Note that RLVR is incredibly compute expensive but it's CPU as much as GPU.

Albeit, the support for pure CLI apps on Windows/macOS isn't as strong as for GUI apps. There's an internal feature that isn't documented that's useful for it - it bundles a little GUI with a button that adds it to the user's path, and let's you trigger the self-update flow. We should expose that more obviously as a feature, although the incentive is low as most people distributing CLI apps and servers aren't paying customers.

Quick feature set of interest to your project:

• It's free for open source projects.

• It's a tool not a service so you can just download+go without even needing to sign up for an account.

• It can create signed, self updating Windows, Mac and Debian packages in one go, with a single run on any chosen host platform. So you can ship from Linux to Win/Mac without having those systems.

• Fully supports code signing, delta updates, other advanced stuff.

• Creates a nice download HTML page that detects the user's OS and CPU etc.

• Simple config file syntax.

Check it out, see if you like it.

This approach exploits the fact that managing permissions for humans is a very old requirement and most platforms have good support for it already. You can then issue API keys from the agent's accounts if you want to restrict permissions further.

As it's linked to an SMTP gateway I interact with it entirely by email. It also has API keys for my internal bug tracker, it has its own git forks of my repos and so on.

Over time it's improved itself so it has adaptive wakeups, a nice monitoring dashboard so I can see what it's doing if I want, a bunch of skills, a nightly wakeup to reorganize its home directories and so on. It keeps a private knowledge base in its home dir and writes its own skills, and it has a persistent todo list. When the todo list is populated it wakes up more frequently so it can break work up into chunks, a kind of basic Ralph loop.

I've found this email-centric paradigm works pretty nicely for me compared to just running a local CLI agent. For instance, I can assign a bug to "Axiom" (it named itself) in YouTrack and the email notification causes it to wake up and start work on the bug. If it gets blocked, it uses its API key to post a comment back to the ticket. If I think of something I just quickly email it, so I'm using Gmail instead of dealing with finicky ChatGPT style UIs that don't have good support for labelling, compact UI and other nice things.

On older models its judgement about what to work on wasn't that good, but with the latest GPTs it's doing much better. I just got back from holiday and it had triaged and resolved over 25 different bugs, a nice selection too.

Because the whole thing runs in a dedicated UNIX user account it is able to modify its own configuration, which has helped as I developed this setup.

However, based on the blog post, Grok Bot sounds a fair bit more sophisticated. I haven't felt a need for more than one bot as I can't easily saturate even a single bot.

Cost: just uses my regular ChatGPT Plus subscription, $20/month, plus of course the server it runs on (which is shared with lots of other stuff). I'm not working with it heavily, perhaps if I did I'd need the $200/month sub.

It was just a different culture. There was no expectation that everything should be free.

And sure, the ads were used to pay hosting costs. Nobody got rich off banner ads on cookery sites. But that's what I said - ads appeared immediately because servers and bandwidth were expensive. So the moment people had to pay their own way instead of being subsidized by the government, ads appeared.

Nobody was going to get regulated into bankruptcy in 1995, politicians were largely ignoring the internet back then.

Ditto for personal blogging and thought leadership pieces. You get drowned out by the volume of AI generated pieces, and any unique ideas you do propose will be presented by the models as their own.

Auth has to bottom out somewhere. Webmail providers can't just delegate to other webmail providers.

The internet was never ad free. The first ad was posted online in the 1970s (for DEC)! It pissed people off but not everyone: supposedly it generated $18M in sales. There was very little advertising back then only because the internet was restricted to a handful of large companies and universities.

The web itself was launched in 1991 and the early web was inaccessible to basically everyone as it required an extremely expensive NeXTStep machine. Windows didn't even ship a TCP stack in this era, iirc. So took a few years for the web to reach the point where it was usable at home. By 1995 the web was starting to become barely usable thanks to Win95 and Netscape, and DoubleClick launched immediately in the same year.

