sgc

For ocr, you have to practice, practice, practice with your prompts on small sample sizes before one-shotting a book. I create a json schema for the output that describes the structure in detail (using ai from my prose descriptions of structure of course). I have specific tough pages I use on openrouter to test every new model I am interested in. Gemini has and does absolutely dominate for this. But flash can be better because it doesn't overthink things.

Please stay away from Dell. Posting from the worst laptop I have ever owned, my Dell XPS that cost over 2.5k usd new. Yes, I hate it enough to rage-post about it on HN.

At least it sort of works after using the best thermal compound available, adding thick feet, and using it only on a cutting board to make sure an uneven surface like my lap doesn't reduce flow. Also, never place the charger on the carpet or it will shut itself off from thermal overload. Everything still randomly overheats with minimal workloads, but mainly it sort of works now :)

System needs to be carefully designed by experts for sure. Apparently that is a hard ask given the current state of affairs.

Well it would receive control signals from an on-site relay server with the hardware to communicate with it, and that would of course also be firewalled and use encryption to communicate with the outside world. That server could always in theory become compromised despite your best efforts, so the benefit is obvious: don't create even the theoretic possibility of being hacked by a class of message that should never exist.

The idea is not to solve everything with one simple trick, but to reduce the attack surface as much as possible in hardware. If you are dumb enough to make your 4 messages something like off, on, hard reset, random dangerous maintenance protocol, then it does nothing. If instead it encodes 4 useful and typically used states that can't destroy your system, then you have massively reduced the requirement for onsite, and accepted that more serious issues require onsite control.

The point is having the digital transmission simple enough that you can use fully *physical* filters to discard everything but the 3-4 messages you need through. Of course it is still digital, but my point is, it is a physically enforced studied, safe subset and doesn't allow for full network access. There are of course a variety of ways to further firewall that, but something needs to act like as much of an air gap as possible while allowing for remote communications.

Seems like 95% of devices could be managed with a hardware filter that only allows through a handful of very rudimentary signals (more like Morse code than a full digital transmission), and then of course a properly designed on-device API that defines those signals and doesn't allow you to trigger a defrost remotely, when that should only ever be done with somebody physically present for example.

Absolutely, what an incredibly low-effort click hijack farm it has become.

Don't do that! v041 has been compromised upstream by a supply chain attack and big fan doesn't want us to know. You have to use v040 until it has been fixed. Also, used the hacked flashing software because the official version is spyware. Did you even read page 723 of the support thread?? I can give you more info when I am done patching my toaster.

Your comment is a bit inaccurate.

1. There are laws already. It is an enforcement problem.

2. The Epstein-type crimes appear to be more prevalent in other branches of government and other parts of society than specifically in the legislative.

3. Enforcement problem is specifically in the executive, of which enforcement is a part.

It's a two-edged sword. Compared to other governments your own government when good is the best to promote your good. But when bad, it is by far the worst abuser.

That is the biggest difference here for me. I have not looked at every solution, but many. They are either much more expensive or garbage quality. For example my next target audiobook is a monster 250k words, 1.5m characters, so about $15 here or $75 using elevenlabs (0.05 per 1k characters). For me that is the difference between I will or I won't use it.

Sorry if this is in that article, but I am on my phone and can't see it. How much would this cost to batch generate an audiobook? Right now I just listen to things in the 11 labs app which is free, but I would rather just generate audio files.

They are saying there appears to be a lot more to these refusals than saying no, and this process appears to only touch the tip of the iceberg; as the refusal seems to run deeper into the token prediction process.

Thank you, I will keep my eye on it.

I went in to cash a check at a bank a few years ago and they made a huge show of it, taking over half an hour and various employees for a couple hundred dollars. They will absolutely win if they want to. A large organization can easily create more friction than an individual can bear.

I need to be able to enforce structured output easily (json, and xml if possible). That would be my number one slightly less than typical requirement.

Other than that, I like to insert agents into deterministic workflows, so some way to easily have deterministic steps (calling a script rather than delegating to an agent, but determined by the workflow rather than the agent). However this second one is likely out of scope for a while, since it is the inverse of the typical agent workflow. I could probably work this out clunkily by polling the cli from an external master script, but it would be cool to have it all inside the box.

