solarkraft

Since agents are more often right than they are wrong, combining them can increase reliability.

> Agents can retrieve documents and query databases, but reliability gates agentic analytics

this reads like an attempt at poetry

That’s not what they’re doing either though. If I’m correctly informed, they are still running bog-standard Google searches under the hood and letting the LLM find the relevant stuff in them. It’s a bit of a godsend not having to do it yourself since the original results’ quality tanked so much. And kinda elegant from an algorithmic perspective actually.

It’s still not as good as actually searching the web semantically as I used to often find myself wanting to get rid of results that sounded vaguely like my search term but actually completely missed the point, but it makes it somewhat bearable again.

Oh yeah. I recently started with engaging with a vendor that has good SEO and I would just instinctively skip past their search result because it was one of those expanded ones.

Consider me confused. Isn’t quantizing a value, e.g. reducing the number of bits in an int, a form of discarding information from that value? Isn’t your comparison just about discarding information on another level (discarding entire values)?

https://handy.computer has Nemotron Streaming and it works fabulously well on any Mac down to the base M1. I also built a little Deepgram compatible API into it that I use with a phone app.

> Transcribe is built for recorded audio, not live streaming.

Bit unfortunate! Nemotron Streaming in Handy has basically solved dictation for me. It recognizes German mixed with English terms and brand names really well.

You see, I will release a really good model some day. It’ll get like 300 on AA.

I heard Argon was bad at executing plans but good at making them. Sounds like a potent combination to me.

They’re not the only ones to do it, so I’m guessing their idea is to double-dip attention.

For me it also works the opposite way, when it actually comes out eventually it’ll already be old news.

> I still broadly just prefer working with Muse Glimmer

Welcome to our very exclusive club of now two people! Ever since it came out I’ve been trying to find a better model, but just can’t.

Please elaborate. Conventional thinking goes that emergency plans must be exercised to make sure they’re actually effective.

Everyone here talking about how hard it is to get one.

My 2018 Samsung TV has smart features, but is not internet connected. It turns on quickly and right to the selected source - no brain cells harmed having to look at ads. It slightly bugs out sometimes, but I assume this to be a general issue.

It’s incredible that you can’t sue them over this.

Every once in a while, public opinion actually gets through to them. It only took months to years of clowning and a viral post for them to finally fix their corner inconsistencies.

For what it’s worth, Muse Glimmer is by far the most usable model on my 32GB M1 Max. It gets stuff done very competently while Qwen never gets done reasoning.

I have what feels like 100 prompts to make it stop doing this. I celebrate whenever it actually works.

The model seems to be so post-training fried that almost nothing works.

After months of using the thumbs-down button didn’t help, I eventually abandoned Gemini to build my own thing, primarily motivated by this obnoxious behavior.

They are pretty clownish on some things and that is one of them. They HAVE at some point started versioning their docker images at least, big win.

Fun fact: The npm release does not contain the code, it is downloaded by the postinsta script. Galaxy brain.

Maybe because of their rich provider ecosystem allowing you to use a whole bunch of subscriptions in it. I use OpenCode Go and OpenAI for almost everything, almost never touching OpenRouter.

Security misalignments OpenCode has a good few of (it used to have more).

I think the benefits of this one vastly outweigh the risks since it enables the self-modification Pi (!) became famous for.

Exactly! Almost all of the time this happens is during the initial exploration when the model decides to take a peek outside of its directory. It’s so annoying that that has to effectively kill the entire session.

From what I read, OpenCode V2’s design is similar. It’s built around Effect, which seems to be gaining steam as a framework for building correct software.

Cordis also looks interesting as an alternative, but appears more bound to DSH rather than a more general thing (which I prefer).

Yes, that’s the thing I had mainly been looking at as an alternative. It probably has a better culture around extensions, but also seemed a good bit more DIY, especially to get it working with a good UI.

Yeah, but only I can click to move a task to the background when it takes longer, which could and IMO should happen automatically. The model can’t always judge that. (I see how my comment wasn’t clear enough that I mean this)

In general, the preventable “model gets stuck” class of annoyances has not been fixed quite yet - there are also stil no timeouts for permission requests, at least by default.

It's good.

Many of the annoyances from V1 are gone. Skills and plugins reload automatically (finally). Tasks can be backgrounded (though I think this doesn't happen automatically yet). There's a built-in skill for editing the configuration (finally I can retire mine). The front-end got some love. Apparently they re-architected the loop around Effect. Sub-sessions can be communicated with better. Configuration is loaded from the folder the session is created in rather than where the server was started.

In general, this update largely takes OpenCode into the current age features-wise and is a good overhaul quality-wise. Just in time to stop my early switching away efforts, which would've been a pain because I know few other harnesses with such a strong server/client model.

I used it a lot in my AI client before I finally got around to hacking a Deepgram API server into Handy: It sucks enough to give me a headache worrying about its unreliability in every given transcription.

They supposedly improved it on later Apple Intelligence enabled devices, but I don’t have one.

Big fan of Handy. It's simple, offers all the models with a great picker and the author is involved in building a general STT model runtime. It does streaming flawlessly with Nemotron, which mind-bogglingly many apps don't offer.

I checked out the alternatives mentioned in this thread and they appear more over-loaded, "sloppy" and commercial. Handy just does the thing without extras.

I was going to say that’s not the level of security I require from that hypothetical organization: OpenAI was subverted by the perspective of many billions in profit, while many projects are running on idealism just fine (because there’s no profit on it).

I stand by that, but then again, this is about secrets management we’re talking about, so the organization would need a pretty high level of seriousness to be able to be trusted.

That’s a high bar for sure, but it looks to me like Bitwarden is trying to approach it from the top.

It can be complicated. I don’t remember the whole story, but I think on the older Xperias you’d have to take care to reinstall the proprietary drivers and there were keys that could be permanently lost and then you’d be out of some of the enhancements.

Maybe as an LLM it is, but I constantly use their streaming ASR model (called Nemotron Streaming) through Handy and it works wonderfully well.