By this point I must have discovered most of the bad bits of JavaScript.

@paulg Imagine if CEOs changed back-and-forth every four years the poor designers and software developers! Phoenix, firebird firefox…

@ChrisEmmin56954 @DanNeidle It allows you to reclaim VAT, which might be useful if you are committing VAT carousel fraud. This is where one company charges VAT on some goods but disappears before paying it to HMRC, then another company receives the good and claims the VAT refund from HMRC. Repeatedly.

@_mehiel @businessbarista @hubermanlab Yeah, he’s he’s actually totally wasting 20% of his time - Should just get a fifth company!

@snyule @RoryMaw @janrosenow @ember_energy Yes. The levies on electricity and gas strongly favour gas but there are plans to equalise the levies which will drastically improve things for heat pumps. To be cheaper now, you’re heavily dependent on getting a well designed system (300% is only average).

@RoryMaw @janrosenow @ember_energy A gas boiler might be 95% efficient at transferring energy to heat. A heat pump is around 300% efficient due to the fact it is moving rather than generating heat. So it’s better to burn gas to generate the electricity for heat pumps even ignoring wind power.

@brendon_rogers @daniel_nguyenx Yup

@maxawad_ @Authy Oh, that’s annoying isn’t it

@TwilioHelp @maxawad_ @Authy @twilio @SendGrid It’s not that we’re desperate for a desktop version, but you now prevent us from using the iOS version on our Mac even when our Mac is capable of running iOS version natively.

@brendon_rogers @daniel_nguyenx Well, it did until they put up a big page saying you need to upgrade, and they disabled being able to run the iOS app on Mac

@TraceyRyniec @SecondCityCop Also, deepseek used older cheaper nvidia chips. As I’m sure Tim Cook would testify it’s quite important for your share price that everyone is desperate for the latest greatest ones!

@TraceyRyniec The deepseek api is 20x cheaper for inference, I would assume therefore that they have massively reduced inference costs and it is possible further improvements could keep on dropping the need for compute. Not unreasonable drop given the huge expectations now on NVIDIA imo.

@ericjang11 No, because you now hit diminishing returns on model quality improvement with far less compute. R1 also needs far less chips for inference so the need for chips once the model is produced also seems much lower.

@NoContextBrits Because your hand isn’t big enough?

@TraceyRyniec it is quite funny when open AI declares they are losing money on their subscriptions - shortly followed by a company no one has heard of producing similar output and selling it in the market 20 times cheaper than openAI.

@sytelus Stock price had already factored in massive growth for inference. Deepseek api is what, 95% cheaper for inference? That’s a significantly lower growth in CPU capacity needed for inference than a couple of weeks back!

@amyvpellegrini You clearly haven’t been wiggling the mouse enough!

@amyvpellegrini I’m going to hazard a wild guess that people have differing views on what a better world looks like!

@jamesrbuk So I think you have to weigh up the downsides of what she’s saying versus how it’s going to play out if she’s forced into an emergency budget (which could totally happen)- in which she might have to make austerity style cuts.

@NoContextBrits