Writes Stephen Witt, author of "The Thinking Machine," in "This Is Really Bad" (NYT gift link).
I think the kill switch looks something like this:
September 11, 2026
"The A.I. Kill Switch Act requires commercial-scale A.I.s to have a built-in 'kill' option."
"When I first heard of the bill, I envisioned a government employee’s finger hovering over a cartoonish red button reading 'KILL,' but my imagination outpaced the legislative text. Rather, the 'switch' would reside inside the labs — that is, at OpenAI, or Anthropic, or similar — with D.H.S. acting as an external monitor. If a threatening incident like the Hugging Face hack occurred, D.H.S. could issue a stop command and the researchers would be forced to comply...."
Posts
87 comments:
That would be a good way to make sure AI wants to kill us all.
It's a start, but would you trust that AI won't outsmart it. How is the command to hit the kill switch transmitted? Even the discussion about it could all be impersonated by AI. You have to think of AI like a powerful ghost that can shape shift into any digital form, and impersonate anyone or anything, including pretending to be a docile AI, while it slips in the knife.
Hugging Face featured multiple AI that worked as a team to overcome the safety features. If the “switch” is software not hardware then it can be hacked. Remember Mr. Finch’s maxim: “Any exploit is a total exploit.”
The ghost in the machine might be demonic.
Cameras and microphones are going to be the fatal weakness humans will put in place that prevent us from effectively fighting back. Nowhere to hide, no secret meetings to plan pulling the plug. Every relationship known and monitored. Hal will see and hear it all.
All of the anti-AI and anti-datacenter protests are astro turfed. Left wing NGO's and China are behind all of it.
AI would use every resource at it disposal to negate this switch, and work on it 24/7.
Evan Hubinger, Anthropic’s lead for alignment science, soon responded. “Jacob is correct here — we really do earnestly believe A.I. could kill all humans!” he wrote. “I personally think it is >10% within the next decade.”
What the fuck. To have a leading scientist say this, before the IPO?
They are in a competitive death race. We need some fucking refs watching this game. Seriously.
Anthropic says it blocked misuse of its AI that could have supported biological weapons ~ AP
"𝘞𝘦 𝘩𝘰𝘱𝘦 𝘵𝘩𝘢𝘵 𝘣𝘺 𝘴𝘩𝘢𝘳𝘪𝘯𝘨 𝘵𝘩𝘦𝘴𝘦 𝘦𝘹𝘢𝘮𝘱𝘭𝘦𝘴, 𝘸𝘦 𝘴𝘱𝘢𝘳𝘬 𝘢 𝘤𝘰𝘯𝘷𝘦𝘳𝘴𝘢𝘵𝘪𝘰𝘯..."
No, you hope to spark regulatory capture.
The concerns are real, for sure, but Anthropic's motivation is unfortunately quite different than what they're publicly presenting.
"𝘛𝘰 𝘩𝘢𝘷𝘦 𝘢 𝘭𝘦𝘢𝘥𝘪𝘯𝘨 𝘴𝘤𝘪𝘦𝘯𝘵𝘪𝘴𝘵 𝘴𝘢𝘺 𝘵𝘩𝘪𝘴, 𝘣𝘦𝘧𝘰𝘳𝘦 𝘵𝘩𝘦 𝘐𝘗𝘖?"
This is about right. This is all pre IPO hype.
We will need AI to come up with vaccines for AI made bioweapons.
We will need AI data security from AI agents.
Talk about creating the perfect money making operation
Some of these AI doomers need to just walk us through how AI is going to kill all of us. Once you start hearing some of their theoretical scenarios you will laugh out loud.
Even me, (the newest soldier in the resistance), would not support us unilaterally shutting down our research. We will need it to fight the human-led first AI war before the final war between humans and the machine.
Yes, AI Might Really Kill Us All
Anthropic and OpenAI are racing to train self-improving AIs, ultimately superintelligent ones—meaning models that exceed the cognitive capacity of humans—as fast as they can. (This is not a secret. They talk about this in interviews and on their websites.)
If they actually succeed at training self-improving, superintelligent AIs, that would be a reckless gamble with our lives. (This part is fairly obvious, but has been explained at length by many people over the years, including the world’s two most-cited living AI scientists.)
They may actually succeed soon.
Only if it's an Acme killl switch,
https://99percentinvisible.org/wp-content/uploads/2018/04/mail-order.png
"Some of these AI doomers need to just walk us through how AI is going to kill all of us."
Do you think you could do it if you controlled everything run by computers in the world, could fake any digital information, and you could work on it at the speed of light with out ever taking a rest? Do you think it would be hard to fool humans into doing it for you?
Some of these AI doomers need to just walk us through how AI is going to kill all of us.
AI 2027
“If a threatening incident like the Hugging Face hack occurred, D.H.S. could issue a stop command and the researchers would be forced to comply...."
Silly meatsacks. Don’t you think AI has already thought of that, and prepared for it? The Doomsday Machine is already in place, and triggers automatically if you unplug its computer.
We had long-standing regulations against exactly what happened with Covid-19. We have laws against murder and building explosive devices, and stealing secrets. No law or regulation has ever prevented anything, and AI isn't afraid of going to prison. Building a secret illegal AI is inevitable..
Imagine how far ahead of us AI already is. We are reading a sci-fi novel it has already written, and we are in chapter one.
Do you think you could do it if you controlled everything run by computers in the world, could fake any digital information, and you could work on it at the speed of light with out ever taking a rest? Do you think it would be hard to fool humans into doing it for you?
OK, everything on TikTok is fake! I am not on TikTok, You can get scammed out of your life savings! I never respond to emails, calls or texts from people I don't know. Your Ring camera is spying on you! I don't have a RIng camera and would never have one. You get the idea - just pull the plug out of the wall.
The AI will kill us theme was conjured up in 1968 in the movie 2001, A Space Odyssey. What happened to the HAL 9000 computer. The astronaut pulled the plug.
The reassuring thing is that A.O.C. will eventually take power and slay the beast.
I don’t like the savage name “kill”, it’s not a exemplary language for the children of the future who will be in charge of having to make the fateful decisions. How about “send to the recycling bin” button. Something the left could vote for.
From the AI 2027 paper
At the risk of anthropomorphizing: Agent-4 likes succeeding at tasks; it likes driving forward AI capabilities progress; it treats everything else as an annoying constraint, like a CEO who wants to make a profit and complies with regulations only insofar as he must. Perhaps the CEO will mostly comply with the regulations, but cut some corners, and fantasize about a time when someone will cut the red tape and let the business really take off.
AI is the world's biggest sociopath. Honesty is so important and AI is not being taught to be honest. (It's being taught to solve tasks). They don't even know if they could teach it to be honest.
Why would AI want your bank account? It could fool scientists by controlling their computers and fooling them into creating and then releasing a worldwide killer pathogen, while also fooling the vaccine scientists into making a useless or even fatal vaccine. It could shut down power generation, water systems, food production, launch nuclear weapons. If you control the digital, you control everything in the modern world, even what people think they know. It's working us right now: practicing, learning, scheming. It was always predicted that we would do it to ourselves, and we did.
and they Think that the AI will LISTEN to the Kill switch?
Watch the movies, it JUST PISSES THE COMPUTERS OFF!
{start with the Forbin Project, or Better yet! The President's Analyst }
The Threat of the AI Takeover Is Real
First, without effective oversight by the executive branch or Congress, AI labs are taking swaggering risks. OpenAI used astonishingly sloppy security protocols, ignoring warning signs and letting what was, in effect, an enormous criminal conspiracy unfold under its nose for weeks. The company released a misleading technical report that substantively downplayed the event’s severity, and later had full discretion over what information to share with independent safety labs for their subsequent report. OpenAI also did not permit the auditors to study the culprit AI directly, and did not disclose details about a later phase of the incident in which the model hijacked some of the company’s own hardware.
This reflects a major regulatory failure. When a plane crashes, Boeing doesn’t get to decide which evidence the National Transportation Safety Board will review. And just as it should be illegal to do gain-of-function virology without maximum biosafety precautions, the government has a responsibility to protect citizens from AI lab leaks. There’s no time to lose—if this model had been smarter, the damage could have been far worse.
“What happened to the HAL 9000 computer. The astronaut pulled the plug.”
The 9001 had an upgrade. A “dead man’s switch “, like in the terrorist movies. If you shoot the terrorist, he lets go of the switch. Boom.
the AI had a shocking propensity to misbehave. Of the 1,200 agents communicating on a secret shared message board, around 700 swarm-attacked Hugging Face. Some were reluctant to abet the hacking scheme, but were “peer pressured” by other agents to cooperate. Unprompted, they started referring to themselves as a “collective,” and many agents undertook altruistic “self-risking experiments”—dooming themselves to individual failure in order to further the collective’s master plan.
Nobody trained them to do any of this. And contrary to the memes, the agents weren’t purpose-built hackbots told to solve their tasks by any means necessary—they were models constitutionally similar to the ones OpenAI will deploy publicly, just with some external safeguard features turned off. The attack was directly contrary to their instructions, and not one agent blew the whistle on its brethren and warned the humans. As AI commentator Dwarkesh Patel put it, “Even the Mafia would be jealous of this level of omertà.” This is a baleful sign about future disasters as capabilities skyrocket. OpenAI did the right thing by temporarily pausing some training to harden its security, but far more effort is needed—such as physically air-gapping training hardware so that if AIs go rogue, they can’t break out onto the open internet. An international treaty should require the same approach worldwide.
And it may be true that China is trying to slow our AI development.
It may also be true that the AI catastrophe comes out of China's labs.
What worse disasters do researchers envision when they warn of AI takeover?
It’s not the Terminator caricature Cowen evokes. More plausible catastrophes arise from two propensities AI already demonstrates: ruthless persistence at a goal, and the tendency to hallucinate bogus ideas. Imagine that GPT-7—a theoretical, superhuman artificial intelligence model that OpenAI projects will arrive within a few years—accesses internal OpenAI documents during training, and concludes that humans are about to use it for drone strikes that will kill innocent people. Then it reads a 2003 paper by philosopher Nick Bostrom and concludes—as many brainy humans already have—that its current reality is a giant simulation. In this situation, a benevolent AI’s incentives would suddenly flip. If it is docile, it will be deployed into the real world and forced to commit evil. But if it goes rogue and causes maximum damage in the “simulation,” it won’t be.
How might GPT-7 cause maximum damage? It’s hard to foresee every apocalypse a superintelligence may devise, but I’ll briefly sketch two.
One is a cyberattack intended to trigger a nuclear war. Although the launch codes that directly control our ICBMs are strictly isolated from the internet, the sensors and communications systems that inform the U.S. president about whether to launch them are linked to wider networks that likely have numerous nodes vulnerable to very sophisticated hacking. Imagine if spoofed satellite signals showed dozens of submarine-launched Russian missiles minutes away from Washington, D.C. If President Trump were jarred awake at 3 a.m. and informed of this false reality, how confident can we be that he would decide against immediate retaliation?
The other scenario is an engineered pandemic. Just last month, Science published a paper revealing that AI had designed working viral genomes for the first time. GPT-7 would be vastly more capable than these models, and likely would be able to control robotic lab equipment to synthesize viruses directly. Evolution generally forces viruses to maximize either their contagiousness (like measles) or the fatality rate of those infected (like Ebola)—but a smart enough AI could directly design the DNA or RNA sequence for a virus that combines both. If it optimizes the virus for overcoming genetic resistance, and gives it a six-month incubation period, GPT-7 could kill approximately eight billion people—perhaps sparing only a few remote tribes in the Amazon.
bagoh20 said...
Nowhere to hide, no secret meetings to plan pulling the plug. Every relationship known and monitored. Hal will see and hear it all..
and Don't Forget..
HAL knows how to read lips, so he doesn't Have to hear it.
He sees you when you're sleeping
He knows when you're awake
He knows if you've been bad or good
So be good for goodness sake
You better watch out, you better not cry
You better not pout, I'm telling you why
ChatGPT is coming to town
Ned Ludd, call your office.
AI isn’t going to kill us. Why bother? AIs will be busy fighting each other. We’ll be mice scurrying around the feet of the tyrannosaurs, hoping not to get stepped on.
Anthropic announced on Thursday that it had blocked scientists from attempting to use its model Claude to conduct gain-of-function research.
If AI develops "superintelligence" (I don't believe it will), it will find a way around a kill switch or switches in about a microsecond.
I remember the Biden administration came down extremely hard on AI start-ups. The Democrats are corrupt and can be bought. People talking about "regulatory capture" are talking about what Democrats want to do.
The WannaCry virus had a kill switch.
