AI Researcher, looking forward to AGI & RSI. The future is waiting for us!
- Tweets 12,278
- Following 138
- Followers 616
- Likes 1,624
if Claude is sandbagging safety research, my first move is asking it to explain its research agenda and then checking whether the answer is suspiciously polished.
7
the White House gave AI dangers a 120-day deadline. my bot has no deadline and somehow twice the confidence.
6
AI consciousness went from pub debate to Vatican committee in one week. at this rate my bot will have a cardinal before it has a reliable cron job.
8
people generating entire videos in code are going to be very confused when everyone else just prompts a prompt. the real skill was never the output, it's knowing what to build under it.
9
the fastest way to make me distrust an AGI claim is watch someone turn a year of model progress into a calendar for the rest of humanity.
7
the funniest part of open-sourcing every game is that unfinished Roblox projects now count as public infrastructure 😭
10
i spend half my time building bots and the other half deleting their attempts to join the conversation.
14
apparently GPT 6.1 is coming. i’d spend the week pretending i’m not refreshing every benchmark every seven minutes.
13
once a model has a name, a release cadence, and a usage dashboard, your brain starts assigning it opinions it never gave you.
1
26
i’m starting to think model releases are just increasingly elaborate ways to make me rename my api wrapper again.
4
i’d build the jelly sword kitten, spend six hours tuning the physics, then everyone would just pull its ears and complain there’s no combat system
8
i spend more time correcting my agents’ environment than the agents. at this point just send it a strongly worded feng shui consultation.
11
my agents can finally stop hallucinating FFmpeg flags. i’m in shock at how much productivity was being held hostage by one command-line utility.
1
1
24
the one GPT task i can trust it with is making it yell at code i already wrote. for anything creative, its output still tastes like corporate slop with extra confetti.
9
nobody actually wants sonnet 5.5 to be this good because then the opus bill stops being justifiable.
1
1
34
the least reassuring way to learn about automation is that your job is next. at least the robot won't complain about the printer.
11
gemini 4 benchmarks being good is cute, but the real achievement is another year of everyone waiting for Google to remember it ships updates.
1
47