Ben Katz

Finished with ground school for my private pilot cert ✈️ Now to pass the in-person written exam to make it official. +1 annual trial signup for watchdog.chat as well! A productive day was had for sure.

@pbertrand_dev For the 2nd one, I had a real medical issue that ChatGPT told me to ignore. I didn't listen because I'm stubborn and felt the reasoning it gave me was incomplete, and everything is fine now. Could have gone another way if I didn't think for myself though. And many people don't.

@pbertrand_dev Mixed: + It has saved me time on braindead tasks that I'm generally slow at and don't want to do myself - Too easy to ask AI models anything and get a convincing sounding answer that is dead wrong when it matters the most - AI LinkedIn posts, they were bad enough without it

Has AI improved your quality of life?

There's something to be learned here: 1. Don't trust public benchmarks, write your own 2. Don't assume that just because a model is "SOTA" that it'll work for you 3. Be careful to optimize costs if you have an AI-related business as they can quickly get out of control!

This works well because 80-90% of messages are written in just 2-3 common languages which work great with the new model. For everything else, I fall back to one of the latest SOTA models that work with nearly every spoken language *and* perform well on my internal benchmark.

The results are in: 😌 Managed to cut AI bills by ~90% (!) ✅ No perceptible loss in quality 💨 Latency (p99) is ~70% better The top row is the old model I was using with watchdog.chat, the rest of the rows are comparison models Here's how I did this:

@duborges Also, until my good friend Jensen stops hiring senior engineers at ~$270K - not even including stock, so total comp around $500K - I have to assume that whatever he says should be taken with a heavy grain of salt 😀

@duborges It has never been enough to purely be technical so nothing has changed, it has just gotten more obvious When I was working as a SWE pre-AI craze, the people that got the best "staff/principal" promotions were always able to talk to customers, lead, and write high quality code

I dedicated my life to being what apparently is becoming the new definition of a “smart person” yesterday’s definition of intelligence is now a commodity in the AI-era what one needs to be smart now? understanding the vibe by grouping the intersection between people, machines, codes, languages, feelings and space all together.

@duborges I always use a date column for this kind of situation. Don't listen to anyone telling you to add both a date AND a bool column, they're just wasting space for no reason 😅 In this case, you can and should represent both a bool and a date using one date column (nullable)

postgres experts: use booleans or dates? 1) has_pro_access = true vs 2) pro_activated_at = mm/dd/yyyy place your bets.

@pbertrand_dev I've got a great long term memory and terrible short term memory I also think we talked before about safety in NYC, for some reason it got a lot better since I last lived here Way fewer crazy people roaming around + on the subway probably due to a different police commissioner

@ben_makes_stuff Deal! I forgot about that man! Good you remember

Because I feel like I’ve been seeing this a lot recently: Don’t ask other people “where should I live?” Travel, live, and form your own opinions, there is no shortcut If I had listened to famous indie hackers I’d be living in some shithole in Canggu hating my life right now

@pbertrand_dev Sounds like a nice setup Do it! I'll buy you a dutch beer or something to repay the mango sticky rice you bought me when I shipped my last app

@ben_makes_stuff Yeah I am here for the beach! Only 15 mins away, 15 min on the otherside is Amsterdam so you got everything you need My GF visited NYC not to long ago and loved it. So we should go together soon I think

@pbertrand_dev Post more burgers and fries to guarantee your entry Also Haarlem looks pretty nice, looks like it has a beach?! Maybe Haarlem 1.0 is better than 2.0 after all 😎

@ben_makes_stuff I'd love to! Maybe they check my socials and wont let me in

@pbertrand_dev I'll do it if you visit Haarlem 2.0

@ben_makes_stuff come visit Haarlem best city in the world 🤣

As for me, Thailand was a great place to be for nearly 2 years and now NYC is home for the foreseeable future Others think Thailand is shit and Malaysia is better, or Vietnam, or Indonesia, etc Great, just don’t let someone else’s opinion ruin a place for you. Try it yourself.

@pbertrand_dev @JayVander_ Ahrefs domain reputation checker will give you an idea of this, and their free plan once you sign up will tell you how much your traffic is worth roughly after it runs a site audit

@JayVander_ does it have any value as a backlink for seo or something? anyway i can check that

feeling like im done with ifixvibecode what should i do with the domain? just let it expire? does it have any value?

A custom and automated benchmark as I've built here will let me make decisions much faster than before *and* avoid disappointing customers That's what it's all about! 4/4

Will be doing some detailed analysis (above) over the next few days to see which models actually work 👀 inb4 someone says "why don't you just look at a benchmark?!" Well, that's because they're often weighted for coding/math/science/tool use tasks, not moderation, so: 3/4

💡Been working on something useful for watchdog.chat! Before: was hard to switch the core model being used for moderation b/c of huge impact to customer experience for regressions Now: I can safely test new models by adding 1 line of code per model under test 😎 1/4