Claude is a UNIX command line tool with an SDK. Yes there's an interactive mode, but it can be invoked as a normal utility too, and piped to other tools and so on.

In that context, I don't understand the difference between a "third party harness" and a shell script.

How are they even detecting OpenClaw?

Im wondering this too. If I have my own local platform similar in nature to openclaw, and am leveraging claude -p through my subscription, is this now against ToS? Or is this just a ban specific to certain services? In which case they're saying 'use -p until you scale and then we'll hammer you'. Either way what a pita.

I guess Anthropic will scrutinize big open source projects for that purpose. The direct official integrations will be removed.

They have terms to not allow `claude -p` used like that. However, people can hide this with the leaked source code. What a funny cat-and-mouse game!

Hi! Do you ever make quants of the base models? I'm interested in experimenting with them in non-chat contexts.

Yes, they are listed on huggingface. The instruction trained models have an 'it' in their name.

https://huggingface.co/collections/unsloth/gemma-4

Edit: Sorry, I'm not sure if this is a quant, but it says 'finetuned' from the Google Gemma 4 parent snapshot. It's the same size as the UD 8-bit quant though.

Only the 'it' models seem to have quants. I was really hoping to try a base model.

Basic quantization is easy if you have enough RAM (not VRAM) to load the weights.

You don't generally commit compiled code to your VCS. If you do need to commit a binary for whatever reason, yeah it makes sense to explain how the binary was generated.

You do usually pin your compiler version though, or at the very least set a minimum version

Back in the day, Objective-C was considered a cool hip language, wasn't it?

I had to double check that they'd removed the non-1M option, and... WTF? This is what's in `/config` → `model`

    1. Default (recommended)    Opus 4.6 with 1M context · Most capable for complex work
    2. Sonnet                   Sonnet 4.6 · Best for everyday tasks
    3. Sonnet (1M context)      Sonnet 4.6 with 1M context · Billed as extra usage · $3/$15 per Mtok
    4. Haiku                    Haiku 4.5 · Fastest for quick answers

So there's an option to use non-1M Sonnet, but not non-1M Opus?

Except wait, I guess that actually makes sense, because it says Sonnet 1M is billed as extra usage... but also WTF, why is Sonnet 1M billed as extra usage? So Opus 1M is included in Max, but if you want the worse model with that much context, you have to pay extra? Why the heck would anyone do that?

The screen does also say "For other/previous model names, specify with --model", so I assume you can use that to get 200K Opus, but I'm very confused why Anthropic wouldn't include that in the list of options.

What a strange UX decision. I'm not personally annoyed, I just think it's bizarre.

`/model opus` sets it to the original non-1M Opus... for now.

Thanks. I quickly burned through $100 in credit when I started using Opus 4.6 in OpenCode via OpenRouter. My session stopped and was getting an error not representative of credit availability, so was surprised after a few minutes when I finally realized Opus just destroyed those credits on a bullshit reasoning loop it got stuck in. Anthropic seems to know that the expanded context is better for their bottom line as they've defaulted it now.

And as others have said it's very easy to burn token usage on the $100/month plan. It's getting to the point where it's going to very much make sense to do model routing when using coding tooling.

Not sure why you were downvoted because this is actually correct. Can also use --model opus

I wouldn't want a touch screen to become the primary input device, but I think it would be useful on occasion. Not entirely unlike how we still have touchpads even though we try to use keyboard commands.

Do you have the brightness on your monitor set really high or something?

I frequently use macOS on a projector, it doesn't quite fill my wall floor to ceiling but it comes close. I don't use full screen often, but I do it occasionally as a focusing strategy, and it's fine.

Projectors are way easier on the eyes than monitors though.

You're shining a bright light on a wall, which you are looking at.

With a monitor you are shining a bright light at your face, while staring directly at the lightbulb!

Doesn't bouncing off the wall just effectively make the "backlight" dimmer? The light reflected off the wall is hitting your face versus the light from the screen hitting your face. It's still light regardless.

