Good (negative) review of Perplexity.

So constant digital interruptions, colliding double-booked appointments, no awareness of last-minute deadlines, phoned in lectures, and using tech to isolate rather than to connect. All in a world driven to the brink by fires and drought. I’ll say one thing for 1987 Apple: they did make a cyberpunk video that’s more of a warning than a utopia.

Standard ML is a general purpose functional programming language. It is statically typed, which prevents a wide array of common programming errors, but also features powerful type inference, requiring few type declarations, if any. This makes it easy to build software that is easy to understand, extensible, and reliable.

Use cases, techniques, alignment, finetuning, and critiques against LLM-evaluators.

What a perfect analogy for how painters felt when the camera was invented.

On the other side of the multiverse, over in the timeline where sharing is caring and where the downhill battle against the copyright trolls was won and where there’s an efficient peer-to-peer system for songs and videos, every now and then they take a look over at our dimension and they glance a li’l incredulously at what the heck we’re doing with our huge fossil-burning wasteful data centers to manage streaming and distribution and then other huge wasteful data centers make sure to keep it that way, to tear down anyone who’d dare to help distribute it.

Always ragging on bad anti-ML arguments but here’s a great one.

[I]f it’s mindlessly generated and thrust upon someone who didn’t ask for it, slop is the perfect term for it.

Happy about this blog post.

While there are many really good arguments against ML, I’m not stoked that opponents to it have been weaponizing copyright. [Copyright is the enemy][Cite], and while I can understand the perspectives of “any port in a storm” or “desperate times call for desperate measures” or “my enemy’s enemy”, I still get to feel a li’l grumpy about it.

This came from that list.

The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin.