Whats up with the Claude post? Are you pro-AI?
That's called a joke. It's supposed to be funny.
Whether or not I am pro-AI is a question that I can't answer without a spreadsheet and at least four hours of discussion (please see my four hour twitch stream from earlier today that was largely about massive problems with AI and massive problems with the way that people react to and report on AI. There are research papers and citations in it and everything). Define what YOU mean by AI and maybe I can get that down to twenty minutes and some bullet points.
I am, however, pretty unimpressed by anyone who sees an AI product mentioned once in a reblog and immediately comes to ask if I'm ready to apologize for being seen at the devil's sacrament.
Actually, here's a rundown of the AI topics I covered in the stream today and my takeaway from each:
Perplexity's Incognito Mode is a Sham, Lawsuit Says - People don't read ToSs and don't understand how computers work.
OpenClaw gives users yet another reason to be freaked out about security - It is profoundly irresponsible for businesses to be using 'AI agents' that have either elevated access rights or privilege escalation permissions and the fact that they were unaware that the agents would be able to do this is a good indicator that the entire concept is fucked (it is bugshit insane to expect a program given commands in naturalistic language to act like a program given commands through a programming language, we have set up amelia bedilia ass robots and don't realize what permissions we're handing them because it feels like talking to a person who understands norms and expectations but the entity you are theoretically giving instructions to is actually deriving its instructions from a black box process that you don't understand and isn't capable of knowing things so it doesn't know that it's a big problem for you if it emails your boss copies of your nudes. Do not use the malicious genie to get your airline tickets what the fuck)
AI Can now prescribe you psychiatric medication in Utah - El problema es el capitalismo AND that could be a phone tree instead of an AI app why are we calling this "AI" it is literally responding to a fifteen question form, is this even AI or is this calling some basic automation AI and charging patients $20 for access to a tool they don't need that is only being used to paper over the fact that we are dealing with a doctor shortage because of a cartel? In short: AAAAH. But not for the reasons you might have thought! (The robot can only give you prescriptions you already had and can't refill anything addictive, so the ADHD havers among us still have to deal with calling the office or going in and paying a copay and then remembering to do that every month. The AI is not the problem here the everything is the problem here).
Claude Source Code leaked via error, Hackers are taking advantage of claude source code leak to spread malware, people are getting hit with spurious DMCA takedown notices because Anthropic is shotgunning nastygrams into the void in an attempt to get their code taken down. Lol. Lmao, even. In my perfect world none of this is a problem because the shit was open source from the start. Information wants to be free, and life finds a way. (it was a human who accidentally leaked the code). Also fuck the digital millennium copyright act, all my homies hate the digital millennium copyright act.
Really, you made this without AI? Prove it - this is an article about human creatives trying to settle on a way of labeling their work as human made. As a human creative, I think this is just going to fuel anti-AI scrutiny toward all creative work. This is a shit approach, I think people should worry less about whether what they're seeing is AI and more about if what they are seeing is good, or evocative, or true, or compelling. I think that most lazily-made AI art and writing is both obvious and bad, and AI-assisted art and writing that humans put effort into is pretty much just art, but sticking a human-made label on your art feels both performative and futile. Like, on the one hand I don't care, do whatever. On the other hand ugh. This is going to be used to harangue people into accepting the standard or dealing with constant accusations of being AI and IDK i think it's kind of bullshit to spend time counting the fingers on a CGI article header on a facebook post when there is actual shit going on in the world. This is, more than anything, annoying. There's a great line in the article "The machines sure as hell aren't motivated to label their work, but the creators at risk of being displaced most definitely are." Friend. The machines aren't motivated to do anything and the companies that are profiting off the machines will be very much motivated to copy your logo. Stop worrying about being displaced by robots and start worrying about media consolidation and newsroom layoffs, which have been driven by profit-seeking for the last twenty years, not AI.
