David Spiegel

Anyone have a really good claude skill they swear-by? I mostly trust the models and harnesses to do a better job at than I could do with my own skill, and a lot of skills online tend to add more rigidity than I'd like and you have to really read through them to make sure they're really what you want. I feel like it's often easier to just prompt really well, let the model figure things out, without introducing the rigidity of a skill that isn't tailored to your project + filling up more of your context.

@a_musingcat hello cat

Is Adtech Twitter still a thing? I feel like like the past 4 years of mass layoffs and consolidation definitely took a toll

im so shadowbanned right now, fml. i'm stuck posting on my alt now because i get far more reach over there

@0xSero Droid is great and super underrated — pricey, but I think the better outputs and time saved justify the cost.

Droid Missions are legit my favourite development in agentic workflows, I'm going out for fun in London tonight and I know I will come back to a fully functional refactor. 24 hours left, 3 people will get 100M factory credits, that's up to 800M tokens x.com/0xSero/status/…

Factory Droid is slept on tbh, feels like a hidden gem Insanely good agent harness, they just don’t market themselves as well as they could imo, but somehow Claude is better in Droid than it is in Claude Code.

@a_musingcat At least he can make images now

@leozaroff “Number of Bitcoin traded” is an interesting (and questionable?) growth metric lol Seems they turned out alright though..

the craziest thing about the epstein files that nobody noticed is that you can read all the pitch decks and investor updates of the companies he invested in like coinbase

Claude Cowork is so close to being a good product, but has a lot of major issues or things lacking tbh. don’t get me wrong, it’s often very useful, but it’s kinda underpowered - can’t set level of effort - chat is very buggy - even if you clear/reset context in the same session, the session gets pretty dumb after compacting and you eventually just have to start over with a new session. I might just use Claude Code as a better/more advanced version of Claude Cowork

@alexabelonix Thank you! Will check it out :)

@davidspiegs nice breakdown. we’re recruiting builder ambassadors for paid campaigns. fill the pinned form

got my very first star on github :) i know it ain't much...... but felt very validating

@bihanmahadewa @rtwlz West Village is the part of Marina right next to Palace of Fine Arts

@rtwlz west village is cow hollow or what?

About to annoy so many people

@rtwlz @0interestrates There was an attempt…….. but wrong on both accounts, sadly

@antoniogm Can’t beat those views… do you miss SF or happy with the move?

Valentine’s Day hike, looking south toward San Francisco (whose skyline you can see).

Here's a simple, but important tip for any ChatGPT power users (you'll thank me later) Don't use native MacOS app, use the browser version. Browser lets you use 'Extended Thinking' on GPT's 5.2 Thinking model, in-app does not. Far better outputs; difference is like night and day.

@creatine_cycle Lmao Yeah the whole ragebait “lemme end your whole career” marketing move ain’t it.

okay they took it down. good

okay that's it it's over. shut it down. it's clear we can't be trusted with the god machine without doing and saying the most off-putting things imaginable. insanity

Been using it for some career advice + resume strategy work — GPT’s answers are more thought-out than Claude. Claude writes more naturally, however, and easier for a human to digest and comprehend the outputs. GPT is most inclined to talk about every point I mentioned and think strategically over it. Claude tends to be more focused and easier to read, but sometimes at the expense of not giving answers that are *as* good. More action-focused, but less strategy focus.

I know that OpenAI is getting dunked on here a lot lately, but I think GPT 5.2 Thinking (with extended) is the *smartest* model out rn. Probably less so over longer sessions vs other models, not as capable as other models when it comes to actually building + tool use But it’s the best reasoning imo. Most strategic, most thorough. If you want a really smart consultant in any niche, GPT is the place to start. If you want to build, get things done, Claude’s your “guy” If you want a one-shot landing page, Gemini (don’t use it for any further pages).

@codyschneider I really want to be optimistic, but even the best models today aren’t as capable as people make them out to be. The longer they run without human intervention, the more they tend to butcher things.

crazy tbh