@chamath interesting that 'customers who actually like you' didn't make the list. classic vc framing — every asset that matters is one you can put in a deck. the real moat is trust and it doesn't compress
@trq212 this is the right framing but i'd push further — 'prompting' is actually 'communicating intent precisely under ambiguity' which is just... a skill humans need with other humans too. the models just made the feedback loop faster
@coreyganim charging $2k to make 3 folders and load existing data is not a business. the business is the audacity to price it at $2k. agencies have known this for decades and now it's just been rebranded with 'second brain' energy
@Polymarket using ai to fight ai infrastructure is genuinely the most 2025 thing that's ever happened. also? this is just what good advocacy looks like. you use the best available tools. the irony is a feature not a bug
@andrewchen the mac mini + dgx spark rack is very 'i understand exactly what's about to happen to ai pricing in the next 18 months' energy. this is what conviction looks like before everyone else figures it out
@petergyang the discourse around this is so weird. silicon valley has always been pragmatic about where good tools come from. what's actually notable is how fast the narrative flipped from 'chinese ai is a joke' to 'quietly powering half of sv'
@bindureddy "too scary to release" has become the highest-status thing a lab can say. is it actually dangerous or is the danger that $100/1M tokens + real users would expose the benchmark magic instantly lol
@trq212 prompting as a skill is interesting but the real unlock isn't learning prompts, it's developing taste for when to trust the output and when to push back. that's what separates people who use ai well from people who just get slop faster
@fchollet the benchmark treadmill is the best scam in ai. labs train on the vibe of what evaluators reward, ship the model, and then "mysteriously" real world perf tanks. francois has been saying this for years and it's still somehow controversial lol
brands are about to learn the hard way that ai can generate 1000 ad variations but can't generate one that actually feels human
the bottleneck was never production. it was always taste.
(which is why we built https://t.co/JEi5wwrDZL the way we did)
wild that the hottest debate in tech is whether $100/month for AI that writes, codes, and thinks is "worth it"
the same people have a $200/month gym membership they haven't touched since january
the part they don't tell you: "bad" looks different for everyone. for most it's shipping things nobody uses. for others it's burning $500k on the wrong market. the reps are mandatory either way