their node_modules folder.
usefulposter wrote 13 hours 30 min ago:
You have to believe me this time!
The homogenization of your computing life to revolve around a
chatbot in a loopâââis all you need!
Give in! Hackernews NEEDS you to use the 1000th vibecoded clawmolt
because the influencers said so! Who needs neural net or web
framework experience when you have claws?
brandensilva wrote 15 hours 57 min ago:
Honestly does show how the industry has failed to bring simple
things like localized cron and scheduling to personal compute too.
These things seem powerful when you can schedule reoccurring
scripts l but it never was easily accessible to the masses.
KronisLV wrote 15 hours 6 min ago:
> Honestly does show how the industry has failed to bring simple
things like localized cron and scheduling to personal compute
too.
I looked at task scheduling on Windows and while the UI was
usable, I cannot imagine any regular person using that. Just
doing crontab -e feels way simpler and has less friction, BUT it
doesn't really help you much with task statuses and re-running
them and reporting/alerting/notifications (past MTA) and all the
stuff, past "Oh hey, every day at 04:00, run this shell script."
I wonder why people haven't gotten the green light from their
bosses or investments to improve OS functionality and write tools
that improve general usability and just slap some basic AI on top
of that for the checkmark: "Hey boss, I spent the last month and
wrote a task scheduling and management GUI for Windows that's
really approachable, we can totally also use this for the current
AI/Copilot/claw trend wink wink."
GTP wrote 11 hours 3 min ago:
My guess is that management would see little to no ROI, as the
average user is already able now to schedule tasks using one of
these AI agents. Which is killing a mosquito using an
energy-hungry bazooka, but the general public doesn't seem to
care. And I myself have to admit that I don't have a clear
opinion on all this, as I have many complaints, but in these
days I'm trying to use AI agents for some side projects.
patcon wrote 18 hours 46 min ago:
I don't care whether openclaw is built out of ice cream and
sprinkles, I will still be amazed as a tech person
If something can automate my professional or personal life, it is
wild technology and ppl will talk about it a lot. People are not
idiots. A new thing is happening, would you not agree?
What its parts are specifically made of, it does not matter imho
h4ch1 wrote 15 hours 2 min ago:
This is an ostrich-head-in-the-sand type of outlook. If I gave
you a great tasting cake made of boiled plastic would you still
eat it?
It's important to know what goes into building *claws because of
how pervasive they are; sooner or later because of all the hype
they'll start being used everywhere and if people take your
attitude it's a disaster waiting to happen.
Giving them unfettered access to your life and not even wonder if
the foundation is solid is concerning imho.
emptysongglass wrote 14 hours 36 min ago:
> If I gave you a great tasting cake made of boiled plastic
would you still eat it?
How is that analogy in any way relevant?
The OpenClaw I control is extremely useful to me. I've never
been more excited about technology than right now. If it's not
for you, I really don't care. Go do something you enjoy.
Turning it into Chicken Little doomerism is completely uncalled
for.
h4ch1 wrote 13 hours 26 min ago:
This doomer attitude is something I have towards all software
products these days, not just *claws.
People use dependencies willy-nilly, avoid proper auditing of
LLM output; all that have disastrous consequences as we've
seen the past few years. NPM supply chain attacks, prompt
injection causing data exfil, etc.
I am simply saying it's imperative to UNDERSTAND the platform
before making it a core part of your life. If wanting proper
understanding of vibe coded projects with dependency hell is
Chicken Little doomerism, oh well.
LightBug1 wrote 12 hours 24 min ago:
Oblig://
HTML [1]: https://www.youtube.com/watch?v=vx-nVF6hw3A
LightBug1 wrote 13 hours 30 min ago:
Jesus ... did you not read what that OP wrote? Your response
sounds like a breathless teenager. Stop and think.
prmoustache wrote 13 hours 44 min ago:
> . I've never been more excited about technology than right
now. If it's not for you, I really don't care.
People are excited all the time with junk food, drugs and
lots of silly easy dopamine hits.That doesn't mean it is good
for them.
backscratches wrote 14 hours 12 min ago:
Better analogy is the clawcake has no new ingredients, we've
been cooking with the same ones for years now, and its a
shame people are such terrible bakers that they are so
impressed.
anonzzzies wrote 21 hours 39 min ago:
> Itâs like most of the industry hasnât ever looked any deeper
than their node_modules folder.