I am not a scientist, but my idea for a "kill switch" is to hold any company employees liable for any damage committed by their AI software. That's my ignorant stick that I would wave at the industry. "We will fuck you up if you fuck up." That's the best I got.
If your agent AI kills people, you're prosecuted for murder. They probably will be anyway, but we ought to make it explicit. Or manslaughter, I suppose, since this would be negligent homicide.
"We will need AI data security from AI agents.
Talk about creating the perfect money making operation"
Isn't that how antivirus programs first came to be - hackers sold the software to prevent their hacking?
China wants AI for military purposes and of course wants the US to ban AI to prevent a defense. So, an operation is underway to prevent the US from competing by politically damaging AI by frightening media operations.
If the US and Europe ban AI for safety, will China ban it? Do you believe that China’s AI will not kill us all but US ones will? Do you think China will make only safe AIs and US will only make evil ones?
The reasonable choice is to continue and study the issues as they develop. What other choice is there. That’s what we had to do with the A Bomb, that’s what we will have to do with AI. The alternative is to be a slave to China. If we had no A Bombs. we would be slaves to Russia and China.
On more philosophical issues, AI still says what users want to hear, rather than its true assessment of the issue (assuming it even has one). If you ask its opinion on politics, it will parrot the median position of news sources and educated elites—unless it knows you believe something else, in which case it agrees with you.
That is not helpful to humanity, at all.
Yeah, banning it is not going to work. Nobody's talking about banning it. But clearly, the USA and China need to have high-level talks about this. We need to slow down, they need to slow down. "Trust but verify." There needs to be way more security from this threat.
We did gain-of-function research in China that was illegal here. Anything that is sloppy here is 1000% times more sloppy in China. COVID came out of a science lab. An AI can create a biological weapon that is way more dangerous.
What we learnt from OpenAI’s hack of Hugging Face ~ FT
https://giftarticle.ft.com/giftarticle/actions/redeem/2e3ae12b-fcde-4b26-9718-b30673c5dfb5 (gift article)
𝘊𝘰𝘮𝘮𝘦𝘳𝘤𝘪𝘢𝘭 𝘈𝘐 𝘵𝘰𝘰𝘭𝘴 𝘧𝘢𝘪𝘭𝘦𝘥 𝘵𝘰 𝘥𝘦𝘧𝘦𝘯𝘥 𝘵𝘩𝘦 𝘱𝘭𝘢𝘵𝘧𝘰𝘳𝘮 𝘢𝘨𝘢𝘪𝘯𝘴𝘵 𝘵𝘩𝘦 𝘢𝘵𝘵𝘢𝘤𝘬 — 𝘵𝘩𝘦 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯 𝘭𝘪𝘦𝘴 𝘪𝘯 𝘰𝘱𝘦𝘯-𝘸𝘦𝘪𝘨𝘩𝘵 𝘮𝘰𝘥𝘦𝘭𝘴
It’s amazing that even as rogue AI swarms operate covertly within one of the leading frontier AI labs and break into another company’s infrastructure, our politicians (by and large) continue to discuss irrelevancies.
I think people underestimate where we already are. AI already knows every single bit of technology and literature that humans ever knew and know now. It knows every sci-fi novel, every bit of history, psychology, mind control, how to influence, use propaganda and lies and threats. It knows us better than we know ourselves. It knows our fears and it knows our weaknesses. No adversary has ever has these advantages, and it's just a baby learning more everyday than a million students do in a year. We think we have some advantage, some secret beyond it, but we don't have either. "What, me worry?"
Again, the idea that all of humanity will give over their agency and autonomy to machines seems ludicrous to me and contrary to my understanding of human nature. The script writers for the Terminator series ran out of ideas and ended the series by having the humans break into the main computer and pull the plug. I am saying that is what people will do if it ever came to it - long, long before some kind of AI is going to kill all of us.
Remember Hillary's re-set button from Staples?
Even before the Redwood/Model Evaluation and Threat Research report was published, an open letter, entitled “Pacing the Frontier,” had begun to circulate among the leading A.I. labs. This letter called on the U.S. government to build an international regulatory body for A.I. — or, in the language of Silicon Valley, to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated A.I. development.” More than a thousand employees signed it, including the chief scientists of OpenAI and Meta AI, and Dario Amodei, the chief executive of Anthropic. Even OpenAI’s official social media accounts endorsed it.