If you're using a monitor in the dark the way you use a projector, you should turn the backlight down. If you're using it in a well lit room, the brighter backlight should have less of an effect.

> The light reflected off the wall is hitting your face versus the light from the screen hitting your face. It's still light regardless.

It sounds to me you've never actually looked at a monitor display large swaths of white before, it's brighter than light hitting a wall for sure, even with the brightness down, extra so when the ambient lightning is dark too.

I've definitely seen large monitors that are unpleasantly bright in the dark, but I've also seen an overly bright projector that was similarly unpleasant. I genuinely don't understand why changing the backlight wouldn't fix everything. A projector's image isn't diffuse like a lightbulb, if it was you wouldn't see an image.

In principle, it's the same as staring at the moon Vs staring at the sun.

The fact that it's bright outside when the sun is up might help, but it's nowhere near enough to compensate!

Look, if this was a project on using DNS to replace Dropbox or something, I'd agree with you.

But the demo version of Doom just isn't that large; Cloudflare will host much larger files than that for free via Cloudflare Pages/Workers. This project is clearly meant as a fun proof of concept, not some novel way to host 3 MB for free.

I’ve heard rumors that DNS records are also sometimes used in some steganography-type communications. Great way of passing small messages in a ubiquitous and innocuous system, unlikely to be blocked or raise eyebrows by accessing.

With iodine you can tunnel TCP/IP over DNS. Really slow but usable for text web sites, gopher, gemini, irc...

gopher -> gopher://magical.fish, gopher://sdf.org...

gemini -> gemini://gemi.dev, it has geminipedia, a web to gemini converter reading sites over gemini at great speeds.

irc -> servers from https://bitlbee.org will allow upon connecting to a registered IRC account to several protocols in the server. For instance, XMPP users will appear as IRC users and groupchat can be created as IRC channels. Ditto with Mastodon, Discord...

mail/usenet -> well, except for big attachments and news binaries (free NNTP servers will just serve text) once you used something like mbsync/msmtp to store your IMAP mail locally and send email ondemand (and ditto with Usenet with slrnpull doing the same exact same task for pushing your writtings and pulling down new articles) everything would just work slower, but usable enough as it can be just batch-uploaded/downloaded overnight.

Iodine it's really great for open but paid wifi services behind portals, such as some hotels, airports...

It won't give you broadband speeds but you can at least chat with people, read some blogs or news at https://lite.cnn.com or https://text.npr.org or get some classic from Gutenberg. That's better than nothing.

I've never had great luck getting iodine running anywhere. The one and only success I've had was on an aircraft where, after numerous attempts at different things, the best I could do is connect to an SMTP server and send an email manually.

A popular use of DNS is for malware to communicate their status. They do this by requesting e.g. "i_am_in_$RANDOM_NUMBER".badplace.ru.cn.cx.

If you consider information theory, when something has states, you can store data in any system that has multiple states, which means you can store data in any system.

The placement of coffee cups on a table can be used to encode data.

At that point, only your audience needs to know that data is there.

I mean, kind of, but they're able to be cached easily and inexpensively in a way that kind of defies the intrinsic values behind steganography.

Not cache-able if no one has seen them before.

it can be used as a novel way to host files.

Airplanes and many other captive portals will allow DNS traffic, but restrict everything else. Such things can be used to get free internet in such environments. It is indeed an abuse of protocol, and future protocols are going to make life difficult for everyone to prevent such abuse.

The fun thing I took away from this was the existence of managed-doom.

> They would have been better off building a war chest and pulling a Valve (though I'm sure they'd hate hearing it that way)

Isn't that exactly what they were trying to do with the Epic Game Store?

Steam is the thing that has made Valve successful. They were great as a game company, but as you said, the games don't last. Steam does, and I don't think Valve would be that successful in a business sense without it.

> Also, why is this the only scenario where they block the mic and camera? Locker rooms are apparently fine.

How would the phone detect that you're in a locker room? Even if it is possible, it seems very hard and likely error prone. Disabling call recording is easy.