nhecker

You get a reasoning trace from Opus 5.5?

Thanks. That's https://lab401.com/products/chameleon-ultra and https://shop.m5stack.com/collections/controllers/CARDPUTER as far as I can tell... I might look into those, especially as my f0 wears out. The f1's feature-set doesn't interest me. Nor does it's price point, which I imagine falls into the "ain't cheap" category.

When folks ask me what it is, my answer is 'an electronic Swiss Army knife', which is the best analogy I've been able to come up with so far. The Flipper Zero isn't necessary the best tool for any one specific job, but it's a really good tool for doing a lot of different jobs acceptably.

WiGLE WiFi is probably something you might find interesting, then. Or possibly Pwnagotchi if you're looking for a more specific use case, to which another comment has linked.

If we're playing the risk game and we assume that this capability exists and is exploitable, then yeah your chats would be just as exposed to the capability as agentic output. So you might be told that the sky is red instead of blue, but only if you previously asked about the freezing point of water.

>will search for backdoors

Or place new backdoors!

But I still don't see the argument for using closed-weights models. Why should I trust a closed model where I can see none of its weights, activations, or reasoning more than a model where I CAN see its weights, activations, and reasoning?

Haven't smallish (16-32 GB VRAM) open-weights models been getting better and more capable over the last few years? Seems like what you can run on 'a laptop' today beats the pants off of whatever you could run on that same laptop two years ago.

> I chose credits, because by the time i spend 10k i suspect the market will have dramatically shifted in requirements, compute, etc.

I agree, at least for inference credits. But anecdotal counterpoint: about ten years ago I put together a system using about $1600 worth of new and used components. Including electricity that TCO has been something like $14/month. I'd be hard pressed to find 128GB of RAM, 12 cores of x86_84, and 40 TiB (raw) worth of disk, and unlimited ingress/egress traffic for rent at that rate. That's what, $0.02 an hour, rounding up?

Yes, a hyperscaler's rent-a-computer would get faster and be more energy efficient. Possibly more carbon efficient over it's lifetime, although I have no way to quantify that. But in raw dollars it would have costed me absurdly more to rent. On the other hand, my machine has sat cold and unused for multiple months when I didn't need it to be running, and I don't consume all of it's resources around the clock. So maybe I could have gotten away with a lesser amount of local compute in a Raspberry Pi or similar, and an equally small cloud box for $5 a month or something. I dunno. I don't even own a laptop or desktop, and try to use 'cloud' services whenever practical. (Edit: always giving priority to privacy in those circumstances; only data I don't care about goes up.)

There are tradeoffs. I don't use my local server for LLM inference, it's far to inefficient for that; so I'll happily pay OpenRouter per token as needed. So I do think that hosted compute for rent does have a purpose. I guess all that's to say that I don't see it in black and white; there are times to invest in local hw and times to ride on the optimized bleeding edge of what other companies are willing to host in big datacenters.

This turned into a bit more of a ramble than I intended.

What makes a ROM hard? I'm ignorant of the details of chip design, but I'd expect just a large array of essentially open or closed connections. Is that incorrect?

Thanks! I'd forgotten that 1.1 and 1.1.1 resolve. I'll not infrequently ping 1.1.1.1.1 with overzealous typing. That will, alas, not resolve.

Anecdotal, but I've heard this too. I just tried with variations of your same prompt on arena.ai, across three different battles (i.e., six LLMs answered, in total.)

Each provided an identity in the first turn, something that they won't do as readily if asked in plain English, and in each case the answer matched the model ID as disclosed by arena.ai after voting -- except in cases where the model ID was a masked/hidden one and then I just had to take it on faith that the model was what it said. (I didn't have much to vote on, but I ended up voting for the answers I felt provided the style, content, and length I was expecting.)

Just chiming in that I've never had this experience with them. I set, hit, and depend on those limits regularly. I would be wildly interested in reading what their support analysis had to say about this situation, if such a postmortem was done.

In a way, this makes me think of Horde AI, but one level of abstraction away. With Horde you trade idle inference for credits you can redeem in the future. So I'd put up credits to send you input tokens, you'd send the response, and you'd get credits to spend for yourself later.

