Subject: complaint about cargo-culting audiophools and the #marketers that exploit them

I saw a new #scam yesterday. Well, new to me. It's probably been around forever.

There are many, many #fraudulent products and ideas out there that target the self-described "audiophiles". And because such #audiophools never do double-blind tests, they always convince themselves of just how well the #scammy products work ("you are the easiest person for you to fool").

Examples of long standing:

Super-thick "oxygen-free copper" #cables for #speakers or #interconnects, being sold at 100x the price of regular cables.

Electrically shielded fiber-optic cables (!).

Paint pens to colour in the outer edge of CDs and DVDs.

Systems to run your equipment from #batteries so as not to contaminate them with that dirty mains #AC.

#Tubes / #valves instead of solid-state #amplification.

... and on and on.

Well, yesterday I was looking for some opamps (tiny little chips, used everywhere), and came across a place selling replacement #opamps for use inside stereo equipment. You can get perfectly good designed-for-audio opamps of good quality for a few bucks each. Upgrading your equipment to better opamps sometimes made sense in the 1970s, but not now. The ones used from the factory are fine.

But no, this one was selling fancy opamps, designed for *RF* use, not audio, which normally cost $10-20 each, in fancy little metal packages, for hundreds of dollars apiece.

"A fool and his money ..."

Another great #Interconnects post, this time on #LLMEvaluation in #AI, and just how hard it is.

https://substack.com/app-link/post?publication_id=48206&post_id=150192338

I found this issue of the #Interconnects newsletter really interesting. Progress in small, #opensource_llm models, but also insights into the dynamics of the whole #foundation_model ecosystem, and most importantly for me, a load on #llm_benchmarking. I'm still keen to find a way to relate the benchmarks to real world tasks in a way which makes sense for people who are trying to build, or are considering building #llm applications.

https://www.interconnects.ai/p/olmoe-and-building-better-llms?publication_id=48206&post_id=148458085

`Moving the incredible amount of #data generated by 528 threads around the die requires an optimized interconnect, so #Intel designed a 2D on-die mesh with 16 routers to shuffle data between the cores, memory controllers, and #silicon photonics #interconnects (eight routers are integrated into the #CPU cores, while six routers are dedicated entirely to just data movement). `

https://www.tomshardware.com/news/intel-demoes-8-core-528-thread-puma-chip-with-1-tbs-silicon-photonics