Brad L.

I'm starting to see signs other people are slowly coming the realization that AI companies have more enshittification cards in their hand. I attended an Anthropic training hosted by Anthropic employees. During the presentation, on almost every slide, the trainer reiterated "If you're experiencing issues with the quality or correctness of the output, you can solve that by giving Claude more access to your internal documentation and intellectual property." Almost every answer to every question included "and make sure you give Claude full access to your files and identity so it see everything you see and act as you." As a nearly 30yr veteran of infosec, I assumed this was an advanced adversary attempting to exploit us. 😂 The AI companies _WILL_ eventually operate solely by extracting value from all their customers and business relationships for their shareholders. We already know what this looks like. Anyone bought an Amazon Basics product? Amazon tracks products that do well, they then rip off the design of those products and launch a cheaper version under the Amazon Basics brand. "But, Brad, AI companies are more ethical than.. no wait, I heard it." Right, so, let's say you're Cursor. You announce a profit for the first time in your history. What happens next? OpenAI raises your rates to nullify your entire profit. This happened, in real life. You're a math researcher working on a research problem with a prize using OpenAI for assistance. What happens next? OpenAI rips off all your work and uses it to train an expert model to solve the problem and win the prize. This happened. Wake up, sheeple! 😂 If you have a good idea and use _any_ thirdparty AI service to create it, you are handing your work over to the AI companies and your competitors. Legal precedent is AI generated code is not IP because a human didn't create it. But every line of code Claude or Codex generates for your proprietary application is stored on their servers for as long as they afford to save it. As these companies pivot to IPO and beyond, just take a moment to consider the history of Silicon Valley and what their next steps are for further monetizing the data they're collecting on you. #fuck_ai #llm #privacy #security #copyright

#protip: Instead of using AI to generate a slide deck, put the prompt you'd use on the slide with a picture of a cat. Your audience will thank you. Another reminder of Oliver Burkeman's "Interest is Everything" https://ckarchive.com/b/92uzhnh695dods3roomrdfzod7d33hwh9nk2g At work and at conferences, even small, local conferences, I'm seeing more and more AI generated slide decks. I am pretty sure I've seen more slop decks than hand made decks in the last 3 months. I heard one speaker mention it cut the time to create the deck from 20 hours to 15 minutes.. Yeah? No shit! You have no idea what's on the slide you have up nor do you know what's on the next slide. This isn't just techbrosephs. I've seen experienced, well-intentioned speakers at OSS conferences on stage, armed with a slop deck and a confused look when the slide deck doesn't match what they're talking about. For me, as a audience member, it should be the first time I see your slides. But for you, as the speaker? You shouldn't act surprised, or have to read all the 18pt font bullet points on the slide for the first time as they're displayed! Stop the slopshow! #fuck_ai #speaking #publicspeaking #public_speaking_101 #llm

@ainmosni@social.ainmosni.eu Most of the pain in YAML comes from the automatic type casting in implementations. I loved YAML until I worked with it in any language other than Perl. Perl famously doesn't have "types", the context determines the type. It's controversial, but it made working with YAML, well, simple and predictable.

YAML is attempting to serialize and abstract (ie, make computer things more human friendly). If you remove the implicit type casting from the deserialization process, I think things get much easier to deal with. Aside from numbers and strings, I don't think YAML or the YAML libraries should deserialize into native types. If you want to use the value as a datetime field, do the type cast explicitly in code. The Norway problem disappears. If you want a boolean, you cast yourself, explicitly. The bad data problem becomes the application's domain to handle. If you expect a Datetime in a certain field, the programmer is responsible for the type casting and error handling, not the library.

I think the serialization aspect of YAML was naive. At this point in history, looking at how YAML is used, it's clear that serialization was a mistake and the minority use case. Most YAML is config. Config doesn't need type casting, it needs application specific validation. The serialization might save some devs a few keystrokes, but at what cost?

I say this as someone who loved YAML when working with it in Perl. I started working with pYAML in Ansible and it broke me. pYAML is fucking terrible. It's slow, incorrect, and volatile. And in the context of the Ansible, the types are fucking worthless anyways.. Inevitably, the YAML serialized number gets converted to a string somewhere between the deserialization and the jinja2 because you have to use quotes to use inline jinja2, so all my fucking templates have to |int or |float anyways. It's mind numbingly stupid.

Apologies if that's too much. I think I may need to talk to a therapist about my YAML trauma.

@nxskok@cupoftea.social When I first started doing public speaking, the general rule of thumb was 1 hour of preparation for every 1 minute of talk. A 20 minute talk would take about 20 hours to prepare. Over time, that dropped a bit, but not by much. I'm not a world class speaker by any means, but the attendees time means something to me. I want to put in the time ahead of the presentation to make sure they're getting their money's worth out of the conference. I do not want to go to conferences where the majority of the speakers are using AI for slide decks and graphics. It's a signal that they don't care about my time. It's the gym metaphor. In the case of preparing a presentation, the point of the exercise is not to simply pick up the weights and move them once. It's to do that over an over again to build strength. Don't use the fork lift if the point is the process of lifting the weight.