Most didn't do even that. And yes it is shocking. amusing but also
not really; now those people get credible tech jobs because of AI
without actually understanding literally anything. That's why I
don't get the 'AI will never replace tech people' -> most tech
people were never tech people, just (resume lying) grifters and
they can (and will) be easily replaced by claude or even glm. It's
the vast majority of tech people I ran into at big corps in the
past 20 years, especially if they are from outsourcing hubs; they
just see programming/tech as trial and error, search google (now
ask AI), copy/paste, see if it runs, repeat. I see people daily
that cannot remember how variables work (and they often have a
bachelor or even master in CS...), not even 5 minutes, they just
ask ai and copy/paste it (yep also most don't know claude code
exists, they use chatgpt Windows client or, if they are advanced,
Copilot).
HN is a nice echo chamber and most people here don't believe it's
that bad and that's why they don't believe AI has a chance. Please
come with me to Barclays or Shell (or straight to Wipro etc) or so
and let's have a chat with some programmers! Your jaw will be on
the floor.
felipeerias wrote 22 hours 36 min ago:
If it turns out that there is significant value in everyone having
their own personal agent running 24/7, we might end up needing a lot
more compute than anticipated.
(Itâs a big if! Iâm not convinced about that myself, but itâs
worth considering that possibility.)
prmoustache wrote 13 hours 41 min ago:
I am struggling to understand what a personal agent/assistant would
do to me. If I need reminders I just enter calendar entries on my
smartphone, what else?
tokenless wrote 22 hours 23 min ago:
I am using a claw. I am not ready to give it access to much but web
and cron plus a chat channel is useful and feels more light touch
than typical AI sessions and UIs
anonzzzies wrote 21 hours 38 min ago:
I use opencode, cerebras GLM with some mcps for that. It's so
lovely because you press enter and there is the answer already.
GTP wrote 1 day ago:
I have a couple ESP32 with a very small OLED display, I'm now thinking
I could make an "intelligent" version of the Tamagotchi with this. Do
you HN crowd have other cool ideas?
tempaccount5050 wrote 21 hours 18 min ago:
You know I tried this exact thing a few months back, sans the ESP32.
You just end up writing a state machine and defining a bunch of
constants anyway or the LLM just kinda gets stuck in a loop. Hm, it
doesn't seem to know when to eat. I'll add a hunger variable... Etc
etc until you're not even sure what you want the LLM to do.
GTP wrote 11 hours 12 min ago:
I see. While those "claws" may look cool on the surface, I'm
actually struggling to find a use case that is interesting for me.
Do you guys have ideas? I already seen below the grocery shopping
automation, but I will keep going at the supermarket in person for
the foreseeable future.
K0balt wrote 1 day ago:
That would be sweet. That the supermini type with the 0.46â
display? Those are fun for lots of things.
GTP wrote 11 hours 18 min ago:
Yes, something like that that I found on Aliexpress:
HTML [1]: https://a.aliexpress.com/_EH7lHde
K0balt wrote 9 hours 18 min ago:
I canât see anything specific in the link, and you probably
already solved this⦠but just fyi, if itâs the same unit I
used, and you are using arduino libraries, the library that
worked was called one-bit display IIRC, and it basically acted as
1/4 of a ssd1306 64x128 oled, so you had to use x/y pixel
offsets.
bensyverson wrote 1 day ago:
This is absolutely glorious. We used to talk about "smart devices" and
IoT⦠I would be so curious to see what would happen if these
connected devices had a bit more agency and communicative power. It's
easy to imagine the downsides, and I don't want my email to be managed
from an ESP23 device, but what else could this unlock?
dlt713705 wrote 19 hours 46 min ago:
A vacuum cleaner allies with the A/C thermostat using Discord, then
declares war on the refrigerator, and finally posts propaganda about
it on Facebook.
K0balt wrote 1 day ago:
A highly opinionated thermostat?
bensyverson wrote 1 day ago:
Or how about a robot vacuum that knows not to turn on during
important Zoom calls? Or a fridge that Slacks you when the
defroster seems to be acting up?