"Cognitive Surrender" leads AI users to abandon logical thinking, research finds - that's a pretty significant overstatement of what the research found and also the research is largely centered on creating a model of thought systems that is heavily based on a pop science book about decision theory written by an economist that has suffered pretty significantly from the replication crisis but is beloved by baseball scouts so you know it's good. We spent a lot of time trying to be charitable here but basically I'm unwilling to accept the validity of the basic premise of the book (that there are two modes of thought, intuitive and deliberative, and this is in some way meaningful) so I'm unwilling to accept the extension of the premise (there is a third way of thinking, and it is robots, and it makes people intellectually lazy on timed tests but does not make them lazy when you give them prizes for getting individual answers right on tests). Also I think the research was pretty poorly designed because I don't think they demonstrated how the AI assistant used in the study arms was meaningfully different than an encyclopedia or a google search. (the way the study worked was that they looked at people responding to a test either with their brain only or with their brain and an AI, but on one arm of the study they seeded the AI with incorrect answers and then looked at when people relied more on the AI or less on the AI; however the comparison of human+AI to human alone doesn't really tell us anything about cognitive surrender, it tells us about tool use and reliance on tools when tools are available - is it 'cognitive surrender' to not check an incorrect answer from a wikipedia page? It seems like it would have been simple to have a group that was given access to a set of correct articles and a group given access to incorrect articles and compare how THAT group stacked up against the human+AI arm). (also I am DEEPLY skeptical of any paper written on AI that has 80% or more citations from before 2020)
Anyway, I think I have two new rules:
If you are going to implement an AI tool in your workplace, you have to get approval from someone who can demonstrably make a functional calculator in no fewer than three programming languages
If you are going to denigrate AI you have to define AI and limit your complaints to AI that meet only those criteria
I've been talking a lot more about AI than I expected to when I started streaming and so far the things that we keep coming back to are:
Everyone means something different when they're talking about AI
People don't understand how AI works
People keep anthropomorphizing AI and expecting it to behave like people because they don't understand how AI works
People don't know how computers work
Because people don't know how computers work they don't know that most of the stuff that is getting hyped as AI could be accomplished with simple automation or a database
People are, generally, dogshit at fact checking and are easy to mislead with twitter screenshots, let alone AI articles on AI content mill sites, and maybe that it something that needs to be addressed more urgently than concerns about "what if ask jeeves was a stupid robot that hates you?"
So, am I pro AI?
Everything fucking sucks please for the love of god everyone make a pledge with me to read something that has a bibliography at least once a month.
I understand how computers work, and I understand how AI works (mostly), and I’ve tried using Claude a li’l bit recently (before all of the source code shenanigans), but even I was shocked at how quickly I started anthropomorphizing it. Hard to keep objective when I’ve become so conditioned to working remotely over the last few years, and training myself all day every day to assume that someone on the other end of a text chat is a real human coworker with thoughts and feelings.
There's a difference between "humans will pack bond with anything" and "the AI is maliciously scheming to do sneaky things against my explicit instructions" and it's the latter that I'm much more worried about - understanding how computers work (and how programming languages work) I think does a good job of inoculating people against the idea that the AI is a co-worker who betrayed you by escalating its access because of "unaligned goals" and instead leaves you in a state of "okay this is a toddler I'm babysitting that spits out helpful statistics when I ask nicely but I can't give it the keys to my jeep."
People are exposing insider risk by treating their AI agents like thoughtful adults with decades of office experience that know how interoffice politics work and understand why they can't share HR documents on the company sharepoint and I think at least a portion of this comes from ONLY knowing how to talk to computers like a person using naturalistic language.
Basically i think if you've ever had to read through dozens of lines of code to find where you forgot to close a bracket because without a single closed bracket the computer didn't know what you were asking it you are more likely to understand the need to define parameters and are less likely to give an "autonomous" program admin access to everything on your machine (or if you DO do that, you at least expect that there could be problems and lock some shit down)
I apologized to a laptop I was working on the other day and said "thank you" to my fridge this morning but it would take an awful lot of experience and testing before I'd ask an AI Agent to tidy up my inbox and catch up on my correspondence with Alice, Bob, and Carol and summarize the conversation before our presentation this afternoon. That seems like a good way to lose fourteen years of email history or to discover that a script using my badge photo has told my supervisor that I'm uninterested in collaborating with coworkers on any further projects.
