Miscellaneous – NearlyFreeSpeech.NET Blog

Is NearlyFreeSpeech.NET anti-AI?

jdw — Wed, 11 Feb 2026 07:14:46 +0000

I read recently that we were.
Are we?
Hmm. Not… exactly? A bit, I guess? But not. It’s complicated.
It’s certainly understandable why it might come across that way. We are, rather notoriously, anti-stupidity. And we’re anti-exploitation. And we’re not super fond of bullshit. LLMs, the specific type of generative AI most prominent in the public consciousness, do seem to be a fascinatingly complex mechanism powered by money and energy (and water, apparently?) that transforms stupidity and exploitation into previously-unimaginable quantities of bullshit. There’s a lot of overlap, is what I’m saying.
We are strongly against AI bots scraping our members’ sites. But that’s less because it’s got anything to do with AI and more because the bots are incredibly stupid and it’s exploitative. Two out of three ain’t… good. They’re also, regrettably, very difficult to stop. But we’re doing our best.
We’re also pretty concerned about the AI surveillance state represented by companies like Palantir and Flock Safety. LLMs get all the press, but (to shamelessly steal a wonderful idiom in their honor) these guys seem like the ones who have dedicated their lives to inventing the Torment Nexus from the sci-fi classic “Don’t Invent the Torment Nexus.” To them, Cyberpunk 2077’s Night City isn’t dystopian, it’s aspirational.
Deepfakes, of course, are pretty terrible. GenAI for the people who looked at the Internet and said, “You know what this needs? More nonconsensual pornography!”
Voice-cloning is also pretty scary. Or, at least, it’s scary that it’s trivially easy. How far are we from kids using their Fisher-Price Forge n’ Fraud play set to convince the school that their parents are keeping them home today?
These are all things that I’ve tried. I want to keep an open mind. And I want to understand what it’s capable of for myself. As hard as it would be to justify the wholesale theft of words, images, videos and likenesses that these models wouldn’t exist without, it’s like, OK, hit me. Show me why it’s all worth it. Show me what it can do!
So I set my home security system up to tell me when my dog is in the backyard, though sometimes it thinks she’s a cat. Or a squirrel. I’ve let coding agents write unit tests for me, which were actually mostly decent, except for that time an incorrect test failed so the agent quietly changed the code to also be wrong so the test would pass. I even vibe-coded a web frontend for a tool I use every day. I will never understand that code (partly because it’s terrible and partly because it’s all in Javascript), which seems like a problem, but it technically works. I’ve messed around with image generation, though I managed to do it without violating anyone. It seems like it only “works” if you give it vague instructions and aren’t invested in obtaining a specific result. I even cloned a voice, though I chose one that was already synthetic. (Ada from Satisfactory, if you must know.)
And, yes, I’ve tried chatbots, both local and the big names. I even found a case where they’re helpful! Say you’re writing a story or trying to plan something out. You get stuck, you give the AI some background and ask it what to do. And what it tells you is dumb and wrong. And then you tell it, “No, that’s dumb and wrong, because (reason).” Repeat a few times, and pay attention to the (reason)s. A lot of the time, you’ll wind up articulating important details about the situation that you didn’t know you knew. This only works with LLMs because, I find, human beings get peevish if you ask them for ideas and then repeatedly tell them why all of their suggestions are dumb and wrong.
I don’t know. The promise of AI seems great. But the current reality is an AI slop, revenge porn, security nightmare, technofascist horror show built by sacrificing every fixed representation of human creativity on the altar of “number go up.” All for what? The most expensive mediocrity the world has ever known?
I think if AI were in the hands of, say, our members, that maybe we’d get more of the promise and less of… that. It is really tempting to try it, because I trust our members and the world they would build vastly more than any of the billionaires who seem half a cackle from a volcano lair. But I went to a performance at a local university’s theater last week. At the beginning, they played a recording about how important it is to acknowledge that the university is built on stolen land. (Feel however you want to feel about that.) If we did try to take some of the more open tools and make them available, knowing how they were trained, damn, but that feels a whole lot like building on stolen land. And I’m not sure a recording would make me OK with it.
Also, as a more practical matter, I’m not thrilled that 1TB of ECC DDR5-6400 RAM costs over $30,000 right now. Guess we’re going to be in the group figuring out how to do more with less while our techno-overlords keep working on adding erotic mode to ChatGPT.
So, yeah. NearlyFreeSpeech.NET anti-AI? Honestly, I don’t know where people get these wild ideas!

Member support position

jdw — Sat, 12 Mar 2011 19:36:24 +0000

We are looking for one person willing to help out with member support on Friday, Saturday and Sunday.

This is a tough rotation to fill, so I wanted to mention it on our blog to get the widest possible audience.

To apply, check out our Work page, and if you’re willing to cover these particular days, please indicate that in your application for fastest consideration.

Quick Quote

jdw — Sat, 12 Feb 2011 02:21:33 +0000

Apropos of nothing, I just ran across this quote:

Plato used the dialogue format because the exchange of views, the posing and answering of questions, showed that understanding is a living, dynamic process. He distrusted writing because the settled character of the written word makes it look as if truth can be fixed and made to stand still. It is worth remembering that this greatest advocate of the objective reality of truth also believed that our access to that truth was sustained in reasoned discussion.