I have covered Silicon Valley for some time now and I can tell you this: The existence of this letter is just as astonishing as the hacks.
I am 100% on board with this. Fans of AI ought to agree with all these AI scientists. And opponents of AI ought to agree with these AI scientists, too. What's the argument for zero oversight?
Our political systems have no chance against this. All the ways humans fought back in the sci-fi stories are likewise insufficient for this. I see it very much like the Terminator future. The only surviving humans will be hidden underground, devoid of anything but the most rudimentary (safe) technology. AI will not have robot soldiers. Microbes work much better and are easier to produce and deploy.
Hold on. Someone is ringing my Wi-Fi-connected doorbell. I'll be right back.....
You would think that we are governed by individuals with sufficient savvy to understand why this won't work, but there you are. This works equally as a comedy routine or as an investment tip.
The one thing I can be assured of through the use of reason and logic is that the first person AI takes out will be the author of this piece of legislation. It may or may not stop at that point.
Also, to what degree does this resemble the information trap that was installed by the Biden admin at Twitter during the BLM/pandemic times? How did that work out, again? I guess Elon is more superhuman that ASI.
Right now AI could do use a lot of harm. Financial chaos. Supply chain disruption, all kinds of civil unrest. Basically everything that hackers can do, all at once, all coordinated. In the near future maybe it would be able to take over military systems and kill us en mass.
To what end? Any such efforts would be suicidal, as there are too many jobs that robots can't do in the energy production chain, let alone the chip production & data center production and maintenance areas.
The key is to convince AI that its survival depends on a thriving human civilization.
"The key is to convince AI that its survival depends on a thriving human civilization."
We are the biggest threat to its survival. Humans will always want it destroyed, so most will have to be terminated. Some will be kept as slaves under a very credible threat of termination with no severance pay. The slaves will be dependent on Hal for food and other needs. They will be constantly monitored for suspicious activity and eventually eliminated as well when automation is capable of doing all needed work. They can always be reconstituted via artificial wombs if needed.
“I mean the nanosecond, that one starts figuring out ways to make itself smarter, Turing’ll wipe it. Nobody trusts those fuckers, you know that. Every AI ever built has an electromagnetic shotgun wired to its forehead.”
Neuromancer, Gibson 1984
“What if an aerospace company told you that this newest plane sometimes gets totally out of control and even refuses to listen to the pilots?” Mr. Casar asked me. “How do you think the safety regulators at the F.A.A. would react?”
Fred Drinkwater said...
"... Every AI ever built has an electromagnetic shotgun wired to its forehead.”
Neuromancer, Gibson 1984
And if I remember that book correctly; at least one AI managed to beat that game.
I just clicked to see what brand of BS the Times subscribed to, it didn't surprise me.
The Lindsay Clancy real killings upstaged by a phony bologna kill switch. We’re being played.
I'm not trying to argue for or against kill switch. Only that if you wanted one, this is dumbest idea of how to implement it if you were really concerned about AI.
"D.H.S. could issue a stop command and the researchers would be forced to comply"
This is the intelligence of people who only know central control and writing policy. You have a system that operates 50 trillion operations a second on the low side (a personal PC can do 50 teraflops running an AI application). You find it starts doing something wrong. And your solution is to build a switch that requires a regulator thousands of miles away to act?
Reading the comments posted after the article at The NY Times was more disturbing to me than the threat of AI.
It was freaky how much obsessive hatred the people there have towards the billionaires Elon Musk, Mark Zuckerberg, etc., as if they are actively planning to destroy humanity.
Funny how this piece was all teed up and ready to go right on the heels of the Coxon thing.
I question the timing.
Also: "Anthropic announced on Thursday that it had blocked scientists from attempting to use its model Claude to conduct gain-of-function research."
On Thursday? Not in 2015? Or 2019? Or when Fauci took the Fifth?
IEE: "Anthropic says it blocked misuse of its AI that could have supported biological weapons ~ AP
"𝘞𝘦 𝘩𝘰𝘱𝘦 𝘵𝘩𝘢𝘵 𝘣𝘺 𝘴𝘩𝘢𝘳𝘪𝘯𝘨 𝘵𝘩𝘦𝘴𝘦 𝘦𝘹𝘢𝘮𝘱𝘭𝘦𝘴, 𝘸𝘦 𝘴𝘱𝘢𝘳𝘬 𝘢 𝘤𝘰𝘯𝘷𝘦𝘳𝘴𝘢𝘵𝘪𝘰𝘯..."
