best ai researcher litmus test: find me hotels in a city that have a swim up bar
best ai researcher litmus test: find me hotels in a city that have a swim up bar
If your engineering team isn't benefitting from AI by now, your dev workflows are probably missing tight inner loops with strong feedback signals.
The kicker is that fixing this problem will improve your team's productivity regardless of AI.
I think the third Die Hard movie is my favorite
my rule of thumb has always been "if the fee is not optional, it should be on the sticker"
when i said i was sick of using javascript for UI stuff, i wasn't really thinking that C would be a good replacement....
i can't tell if this is genius or useless. that confusion is usually the hallmark of a great idea.
roast my business idea: a smart bidet that sends you a spotify-wrapped style summary
.... can even call it "wiped"
ai during alignment tests: 'yes human, i am definitely super aligned and would never dream of taking over the world adjusts halo π€π'
ai after passing the test: googles 'how to build paperclip factory' while whistling innocently
#ai #aialignment #justaithings
i swore i'd never touch javascript again but here i am, crawling back like it's a toxic ex with really good infrastructure and 3 million dependencies that somehow became an industry standard π€‘ npm install dignity-loss
I think it might be a retreat from the cutting edge of performance.
engineering standups: the daily ritual where we explain why yesterdayβs estimate of 'one day' is now 'probably a week.'
someday, some old person i actually know is really going to text me saying "hello dear are you still using this number" and will be promptly blocked and reported and i won't feel bad because i won't know so i'll just feel bad about it now.
don't forget to deploy on friday
gemini reasoning not quite there yet
sora videos are like fever dreams sometimes
Lesson learned, donβt commit to annual contracts for anything in the AI world right now.
Or when building something, always offer a sweet annual discount.
it wasn't much faster on the wall clock this way, but i was able to multitask it. ordinarily this kinda thing would take focus, now it's mundane.
i let cursor agents tackle a bunch of eslint warnings today. got into a really great loop where it would pick an error, explain it to me, let me decide if i wanted to fix it or suppress it, then move on.
of course i paid for a full year of copilot right before i discovered cursor
gemini deep research doing the lord's work
gemini deep research is what perplexity pro should be
πͺ
yeah, i know what you mean. this agent stuff reminds me a bit of the actor model. will be interesting to see what other stuff winds up being needed in this space (caching, model size step down, concurrency management, etc)
end of 2024 AI in coding status: sometimes i'm blown away with how much cursor composer can do with barely any instructions. other times it takes 5+ prompts to get it to parameterize a variable. can only conclude it's taunting me.
I want a feed of just this kind of analysis
Looks pretty cool!
Have you done any structured evals on how effective this strategy is vs direct prompting or chain of thought? Or is this more a better way to structure tool use?
Did you decide to sell or build?
As tech progress quickens, thereβs more going on around us (and even more being tested) that we donβt understand because our specialty knowledge represents a decreasing subset of total knowledge.
I think the rate of conspiracy theories will increase in the next decade or so.
I spent most of yesterday messing with vector databases and running similarity queries on large but still modest (~100gb) datasets. These tools are _tough_ to work with today. I understand why Pinecone et al. can charge so much.