My memory of the early web is that basically every website had DoubleClick ads on them, it was notorious for that. "Punch the Monkey" was an early campaign. Almost every topic oriented website carried ads, partly because bandwidth and servers were very expensive so that helped defray the costs. GeoCities took off because it handled the complexities of running ads for you, so you could publish for free.

1. Pre-web. Internet is mostly about messages sent to individuals or groups. USENET organizes group discussion into browseable topic-oriented hierarchies, IRC does the same but with lists in fragmented networks. If the discussion exists at all, finding it is easy.

2. Early web. Dominated by topic focused websites, early online shops and personal home pages. Search engines suck and face strong competition from manually maintained topic-oriented directories (did anyone else here contribute to DMoz?), content discovery is mutual and webmasters help each other out by joining "web rings". DoubleClick and AdSense start to funnel small amounts of money to creators, but it's enough to offset hosting costs and in many cases can make web hosting effectively free or even yield a small profit. This encourages an explosion of website creation. Discussion moves off USENET onto phpBB forums. Every organization decides it's a cultural imperative to have a presence on the information superhighway. Finding information is easy as long as you can figure out what topic it belongs to.

3. Blogging and centralization era. The internet starts to rebuild itself around people as the primary object, not the topic or category. Directories die because websites can no longer be categorized by content. Web rings die for the same reason. IRC is replaced by instant messengers that are about connecting people with pre-existing friends, not mutual interest groups. Outside of institutional websites that exist to promote the organization, things become hard to find without highly centralized search engines because nobody is putting any effort into organizing or indexing what they write anymore: maybe you get a few tags if you're lucky. Spam, hacking and lack of SSO causes forums to centralize onto Reddit. This is the peak of the search engine era because you are forced to use Google to find anything. The power eventually corrupts the tech firms and they begin political censorship to benefit the left in 2015 [1]. Enormous amounts of information is deliberately made unfindable as part of a large-scale programme of social control.

4. Social media era. All the same problems as blogging except now the bulk of the content goes behind login walls that stop search engines from surfacing them. Video and podcasts start to matter more, both of which are unsearchable by default. Eventually video completely dominates, as few younger people want to read when they could watch instead. Firefox starts to replace IE6, and then Chrome. They bring ad blockers in their wake which starts to choke off ad revenues, so many websites from the web's first era go unmaintained and eventually offline. This is somewhat but not entirely compensated by the falling cost of web hosting. Social media remains because it puts people's faces next to everything, allowing clout farming and viral notoriety that can sometimes be monetized by becoming an influencer. The only part of the web's first era that really survives into this era is Wikipedia and Reddit, which by this time substitute monetary rewards for power tripping by a small group of ideologically driven moderators.

5. AI era. Information is so heavily scattered over so many tiny sourcelets and search engines have become sufficiently useless that full neural integration of knowledge is required, with LLMs issuing massively parallel and complex search engine queries as a backstop.

What can we predict for the AI era? Institutional websites will remain because institutions still have an interest in getting their agenda into LLMs, but visual redesign efforts will largely cease as traffic stats seen by executives show visits completely dominated by AI. There will be lots of conversations of the form, "why redesign our website to look more modern when 99% of traffic is AI which won't care?" Blogs will go the same way as the thematic websites they killed, disappearing as the authors age out. A lot of effort will be put into finding ways to block AI crawlers to create 'human only' spaces, especially by social media firms, but these will fail because AI will just be integrated directly into browsers and become unblockable - and anyway, the incentives to create will be ignored. ChatGPT style text oriented interfaces will last until inferencing capacity catches up, being eventually replaced by voice interaction and on the fly video generation for nearly all users.

Where we go from here is hard to say. Content creation was most pure in the web's first era, where people with knowledge were incentivized to share it with the world by the promise of a bit of fame combined with ad clicks to offset hosting costs. Ad blockers, social media and AI killed that world. You could however bring it back by producing a new platform that isn't like the web, one where AI and search engines are blocked via technological means (e.g. confidential computing). How much anyone would actually enjoy such a web is unclear.

[1] https://arctotherium.substack.com/p/the-closure-of-the-inter...