I'm still looking through your docs, but I'm curious how e.g., you'd prevent my job you run on your idle agent from contaminating your agent's memory or history, or in the opposite direction how you'd guarantee that your idle agent's memory or skills or system prompts or whatever aren't corrupting or impacting my job?

(Edit: future is only spelled with one f)

Reads much like it was LLM-written, but the facts seem to indicate it was a single-vehicle crash at 73mph in a 70mph zone where the vehicle veered off the highway in to a guardrail. The sole occupant was wearing a seat belt at the time of the accident. I was expecting speeding or some other kind of reckless behavior, but none was evident from what I could tell. It's sad that this was fatal.

Keep your hands on the wheel, y'all. :-(

I really like this website design. That's all I came here to say :) Well, and keyboard mods are always fun to read about (I too know the struggle of a PS2 keyboard, hah) but I really just dig the text-first layout.

Homepage is cool too.

That sounds like large amounts of context (maybe from memory or history?) are chewing through your quota, context that hadn't yet had a chance to build up in your newer account. That's just a guess, I'm not an account holder there or anything so I can't try to re-create it on my end. You could try to erase history on your old account too, too see if it makes a difference. 'Course, then you'd lose all that history!

I feel much the same way. Actions speak louder than words, and when was the last time that OpenAI published open weight LLMs? Anthropic?

I agree with the root cause generally, but balk at the 'if we had' phrasing. We still can have these things, it just takes our own time and energy to get out and be with people face-to-face. Nothing about that is impossible. It is not as easy as sitting behind a keyboard though.

Another path may be to add the business to OpenStreetMap first; I get the impression that Google will scrape changes to OSM and add them to it's dataset. (I've seen several small changes I've made only to OSM get propagated to GMaps a few weeks later. Possibly it's a coincidence, but when the features are exactly the same shape as the ones I drew in OSM, I get suspicious.)

Between the streetcomplete and everydoor apps, it is very very easy to do almost any OSM map editing on a mobile. Osmand (the offline map & nav app) is also useful for some edits when I don't want to leave that app.

Oh, for the simpler days when not everything required an app.

That's a fun li'l read, thanks for sharing.

I'm sure there's a way to pay for Hermes Agent. https://portal.nousresearch.com/cloud

> People can change access or disconnect a service whenever they want. People can also opt out of their interactions being used to train Meta’s AI models.

-- https://about.fb.com/news/2026/09/introducing-muse-personal-...

I normally don't appreciate snarky comments on HN, me being but a simple curmudgeon wanting pure information and the occasional opinion. but this one actually makes a good point, even if it's not stated explicitly. Reading this book is a wonderful experience because of the way it's written. Reading the plot summary (from an AI or otherwise) yields nothing but the dry husk of that experience.

https://www.youtube.com/watch?v=pOWwx7nAPvk&t=45s shows wind turbine bolts evaluated in that same way. (Mind the NN voiceover.)

I've been watching the number of open PRs on this repo over the past few months and it just keeps climbing. As far as I can tell, they're merging about as fast as they can, several times an hour around the clock.

I'm sure there are other examples of repos that move this quickly out there, but I guess this is the first one that I've become aware of. On one hand it feels good that opensource software gets this much attention; on the other hand it seems like it must (at times) feel like a Sisyphean task to keep up both on the dev side and to keep the client-side software up-to-date.

I've had the same thoughts, and I think you articulated them pretty well; maybe better than I could have. I.e., for me it's something like: I'm much more willing to share if I know the party I share with will share back, in turn.

I know it's more complicated than that, with economics and privacy factors involved. But if I operate on the assumption that real privacy is real hard I'd rather just be guarded and careful with my prompts and whatever output I give to the LLM and expect that it will be training on that, rather than spill all my darkest secrets to some other LLM provider that pinkie-promises privacy only to leak it publicly later, anyway.

But you /could/ make a pruned plum tree plumb.

Yes, those tokens cost money|credits|whatever too.

On either side of the front wheel is a perfectly reasonable place to carry cargo. I think I'd have taken more issue with the spokes, or at least that's what stood out to me. The chain is indeed nice, however.