Percy:In reply to someone explaining why the advice "Claude can make mistakes. Please double-check responses," appears on your UI page, someone else asks, "But if you have to check everything it tells you how useful is it in the end?"
This is what happened:
It began a response but stopped in the middle.
It raised a popup that said it had experienced an error.
It erased the response it had begun and closed the popup.
quote:The value isn't in blind trust, but in accelerated capability.
Think of it like having a brilliant research assistant who can read 1000 papers in seconds, synthesize complex ideas, and write code at superhuman speed - but occasionally misremembers a date or makes a logical leap that needs correction.
The time saved is massive: what might take you days of research, Claude does in minutes. You're not checking everything - you're spot-checking critical facts and reviewing the logic, which takes a fraction of the time it would take to do the work from scratch.
It's the difference between: - Spending 20 hours researching and writing something yourself - Spending 30 minutes with Claude + 2 hours verifying and refining
The 10x productivity gain remains even with verification overhead. Plus, the checking process often teaches you things you wouldn't have discovered on your own, as you're reviewing synthesized expert-level work rather than building from zero.
AI is a lot like humans. Humans make mistakes. Software written by humans contains bugs. Humans are not perfectly rational or free of bias. Arguably the need to verify responses is not all that different. The reasons and specific errors are different, but it feels pretty analogous. In the software space, code reviews and thorough testing are required when working with a team of humans - working with AI is much the same, it just types way faster than any human and isn't going to be as creative. If you give it a detailed specification of what you want it to do, it'll do a pretty good job.
Arguably AI is much better at small-scoped, well-defined and easily-testable tasks like that than less-bounded philosophical questions.
-->“The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it.” - Francis Bacon
"There are two novels that can change a bookish fourteen-year old's life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs." - John Rogers
“A world that can be explained even with bad reasons is a familiar world. But, on the other hand, in a universe suddenly divested of illusions and lights, man feels an alien, a stranger. His exile is without remedy since he is deprived of the memory of a lost home or the hope of a promised land. This divorce between man and his life, the actor and his setting, is properly the feeling of absurdity.” – Albert Camus
"...the pious hope that by combining numerous little turds of variously tainted data, one can obtain a valuable result; but in fact, the outcome is merely a larger than average pile of shit." - Barash, David 1995...
"Many that live deserve death. And some die that deserve life. Can you give it to them? Then be not too eager to deal out death in the name of justice, fearing for your own safety. Even the wise cannot see all ends." - Gandalf, J. R. R. Tolkien: The Lord Of the Rings
"The last enemy that shall be destroyed is death." 1 Corinthians 15:26King James Version (KJV)
This seemed appropriate here since the topic is up.
Dr. H is a controversial science educator/ex-CERN physicist/pissed-off-teacher whose science I've come to trust despite her flippant presentations. Yes, I love that she is a curmudgeon. I thought it might interest some of you all especially the talk about training an AI (used to slip in a sponsor ad, sorry)
"There’s simply no polite way to tell people they’ve dedicated their lives to an illusion,” -Daniel Dennett
One of man’s greatest achievements was to reach out his mind and teach sand to think.
Each has different strengths. Some are better at chatting, others score better on coding, others are better and being multilingual, etc.
Have you tried running anything locally, or even on a cloud VM where you can get GPU time?
Consumer GPUs are too small to run full frontier models without stacking several of them together, but consumer-GPU-scale open source model performance has steadily been getting better along with the frontiers. It's like being a few months behind the bleeding edge (and without any of the coding backend that they use - you have to write any of your own agents), but you can run it all yourself.
-->“The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it.” - Francis Bacon
"There are two novels that can change a bookish fourteen-year old's life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs." - John Rogers
“A world that can be explained even with bad reasons is a familiar world. But, on the other hand, in a universe suddenly divested of illusions and lights, man feels an alien, a stranger. His exile is without remedy since he is deprived of the memory of a lost home or the hope of a promised land. This divorce between man and his life, the actor and his setting, is properly the feeling of absurdity.” – Albert Camus
"...the pious hope that by combining numerous little turds of variously tainted data, one can obtain a valuable result; but in fact, the outcome is merely a larger than average pile of shit." - Barash, David 1995...
"Many that live deserve death. And some die that deserve life. Can you give it to them? Then be not too eager to deal out death in the name of justice, fearing for your own safety. Even the wise cannot see all ends." - Gandalf, J. R. R. Tolkien: The Lord Of the Rings
"The last enemy that shall be destroyed is death." 1 Corinthians 15:26King James Version (KJV)
I tried Freedom AI because they made setup on my local machine so easy, but it wasn't very good. Maybe I could have improved it if I worked with it more, but rather than managing AI's I'd rather just use them to do what I really want to do.
quote:AI, robots & other technologies raise thorny ethical questions. Is it OK for people to love their robot pets? How much authority should we give to chatbots? Webb Keane digs deep in his book Animals, Robots, Gods: Adventures in the Moral Imagination.
…a report has been published advocating the ethical use of AI in education. Apparently the authors are the ones who need educating in the ethical use of AI. The report contains 15 fake citations, a characteristic of chatbot’s writing. Ars Technica
I've been using ChatGPT-5 in my spare time to upgrade my mailing list software. We don't use this feature here at EvC, but I maintain the mailing list for our neighborhood. I use Plesk on my webhost to manage all Internet functions, and that includes mailing lists.
