tough

There's a meta-recursion there that's kinda funny innit

New Amsterdam TV Show is a great insight into this POV imho

mcnamara / goodfire couple weeks ago

maybe both tried to read/write it and saw each other on the bench

given than LLM training doesn't "need" yet to adhere to much licensing, i'd say any source available, which with "reverse engineering" as skill anything is, makes the field ripe for disruption by llm's

claude code "auto-mode" uses a smaller sonnet-like model for classifying, which means it's basically dumb, and the harness biases it into denials

> nowadays selling your tesla involves the new owner identifying themselves and uploading their identity documents to tesla.

wait does this mean you cannot buy a used tesla anonymously (to tesla), not the reseller (?)

If so, that's crazy

amazon famously let you swap products of an id/ad/product without loosing the reviews so you can even find literal unrelated reviews under such fake product amazon listing pages

One would think most jobs are "by order" and thus the customer comes before the risk

Now, look who folded.

It's also a great "telephone" between the low-ranks and the C-suite if the comms are broken on a company.

meaning: If your boss won't bother to ask it's own employees or trust their feedback, he might choose a well kept, well presented third party to gather that feedback from their employees, and then package it nicely with some cool logos and slides and decks, and charge you a shit ton of money out of it.

We use GLM at work and YMMV a lot depending on the provider. So i guess some have bugs on their inference pipeline, or serve too agressive quants

Yes they're basically glorified "insurance" for the c suite