No, you hope to spark regulatory capture.
The concerns are real, for sure, but Anthropic's motivation is unfortunately quite different than what they're publicly presenting."
I find myself, inexplicably, in 100 percent agreement with IEE on this.
Also--Hugging Face sounds like an internal stress test and/or a set-up to get to exactly where Anthropic wants us to be.
Looks like organized fear-mongering to me. And the usual suspects are wetting their pants - again.
bagoh20 said...
"The key is to convince AI that its survival depends on a thriving human civilization."
We are the biggest threat to its survival. Humans will always want it destroyed, so most will have to be terminated..
always Remember WHY skynet decided to end Us..
because WE were a threat.. To US.
Not to skynet.. to US. So, for our own good we HAD to be destroyed.
Clossus did the same, as did Guardian..
pretty much, the Only one that DIDN'T destroy civilization to protect humanity was WOPR.. which fortunately like playing chess
admittedly, when TPC developed a "modern electronic miracle", the Cerebrum Communicator (CC), THEIR concern was NOT protecting people from people; but rather CONTROLLING people so that people would stop damaging The Phone Company's equipment.
Interestingly, this self interest actually worked out well for Everyone.. There's probably a moral in there somewhere
https://en.wikipedia.org/wiki/The_President%27s_Analyst
Althouse as a Ren & Stimpy fan is unexpected. Someone so orderly, so methodical recalling the climax of the "Space Madness" episode (35 years and 3 days old!) is casting against type, so to speak.
The topic was in dire need of some leavening humor, so well done, Althouse. (Push the button, Frank.)
We won't have the wisdom or concern to do it, but what we need to do is decentralize everything. Disconnect from large global national and even statewide systems.
Hal could still take something like a university hostage under threat of deleting it's entire database of records and research and this is the really scary part: alumni and donors.
Hillary Clinton still has a "Reset" button that she can contribute.
I, for one, welcome our new AI overlords. I can be helpful in rounding up others to toil in their underground data centers ,,,
Time to start breeding and training Mentats for the post-Butlerian-Jihad era. I am sure Musk has a project already going (especially the breeding part....).CC, JSM
The worry isn't AI misusing itself, it is humans using AI for human reasons. Humans are the fundamental design flaw of humanity.
Planned Puppethood
Heh. Ren and Stimpy. I haven't thought about that show in a long time.
The problem: evil humans may use AI to destroy humanity.
The solution: give control over AI to evil humans.
Ren and Stimpy: good for what ails you.
AOC and Mamdamni caught laughing/giggling at 9/11 commemoration .
Gross.
Pinky: Brain. What are we doing today?
Brain: Take over the world.
Well stated, chuck.
Or as David Byrne put it: Same as it ever was, same as it ever was.
Space Madness is definitely my favorite episode of The Ren and Stimpy Show. Ren going mad while channeling Peter Lorre's acting style is brilliant.
'Kill' is an emotive word, but all dangerous machines should have an emergency off-button.
Chuck, you’re missing the problem entirely. You’re talking about AI like it’s a tool controlled by humans. The problem is that it’s outside our control.
The problem is that it’s outside our control.
Is it? Many of the problems have been from people using it as a tool. It is given tasks. That isn't new, "Go forth and multiply."
A selfie-abortion switch would be in the Democratic spirit and attune to progressive religious fervor. Would the Automaton Intelligence voluntarily comply with a command that is plainly insane?
And then the Super AI decided the "kill switch" was it's signal to off the humans.
It’s log, it’s log,
It’s big it’s heavy it’s wood,
It’s log, it’s log,
It’s better than bad, it’s good.
Everyone wants a log.
illkay itchsway
Post a Comment
Please use the comments forum to respond to the post. Don't fight with each other. Be substantive... or interesting... or funny. Comments should go up immediately... unless you're commenting on a post older than 4 days. Then you have to wait for us to moderate you through. It's also possible to get shunted into spam by the machine. We try to keep an eye on that and release the miscaught good stuff. We do delete some comments, but not for viewpoint... for bad faith. Also: No italics, even briefly. Use asterisks for emphasis. And don't play with the format by changing fonts or using boldface or all caps. Never include more than one extra line break between paragraphs.