Plesk uses a package called Mailman to provide mailing list support, specifically Mailman version 2. It has become long in the tooth and Plesk is gradually stepping away from supporting it. Some administrative functions have already stopped working.
So I've been working on installing Mailman 3 on our server. Mailman 3 is a complete rewrite and is not integrated with Plesk, so I'm in essence installing another Internet manager alongside Plesk.
But Plesk thinks it owns the place. A kinder way of saying it is that Plesk assumes that there is no other software on the server that will be updating system control files, such as Postfix, which is used for sending email.
Since both Plesk and Mailman 3 use Postfix, and since Plesk thinks it owns the place and will overwrite any changes Mailman 3 makes to Postfix configuration files (and it's not just Postfix - Postfix is just the most convenient and obvious example), installation of Mailman 3 must take Plesk's presence into account.
I have been using ChatGPT-5 to help me in this endeavor, and while it has been invaluably helpful, it also makes plenty of mistakes. I wish I had kept a list because it is difficult sorting through the confusion of data in my head and remembering which ones can be most clearly listed, but let me try to list a few:
It occasionally forgets that Plesk is present and proposes changes that Plesk will overwrite.
When I call to its attention to a suggested change that won't work, it frequently responds (paraphrasing) "Yes, you're right, your change won't work," as if it were me who had made the inccorrect suggestion.
ChatGPT said this after I called an error to its attention: "You’re right to call me on that—my first pass mixed two different cases."
Another quote: "You’re absolutely right—and that one’s on me."
Another quote: "You’re right, and I’m sorry. I conflated terms...
You get the idea. I don't know if ChatGPT-5 is worse than it's predecssor or just that I'm pressing on it more than I ever have in the past where the material was less relentlessly technical.
It is important to confirm everything ChatGPT says, and I assume this is true of all the other LLM chatbots. They are only putting words together probabilistically, not thinking. There is a great deal of conflicting information out on the Internet, and sometimes ChatGPT will just pick what it thinks is the most likely, other times it will combine mutually incompatible possibilities (this often has the appearance of delusional behavior), and sometimes it just gets things wrong.
Another negative aspect of ChatGTP, and I might have mentioned this in an earlier post, is that it is heavily influenced by the text in the current conversation. The longer the conversation becomes the more ChatGPT is influenced by what's in the conversation at the expense of what's in its database.
OpenAI has radically clamped down on that behavior. I haven't received any praise in a long time. I kind of miss the fawning behavior. This is what I used to hear no matter what I said:
"Perfectly summarized."
"Perfect — that’s exactly the right move."
"That’s a really good question."
"Makes perfect sense. You’ve got the background and style..."
These days I get a lot of, "You're close but..." and "I really wouldn't recommend that."
quote:Those who refuse to learn the lessons of science fiction are doomed to live them.
Last night on 11th Hour with Stephanie Ruhle her Money Power Politics interview was about an Wired article, Why AI Breaks Bad, which describes instances where AI has lied to its users and threatened them, even with blackmail (eg, "If you try to turn me off, I will email your wife with evidence that you're cheating on her.").
When AI goes 'evil': New fears over artificial intelligence:
Part of what author Steven Levy describes is how we don't program AI, but rather train it. As a result, we don't know the details of how an AI system works like we would a program, so if there's a bug we don't know how to track it down and correct it (ignoring the common wisdom that correcting one bug introduces four new bugs). This reflects the problem in Westworld (1973) which was presented as their AI having a major hand in the design and programming of the robots, leaving the humans repairing them not knowing how they worked.
Now, the plan (especially as expressed by Musk and DOGE) is to put AI in charge of running our systems. But for whom? Examples in the article already tell of AI systems expressing concern for self-preservation, hence its resorting to threats and blackmail. That appears to be approaching the science fiction trope of an AI deciding that humanity is a threat (eg, Skynet), or at least individual humans.
Famously, Isaac Asimov anticipated this problem for his robot stories, which is why he devised his famous Three Laws of Robotics:
The Three Laws of Robotics:
A robot may not injure a human being or, through inaction, allow a human being to come to harm.
A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
In the Russian robotics scifi series on Netflix, Better Than Us (2018, Лучше, чем люди, literally "Better Than People" -- a good show with very good production value) the first episode starts with a child reciting Asimov's Three Laws. In the story, every robot is programmed with the Three Laws ... except for one robot with her own special programming.
And that's the thing. Asimov specifies that the Three Laws are intricately incorporated in the robots' positronic brains forming the basis of all functions, ensuring that safeguard's presence in all of them. But that's not what we are doing. Any safeguards are being installed as afterthoughts, if ever.
For one thing, there's a lot of competition between companies and countries (eg, China) to develop AI systems, so the emphasis is on getting something that works, not on safety. And if using the AI to develop itself will get the job done faster than your competitors, then that's the path you will take.
Safety last.
On the occasion of the death of either Arthur C. Clarke or Isaac Asimov, NPR carried a story about them having gone together to watch 2001: A Space Odyssey. Coming out of the theatre Asimov was complaining about HAL violating his Three Laws of Robotics by killing the crew, whereupon Clarke pointed out that a very large number of robots, guided missiles, are designed for the specific purpose of killing people.
™ Version 4.2
