@emubrigadier People often say 0 Fahrenheit is fucking cold and 100 is fucking hot, and that's not wrong, but I'd argue 20 is also pretty fucking cold and 90 is pretty fucking hot. As someone used to Celsius, Fahrenheit is absolutely not intuitive for the temperatures I care about.
@drbarnard My takeaway from Bryan is to do your own n=1 trials and see what works for you. He pretty much says this, recommending that the main metric to optimise should be resting heart rate before bed.
@Necmttn I also have a custom task that generates coverage reports that are stored under version control. This way I can see the diff of the coverage in the PR. Also have another agent stop hook that checks the coverage reports for gaps. Coverage also isn't everything, but it helps.
@Necmttn I'm doing BDD using cucumber, with e2e tests that are kept as realistic as possible using real dependencies wherever possible (e.g. test containers for the db)
I also have a stop hook that runs an agent focussed on checking that the tests are good which helps a lot
Then there is tactical voting. A significant chunk of the people who voted for the party that wins did so to stop another party winning, not because they expected the party they voted for to be good. So it's actually >80% of people that expect things to go badly.
I think this is the main problem in the UK these days. Politics is so fractured that no party can get much more than 1 in 5 people to vote for them, meaning 4 in 5 people will think things are going poorly no matter who wins or what they do.
@andrestaltz I've been wondering this. I get why not everyone would want to use them, but it feels like nobody does. It feels like people don't realise they exist.
@andrew_lilico 1) Expand permitted development so more can be built without planning permission
2) Customs deal with EU that removes all customs checks and paperwork
3) Not policy, but a period of stability where we keep the same prime minister for more than 2 years would be hugely beneficial
My setup for Android and backend dev:
- Claude code cloud environment
- Integration tests for Claude to check its work
- Continuous delivery including spinning up a version for each PR for manual testing
If I need to run something or edit some code myself, it's easy and quick to just checkout the branch locally. Although I find these days I almost never need to do this.
@BenjaminDEKR Same as now. Datacenters will have models that use 1000 times the compute of models that can run on $5000 dollar hardware, and those models will be much more capable, so most people will use them.
@MichaelAlbertMD And if it doesn't improve outcomes, to ask why. Is it that the test just doesn't provide any actionable info and therefore isn't useful, or is it that doctors need training in how to interpret the results?
@anishmoonka I was doing 30 minutes per day for about 2 years tho. 5 minutes per day isn't enough to learn much. Also, learning to understand real speech is hard, just nowhere near as hard as learning all the vocab imo.
@anishmoonka I got up to about B1 level German primarily through Duolingo. In the long run, learning vocabulary is the hardest part of language learning, and if you do enough Duolingo you will learn a fair bit of vocab.
@Robotbeat Absolutely. Although, I get it if the odds of a false positive are high enough and the only follow up option for a positive result is a potentially hazardous test. If it doesn't matter if the result is positive or negative, both should be ignored, then the test has no value.