lordmauve

We need to see DeepSWE scores. SWE Bench Pro is junk.

I don't know if DeepSWE is genuinely a good benchmark. It's more important that their analysis demolished the validity of SWE-Bench Pro, objectively: it is being mismarked.

I think that buys enough credibility to propose an alternative.

I think there's a case to answer if Anthropic models underperform on a novel benchmark. I'd like to see more novel benchmarks to get a clearer picture.

Given DeepSWE just blew apart the SWE-Bench Pro benchmark and handed a 14-point lead to GPT-5.5, it looks pretty bad that they've listed SWE-Bench first in the model release and no DeepSWE. Like, this isn't obviously an answer.

Or maybe it is, but publish the DeepSWE numbers so we can see for ourselves.

From a lot of game jam games with custom engines, I have a pretty standard approach: I set the game clock to zero timestep, and push a pause gamestate onto the gamestate stack, where top gamestate gets all input. The trick is that you need many clocks, clocks for the gameplay/game animations, HUD animations, menu animations.

Are you using Stop hooks to keep Claude running on a task until it completes, or is it doing that by itself?

It's written by AI. I have made LLMs adopt the same tone by prompting to be engaging. Short sentences. Every point intended to land with impact. Artificial gravitas. I consider that a failed experiment and rewrote it, rather than posting to Medium

Finally, the "AI" turns out to be 700 Indians. We now have the full loop of humans asking machines asking humans pretending to be machines. Civilisation collapses

I'd argue this is the wrong design goal: correctness is more important.

I'm in the UK but my work PC's Internet exit node is in New York due to enforced use of corporate proxies, so the time shown to me is 5 hours out. Javascript would report the correct timezone.

It is not possible to correctly identify physical location from IP addresses. Not just because of proxies and VPNs and the accuracy of the data: you can go near a border and find your mobile phone connects to a cell tower in a neighbouring country, without even visiting! IP Geolocation is accurate enough for statistics and marketing but probably shouldn't be used for anything user-facing.

I don't think that is incompatible at all. It's a restatement of the same thing:

If someone is right a lot they are likely not making assertions about things that they know nothing about.

If they are making assertions about a domain and those assertions are correct then that domain is one of their competencies.

"Utterly useless" is a bit extreme but it's a reasonable observation to say it doesn't have predictive power.

> > Nobody in their right mind would want to learn programming in BCPL

> I agree, but that's not what the author thinks:

The author of this document is Martin Richards, the creator of BCPL. Of course he thinks you would want to learn it.

What we call cornflour in British English, and cornstarch in US English, is flour heavily processed to remove everything except starch, giving a flavourless thickener.

If you use wheat flour you will change the flavour and also add a slight graininess as the flour grains don't completely homogenise.

Also pipx isn't reproducible - it re-resolves dependencies so you may end up with different versions over time or in different places, eventually causing something to break.

If you have a shiv working it stays working, assuming you have a solution to distribute the required interpreter version.

The finance industry mostly uses floats for currency, up until settlement etc.

"What would I get for this share?" can be answered with a float.

"What did I get for selling this share?" should probably be a fixed point value.

> However the following is fine: > > case $a in > > No field splitting occurs here

This kind of bullshit is how I made a career rewriting people's buggy shell scripts in Python