— John Churchill, From the Secretary: Inspiring Conversations in The Key Reporter. Vol 67, Number 4. P. 2., Summer 2002

I think this is one of the strongest, most concise arguments in favor of free speech and open debate that I have ever heard. I’d never heard of this guy, but it seems he’s the secretary of Phi Beta Kappa. Sounds like a smart cookie.

(Posted at the very lowest possible priority on our blog because it doesn’t have anything to do anything, bears repeating, and won’t fit in a tweet.)

Bug with MSNBot/1.1

jdw — Thu, 03 Apr 2008 08:41:45 +0000

Recently, Microsoft announced the release of MSNBot/1.1, which is designed to save bandwidth and improve their search rankings. We applaud this effort. However, we’ve discovered what we believe to be a bug in MSNBot/1.1.

MSNBot/1.1 is using the If-Modified-Since: HTTP header to save bandwidth. This is a good idea; it allows the server to say “nope, hasn’t changed” rather than sending back an entire huge file and letting Microsoft figure out if it changed or not. Better for them, better for you.

However, when MSNBot/1.1 issues an If-Modified-Since: header, they’re garbling the HTTP request. Here’s an example:

GET / HTTP/1.1
Accept: text/html, text/plain, text/xml, application/*, Model/vnd.dwf, drawing/x-dwf
Host: www.xxxxxxxxxxxx.com
Accept-Encoding: gzip, deflate
From: msnbot(at)microsoft.com
If-Modified-Since: Sun, 17 Dec 2006 23:34:11 GMT
 
User-Agent: msnbot/1.1 (+http://search.msn.com/msnbot.htm)
Connection: Close

Notice the extra line between the If-Modified-Since: header and the User-Agent: header. If it were just a blank line, it would be mostly fine; the User-Agent: and Connection: headers would just be regarded as part of a useless request body. Unfortunately, that extra line is actually a space on a line by itself. To be sure, we confirmed that with tcpdump:

	0x0020:  5018 ffff 4296 0000 4745 5420 2f20 4854  P...B...GET./.HT
	0x0030:  5450 2f31 2e31 0d0a 4163 6365 7074 3a20  TP/1.1..Accept:.
	0x0040:  7465 7874 2f68 746d 6c2c 2074 6578 742f  text/html,.text/
	0x0050:  706c 6169 6e2c 2074 6578 742f 786d 6c2c  plain,.text/xml,
	0x0060:  2061 7070 6c69 6361 7469 6f6e 2f2a 2c20  .application/*,.
	0x0070:  4d6f 6465 6c2f 766e 642e 6477 662c 2064  Model/vnd.dwf,.d
	0x0080:  7261 7769 6e67 2f78 2d64 7766 0d0a 486f  rawing/x-dwf..Ho
	0x0090:  7374 3a20 7777 772e xxxx xxxx xxxx xxxx  st:.www.xxxxxxxx
	0x00a0:  xxxx xxxx 2e63 6f6d 0d0a 4163 6365 7074  xxxx.com..Accept
	0x00b0:  2d45 6e63 6f64 696e 673a 2067 7a69 702c  -Encoding:.gzip,
	0x00c0:  2064 6566 6c61 7465 0d0a 4672 6f6d 3a20  .deflate..From:.
	0x00d0:  6d73 6e62 6f74 2861 7429 6d69 6372 6f73  msnbot(at)micros
	0x00e0:  6f66 742e 636f 6d0d 0a49 662d 4d6f 6469  oft.com..If-Modi
	0x00f0:  6669 6564 2d53 696e 6365 3a20 5375 6e2c  fied-Since:.Sun,
	0x0100:  2031 3720 4465 6320 3230 3036 2032 333a  .17.Dec.2006.23:
	0x0110:  3334 3a31 3120 474d 540d 0a20 0d0a 5573  34:11.GMT.....Us
	0x0120:  6572 2d41 6765 6e74 3a20 6d73 6e62 6f74  er-Agent:.msnbot
	0x0130:  2f31 2e31 2028 2b68 7474 703a 2f2f 7365  /1.1.(+http://se
	0x0140:  6172 6368 2e6d 736e 2e63 6f6d 2f6d 736e  arch.msn.com/msn
	0x0150:  626f 742e 6874 6d29 0d0a 436f 6e6e 6563  bot.htm)..Connec
	0x0160:  7469 6f6e 3a20 436c 6f73 650d 0a0d 0a    tion:.Close....

(Look for the “0d 0a20 0d0a” after If-Modfied-Since:.)

That’s a nasty violation of the HTTP spec, and causes the request to be dropped (and, on our system, logged, which is how we found out about it). We’d like Microsoft to fix this.

I’m posting about it here because, as with most large companies, the people responsible for MSNBot appear to be pretty well insulated from the outside world, no doubt due to several billion “your crap search engine lists my site as #372 when you search for cute kittens!” complaints. We don’t have a lot of members up in Redmond, for obvious reasons, but we do have some. And we have a bunch more who know people up there. So this is sort of a throwback to the days when mail was delivered by handing it to anyone who looked like they might be headed in the right direction.

So…. Where do you want to go today? Can you pass along a message when you get there?

Update: It’s resolved; see the comments for full details.