For a long time the conventional wisdom was: "don't write asm by hand, you will never beat the compiler" Now we have AGI beating the compiler When optimizing for perf it can just read the compiler generated asm and map it back to the source instantly, 1000x faster than a human could And I notice it finds lots of stupid shit the compiler does like spilling registers to the stack, not keeping hot values in registers etc Instead of struggling to massage and coerce the compiler to generate the optimal output, sometimes it's just better to just straight up get the model to generate the asm And any perf optimizations it misses it can just steal back from the original compiler output And obv it can generate it for every CPU architecture and know the intimate details of each one and optimize Probably not practical yet for all software due to context window size and speed, but super useful for generating optimal code for some small to mid sized functions

So much of life is luck: meeting the right person at the right time, finding the right opportunity for your talents. Fortunately, you can develop an unfair advantage if you learn to increase your surface area for luck. More below.

If you understand your whole codebase, the codebase isn't that important. Real software is larger than what your brain can comprehend.

I love @Blick_Editor holy shit this is awesome

GPT-6 Astra is our best model yet for computer use, software engineering, and visual understanding to date. Hear from the builders putting it to work:

forced deprecation of mv2 extensions is only part of the problem. the chrome web store is generally hostile to everything that doesn't benefit google, and i don't think it should be the only place for all chromium browsers to get extensions. google likes to block genuine extensions from releasing updates (or from being published at all), yet it rarely takes down active malware in a timely manner, if at all. it's surprising an alternative web store hasn't been created yet, but not unexpected, considering browser vendors rarely do anything for everyone's benefit without a hidden incentive. see "microsoft edge add-ons" or "opera extensions", for example. they're locked to a vendor's "ecosystem" for no real reason. the biggest blocker here is safety and moderation, because every extension would have to be thoroughly scanned for malicious behavior, and i don't think existing software is good enough for that. it would likely be the biggest expense for this project. it'd be great if we had someone dedicated to this on the helium team, but since we're *really* tight on budget, we'd have to work even harder and stretch ourselves even thinner for the benefit of an open web. i'm not sure if we'll be able to do this, so don't take this as a promise or announcement or anything serious, just thinking out loud lol