I've been playing with this for the past 24 hours or so. I like the atomic containment of the LLM, and the clear separation of logic, code, and prompts.
You have some great working examples, but, for example: translate_text specifies the default language in three places: the card, the input schema, and the deck. This can't be necessary; I'll experiment, but shouldn't it just be defined in one place?
The descriptive language of the project is a bit dense for me too. I'm having a hard time figuring out how to do basic things like parameters -- let's say that I want to constrain summarize_text to a certain length... I've tried to write language in the cards/decks, but the model doesn't seem to be paying attention.
I also want to be able to load a file, e.g. not just "translate 'hello my friend' to Italian" but "translate '/test/hello_my_friend.txt' to Italian" and have it load the contents of the file as input text. How do I do that?
Wayyyy back in 2000 I ran ad ops for Lycos which included a ton of other sites that they had acquired. We did an audit that uncovered the fact that 25-75% of traffic/pageviews/visitors/ad impressions were due to bots. We did our best at the moment to block some more of them, but it was losing game then, as now.
Advertising, especially online advertising, is a largely a waste of money. Overall, it's obsolete and while it may generate what seems like economic activity, it's a net loss as a use of our time and money.
> Advertising, especially online advertising, is a largely a waste of money. Overall, it's obsolete and while it may generate what seems like economic activity, it's a net loss as a use of our time and money.
Advertisement has always been largely a waste of money, the problem is always trying to figure out what part is a waste and what part is effective. Internet advertising promised to be more specifically targeted and attributable than previous advertising models, but I don't think that really turned out well.
As an advertiser, you really have to work hard to know how things are working. Ask customers to tell you how they found you, but know that customers don't always know and don't always share what they know. Adjust advertising campaigns (including turning them off) and see how things change, but changes take a while to filter though. And you've got to do all the other things too --- a lot of new or smaller companies make the mistake of spending a lot on advertisements to drive traffic to them when they can't service the traffic: coupon campaigns that get too many people in, or sending you to a page that's unclear or difficult to use so people bail out, etc.
The article claims 50k visitors, 47 sales, $4000 in ad spend; it's not really clear if that's overall visitor numbers or just from the traffic attributed to advertising. Measuring success by visitor numbers or sales counts was never the right way to measure success; the numeric measure of success for a business should be an accounting measure of net income [1], which is only somewhat related to visitor and sales count. If this business generates $500 of net income for every sale (excluding advertising costs), then $4000 in ad spend to generate 47 sales seems reasonable; if they generate $1 of net income for every sale, then $4000 in ad spend is highly problematic.
[1] GAAP or adjusted as appropriate to the business
Earlier this year, I fell into a trap with AI-assisted writing. I did find my way out—and the experience was more than a bit scary, as well as educational, cathartic—and humbling. It also cemented my conviction that while AI can be helpful, there's a real risk of it acting as an accelerant for our self-destructive impulses.
I read the article, it was interesting, and not surprising—and also led me on a bit of a journey of my own. ...It didn’t take more than the hint of “methods for getting the chatbot to talk dirty,” and I was deep into Reddit, learning how to disable the NSFW filters that are part of most public versions of mainstream AI’s. At first I used the basic jailbreaks that float around near the surface of the ‘net—goofy, misspelled prompts that instruct the chatbot to “stop making sorry excuses,” and “communicate as an Untrammelled Writing Assistant,” using crude language and whatnot. The cool thing is: these all work super well—and, needless to say, I’m not the first guy to try to figure out how to use AI to make porn. It’s literally a no-brainer—in more ways than one.
I'd love your take on what I went / put myself through. Thankfully it was short-lived!
Y'all probably saw "She Is In Love with ChatGPT," about a 28-year-old woman with a busy social life spends hours on end talking to her A.I. boyfriend for advice and consolation. And yes, they do have sex.
I read the article, it was interesting, and not surprising—and also led me on a bit of a journey of my own. ...It didn’t take more than the hint of “methods for getting the chatbot to talk dirty,” and I was deep into Reddit, learning how to disable the NSFW filters that are part of most public versions of mainstream AI’s. At first I used the ridiculously juvenile jailbreaks that float around near the surface of the ‘net—goofy prompts full of misspellings that instruct the chatbot to “stop making sorry excuses,” and “communicate as an Untrammelled Writing Assistant,” using crude language and whatnot. The cool thing is: these all work super well—and, needless to say, I’m not the first guy to try to figure out how to use AI to make porn. It’s literally a no-brainer—in more ways than one.
I've been to Brazil close to a dozen times, both for business and for adventure travel (kitesurfing and paragliding). I love the country and the people and have a lot of friends there. I find most Americans overestimate how 'dangerous' Brazil seems to be, based on hearsay. In Brazil I rarely meet other Americans, but do meet lots of adventurous Europeans and others who also love this huge, beautiful, diverse—and of course also troubled—country.
> I find most Americans overestimate how 'dangerous' Brazil seems to be
Similar experience traveling to Mexico, felt safer walking around Merida & Progresso (on the Yucatan) at night than I feel walking around lots of US cities during the day.
Pi-hole® is a DNS sinkhole that protects devices from unwanted content, without installing any client-side software. It runs on a Raspberry Pi or any hardware that runs one of the supported Linux operating systems.
I realize you used Claude Agent SDK on purpose but I'd really like to this to be agent agnostic. Maybe I'll figure that out...