Do you have a list of a few mcp servers or other tools you would add to help with the pizza-bot specific workflow? For example, how would you enforce structured output, should that be an mcp server, a plugin, already in the box?

I think there should be some focus on nature vs nurture here. At least some people have some degree of control over how they imagine things, and some training of when to use the best imaginative process as the tool it is would be useful. Struggling with a problem? Perhaps stop visualizing and try linearizing in monologue, or vice-versa, etc.

This entire conversation is new and fascinating to me. I used to be much more visual but I would say that my primary mode of thinking now is monologic. However I would never think to describe it as hearing. It is more like an imaginary 'voice' that is not the one I use with other people (it is absent timbre etc and just a much simpler more efficient voice abstracted from physical constraints). I use it to write out my thoughts in my mind. I quasi-visualize the chain of words just in a very abstract, largely amorphous way using this 'voice' and play them through my mind while 'speaking'/'hearing' them. In this sense I pretty literally get 'lost in my train of thought' when thinking about something interesting.

I don't know how much this style of thought came about because I really didn't enjoy constantly visualizing things and consistently fought against it; mainly due to the constant intrusive visual sequences of various types, of which I still have far too many for my taste. I use schematic and visual imaginations all the time, but I guess I feel safer using the monologue and so it has become the bulk of my thought. At the same time the fact that I have learned several languages and have had to consciously focus on language over the years undoubtedly plays a big part. I wonder how many polyglots who learned languages later in life use a similar monologic imagination style?

Why are you giving a shady company with shady terms and shady history the benefit of the doubt?

One can still note a poor decision based on its merits.

An equal area projection map is obviously valuable. But it's practically asinine to pick a projection that distorts land features when there are several equal area projections that do not.

Further, Ukraine had to abstain from this vote because they shoehorned in something ridiculous considering the official position of the UN members that Crimea is Ukraine [1], and Crimea is depicted as outside of Ukraine.

I have no idea if there are other hidden yet important changes, since there is hardly any commentary apart from size and shape.

[1] https://ukraine-war-analytics.com/history/international-reco...

I stay purely home network, as I suspect most do. So although I can see your use case, I can also understand why it's not a bigger priority than those other features.

It's probably the software I trust the most on my network to 'just work', and with the local caching I can use slower upstream dns providers and still have sub 1ms average latency (no performance worries by excluding google and cloudflare). I don't use it for dhcp, but it is a good fit that it's available as well. Couldn't be happier.

Rather late to comment here, but are there plans to support Hebrew, Syriac, Arabic, and other scripts? I love the readability and elegance of this font in English and Greek, and would like to use it at CatholicLibrary.org.

It seems like the best course of action would be to argue he did not destroy evidence, just made it unavailable at the location to force the requirement for a search warrant. It would probably be a hard sell, but I can't think of a better argument (not a lawyer).

I became so annoyed by google's ai overview and front page results that I am using 99% duckduckgo as the least bad free alternative for the other leg from my chatgpt usage in this context.

I asked a question once and now there is a effing alexa for shopping toolbar that takes a quarter of the screen that will not go away no matter how many times I close it, and the space remains taken even if I adblock it. Absolutely hostile implementation. I have words for this I cannot type out.

what else did I miss today?

fail2ban

Yes of course, the secrets have to be isolated from the VMs, preferably at the network gateway and not on the same KVM host. But as far as I know there is no safer containerization technology than a VM, and the only way to be more secure would be to have a physical computer per agent process? A KVM does not use the host kernel and user space, and provides hardware level isolation (the CPU hypervisor etc), that's the point. I think that once you are inside of vm, just using firejail is the better approach, since you have more direct control over the OS level controls that are being leveraged by various container solutions anyways. At any rate, that's what I did.

If you can't isolate a computer on your network, you probably can't isolate your network from the internet, so it's an irrelevant exercise at that point. And yes, probably you can't do any of those things and any frontier model could technically hack your network, but I don't think there's a better way to do it?