Forgot your password?
typodupeerror

Comment TIOBE is measuring a world that no longer exists (Score 4, Insightful) 17

TIOBE is based on counts of web search hits using the query "[language name] programming".

This methodology (that was questionable even when humans turned to web content created by other humans to seek help about programming languages) is now absolutely useless when humans use AI coding agents for programming.

The interactions that used to appear publicly in sites like Stack Overflow (new monthly questions posted down 98% from its peak) now belong in private chats between developers and their AIs, out of reach from web search indexers and thus invisible to TIOBE.

What TIOBE sees is mostly the fossilized landscape of two years ago. The rest is noise.

Comment Re:A good step, but... (Score 1) 32

Sure, go far back enough (and yes, I was on computers at that time) but surely he's not saying, "Man, computers were better when you could only run one process at a time, and the operating system let you do literally anything" etc etc and if so, the answer to that is, you can still do that if you want .. you just can't do it *inside* somebody else's software architecture. You still own the hardware, do whatever you want.

Comment Re:A good step, but... (Score 1) 32

huh? it's trivial to write any code you want and run it on your phone

But with respect to your "dawn of the PC era", unless you wrote the operating system, you couldn't write and run code that could do anything and everything your computer was capable of - only code that could do what the operating system would permit you to do.

That's no different today.

Submission + - Leak-Proof AI Benchmark Posts Biggest Gain in 20 Years, on One CPU Core (spasim.org)

Baldrson writes: The Hutter Prize for lossless compression of human knowledge has paid out nearly €50,000: its largest advance in 20 years: three entries together cut the record for compressing a 1GB Wikipedia snapshot by nearly 10%: a compression ratio of more than a factor of 10. Every entry ran on a single CPU core, in about 50 hours and under 10GB of RAM, so the gain came from algorithms.

Unlike most AI benchmarks, the prize requires no validation data, so a leak can't inflate a score. The full file is public, and the size of the decompressor counts against the entry, so anything memorized is paid for in bits. Vladimir Ivanov's fx2-cmix-T now holds the record, after earlier 2026 gains by Ibrahim Marcouch and Kaido Orav (cmix-lex) and David Freelan (cmix-obias). Sponsor Marcus Hutter, who advised DeepMind co-founder Shane Legg's 2008 thesis "Machine Super Intelligence," calls it "the largest progress in its 20-year history." With trillions riding on claims of AI progress and data-center power demand, is a leak-proof, energy-limited benchmark the yardstick the industry should be watching?

(Disclosure: the submitter is on the Hutter Prize committee.)

Slashdot Top Deals

Algebraic symbols are used when you do not know what you are talking about. -- Philippe Schnoebelen

Working...