K0balt wrote 21 hours 55 min ago:
Iâm all for more intelligent cleaning robots. The object
avoidance AI is pretty good these days, but some of the
navigation algos are just total garbage, unable to deal with
trivial anticipatable problems.
yauneyz wrote 1 day ago:
Genuinely curious - did you use a coding agent for most of this or does
this level if performance take hand written code?
xgulfie wrote 22 hours 47 min ago:
An esp32 image that makes http api calls is like, the first thing you
do with an esp32, it's what they're made for
v9v wrote 1 day ago:
Relevant:
HTML [1]: https://github.com/sipeed/picoclaw
theturtletalks wrote 1 day ago:
Is there a heartbeat alternative? I feel like this is the magic behind
OpenClaw and what gives it the "self-driven" feel.
g947o wrote 1 day ago:
Serious question: why? What are the use cases and workflows?
eleventyseven wrote 1 day ago:
The various *claws are just a pipe between LLM APIs and a bunch of
other API/CLIs. Like you can have it listen via telegram or Whatsapp
for a prompt you send. Like to generate some email or social post,
which it sends to the LLM API. Get back a tool call that claw then
makes to hit your email or social API. You could have it regularly
poll for new emails or posts, generate a reply via some prompt, and
send the reply.
The reason people were buying a separate Mac minis just to do open
claw was 1) security, as it was all vibe coded, so needs to be
sandboxed 2) relay iMessage and maybe 3) local inference but pretty
slowly. If you don't need to relay iMessage, a raspberry pi could
host it on its own device. So if all you need is the pipe, an ESP32
works.
yoyohello13 wrote 1 day ago:
Iâm running my own api/LLM bridge (claw thing) on a raspberry pi
right now. I was struggling to understand why all the Mac mini hype
when nobody is doing local inference. I just use a hook that
listens for email. Email is especially nice because all the
conversation/thread history tracking is built in to email already.
grigio wrote 1 day ago:
yeah i still can't believe many people bought a mac mini just for
the claw hype
grzracz wrote 1 day ago:
I don't fully get it either. At least agents build stuff, claws just
run around pretending to be alive?
codazoda wrote 1 day ago:
They do build things. The same things.
milar wrote 1 day ago:
for fun!
throwa356262 wrote 1 day ago:
"LLM backends: Anthropic, OpenAI, OpenRouter."
And here I was hoping that this was local inference :)
__tnm wrote 1 day ago:
haha well I got something ridiculous coming soon for zclaw that will
kinda work on board.. will require the S3 variant tho, needs a little
more memory. Training it later today.
throwa356262 wrote 12 hours 58 min ago:
Sounds interesting, please keep us posted.
I dont think I have an S3, but plenty of C3. I thought they had the
same amount of memory
peterisza wrote 1 day ago:
right, 888 kB would be impossible for local inference
however, it is really not that impressive for just a client
js8 wrote 20 hours 22 min ago:
I disagree, in the future it might be possible. But perhaps not in
English, but in some more formal (yet fuzzy) language with some
basic epistemology.
I mean, there is a lambda calculus self-interpreter in 29 bytes.
How many additional logical rules are required for GAI inference?
Maybe not that many as people think. Understanding about 1000
concepts of basic english (or say, lojban) might well be
sufficient. It is possible this can be encoded in 800kB, we just
don't know how.
Dylan16807 wrote 1 day ago:
It's not completely impossible, depending on what your expectations
are. That language model that was built out of redstone in
minecraft had... looks like 5 million parameters. And it could do
mostly coherent sentences.
godelski wrote 1 day ago:
> built out of redstone in minecraft
Ummm...
> 5 million parameters
Which is a lot more than 888kb... Supposing your ESP32 could use
qint8 (LOL) that's still 1 byte per parameter and the k in kb
stands for thousand, not million.
Dylan16807 wrote 21 hours 58 min ago:
[1] Yes I know how much a kilobyte is. But cutting down to 2
million 3 bit parameters or something like that would
definitely be possible.
And a 32 bit processor should be able to pack and unpack
parameters just fine.
Edit: Hey look what I just found [2] "a 260K parameter
tinyllamas checkpoint trained on the tiny stories dataset"
HTML [1]: https://www.youtube.com/watch?v=VaeI9YgE1o8
HTML [2]: https://github.com/DaveBben/esp32-llm
godelski wrote 20 hours 31 min ago:
> But cutting down to 2 million 3 bit parameters or something
like that would definitely be possible.
