mil22

Facebook is actually a good counterexample to the claim that language choice doesn't matter. PHP's execution overhead was significant enough that Facebook developed HipHop, a PHP-to-C++ compiler, which reduced their average CPU consumption by around 50%. They subsequently developed HHVM and Hack to address further performance and maintainability issues.

That doesn't mean PHP prevented Facebook from scaling, just as C# didn't inherently prevent Orkut from scaling. But being able to scale an application and being able to scale it efficiently are different things. You can compensate for language and runtime inefficiencies with more hardware and better architecture, but that doesn't make those inefficiencies irrelevant.

C# would arguably have given Facebook a better starting point in terms of execution performance and static typing than PHP, but claiming they could have run the original application unchanged today is a bit of a stretch. Facebook's architecture, infrastructure and requirements have evolved enormously, regardless of language.

Not entirely orthogonal. Orkut's original C#/ASP.NET/SQL Server stack was an awkward fit for Google's predominantly Linux-based infrastructure, which made scaling and operating it more difficult. Almost all of Google's other web services were written in tight, optimized C++. David Jeske, Orkut's former technical architect, described problems with the .NET runtime, difficulties integrating with Google's infrastructure, and an eventual migration to Java and C++.

That doesn't mean C#'s execution speed was the primary bottleneck. In fact, Jeske's team reportedly made dramatic performance improvements while retaining .NET. But arguing that the choice of technology had no relationship to Orkut's scalability problems is just as misleading as blaming everything on C#.

The original Orkut site was written in C# - it was very slow and couldn't handle the growth. Orkut was only a big success in countries like Brazil and India where internet connections were already so poor that nobody noticed how slow the site was.

Deactivating my LinkedIn account was one of the best things I did for my mental health. I have not missed it even once, and my career has never been better.

Yet another scramble to use up weekly quota before an Anthropic cliff. This isn't fun.

Israeli hegemonic ambition. Zionist funding of Trump's campaign. Corrupt profiteering from market manipulation. Distraction from the Epstein files. Gullibility of a large fraction of the US public. Trump's ego and stupidity. And when a nuclear superpower starts a war like this, there's not much anyone else can do to stop it. The result: a massive strategic blunder for the US, the true cost of which may not be fully understood for many years.

From the original paper abstract:

“Claude review raises Codex drafts from 71.6% to 89.7%; Codex self review raises them to 84.5%. The reverse direction does not pay off: Codex reviewing Claude drafts drops the pass rate from 91.4% to 82.8%, and Claude self review leaves the 91.4% baseline unchanged.”

The highest pass rate was to use just Claude, at 91.4%. Claude reviewing Claude did not improve the pass rate.

How well do models like these generalize to robot embodiments they weren't trained on?

Now if they could just extend the 50% weekly limit increase promotion which will otherwise expire on July 19th at 11:59PM Pacific, they might actually remain competitive.

Finally! Definitive, tangible, tactile proof that we're near the top of the bubble. /s

Hopefully the same will happen this time but there’s no announcement and it’s uncertain and unpredictable, which is less than ideal. Personally I am using all my quota before Monday morning Pacific time anyway, hoping they will do a reset when the team is back in the office. If I’ve bet incorrectly, I will have wasted my 25% by using it all in one day, in a rush and therefore inefficiently, rather than being able to take my time and plan it out. That’s what irks me the most.

In fact I’d say the entire system of weekly quotas and resets seems quite poorly designed. A constant rate continuous recharging model would work much better, like a battery. No arbitrary usage period phasing, no waiting for 5:45am on Sunday morning for your quota to reset. There would be no worrying about how your specific reset time aligns with the additional random resets or model availability. That would be a much better user experience and would cost them nothing.