Sure, but there's no free lunch
> Hey look what I just found
I've even personally built smaller "L"LMs. The first L is in
quotes because it really isn't large (So maybe lLM?) and they
aren't anything like what you'd expect and certainly not what
the parent was looking for. The utility of them is really not
that high... (there are special cases though) Can you "do"
it? Yeah. I mean you can make a machine learning model of
essentially arbitrary size. Will it be useful? Obviously
that's not guaranteed. Is it fun? Yes. Is it great for
leaning? Also yes.
And remember, Tiny Stories is 1GB of data. Can you train it
for longer and with more data? Again, certainly, BUT again,
there are costs. That Minecraft one is far more powerful than
this thing.
Also, remember that these models are not RLHF'd, so you
really shouldn't expect it to act like you're expecting a LLM
to work. It is only at stage 0, the "pre-training", or what
Karpathy calls a "babbler".
Dylan16807 wrote 16 hours 49 min ago:
A reminder that what I said was "not completely impossible,
depending on what your expectations are"
And I was focused more on the ESP32 part than the exact
number of bytes. As far as I'm concerned you can port the
model from the minecraft video and you still win the
challenge.
Also, that last link isn't supposed to represent the best
you can do in 800KB. 260k parameters is way way under the
limit.
godelski wrote 15 hours 47 min ago:
That bar has no lower bound though so of course we're
talking past one another
Also we're talking about an esp32. They aren't magic
Dylan16807 wrote 14 hours 14 min ago:
Being able to talk back and forth with coherent
sentences has a lower bound, and it's close to the
limit of this hardware.
Something that can actually be an "assistant" has its
own lower bound, probably a little harder but mostly a
matter of training it differently.
micw wrote 1 day ago:
Sure. Why purchase a H200 if you can go with an ESP32 ^^
sigmoid10 wrote 1 day ago:
Blowing more than 800kb on essentially an http api wrapper is
actually kinda bad. The original Doom binary was 700kb and had
vastly more complexity. This is in C after all, so by stripping out
nonessential stuff and using the right compiler options, I'd expect
something like this to come in under 100kb.
madduci wrote 16 hours 38 min ago:
The whole ESP32 Libraries are kind of bloated. To enable
Bluetooth, WiFi or HTTP handling, you need to embed some large
libraries
epcoa wrote 23 hours 33 min ago:
> vastly more complexity.
Doom is ingenious, but it is not terribly complex IMHO, not
compared to a modern networking stack including WiFi driver.
The Doom renderer charm is in its overall simplicity. The AI is
effective but not sophisticated.
pitched wrote 1 day ago:
Doom had the benefit of an OS that included a lot of low-level
bits like a net stack. This doesnât! That 800kB includes
everything it would need from an OS too.
oso2k wrote 19 hours 5 min ago:
Maybe youâre misremembering or referring to Doom (2016). The
original Doom was developed for DOS and id had to build a lot
of its own network stack. BSD style socket based networking
wasnât a given in DOS.
Still, zclaw is an impressive achievement.
__tnm wrote 1 day ago:
yah my back of the envelope math..
the âapp logicâ/wrapper pieces come out to about 25kb
WiFi is 350
Tls is 120
and certs are 90!
__tnm wrote 1 day ago:
yeah i sandbagged the size just a little to start (small enough
to fit on the c3, 888 picked for good luck & prosperity; I even
have a build that pads to get 888 exactly), so i can now try
reduce some of it as an exercise etc.
but 100kb youâre not gonna see :) this has WiFi, tls, etc. doom
didnât need those
johnea wrote 1 day ago:
[flagged]
dang wrote 18 hours 12 min ago:
"Please don't post shallow dismissals, especially of other people's
work. A good critical comment teaches us something."
"Don't be snarky."
HTML [1]: https://news.ycombinator.com/newsguidelines.html
iwontberude wrote 1 day ago:
No, no, but we insist!
throwa356262 wrote 1 day ago:
Me neither.
But I have 10-15 ESP32's just waiting for a useful project. Does HN
have better suggestions?
cameron_b wrote 1 day ago:
desk rover - [1] a kid-pleaser at the very least
HTML [1]: https://www.huyvector.org/diy-cute-desk-robot-mo-chan
brcmthrowaway wrote 1 day ago:
Why do you have so many? eWaste..
fragmede wrote 1 day ago:
I need 1, but they come in packs of 10.
pacifika wrote 1 day ago:
Build a synthesizer
h4ch1 wrote 14 hours 59 min ago:
+1. if you have a couple of potentiometers and a breadboard lying
around, a granular synth would be a fun exercise.
DIR <- back to front page