("It actually makes RAM needs worse" -> "MoEs actually make RAM needs worse")

It actually makes RAM needs worse than a dense model, for a given quality.

That said, inference techniques like FreeToken are improving MoE swapping performance, so it's not as bad as it once was. Honestly, I would not be at all surprised if we start doing training in a swap-aware manner, where expert cache misses count as loss and models can queue experts to start loading before they're needed. Could even train in two tiers of loading - RAM and disk. Wouldn't that be great? Could run multi-terabyte models on consumer hardware. Training would tend to tend to concentrate key logic and reasoning on small number of very active experts, mixed reasoning/knowledge on less common experts that are usually kept in RAM, and rarer knowledge on experts that usually remain on-disk until needed.

There are well known, reputable abliterators on HuggingFace.

Also, censorship usually (not always) is something you care about for chats, not agentic work.

A .safetensors or .gguf model cannot run code unless you let it. No agentic framework = all it can do is write text. It can't hurt you directly. At worst, it can only lie to you.

It's big, but not good for its size. They're using tricks to pretend that they're better than they are. For example, compare the numbers that they list for the competion on DeepSWE up against the actual DeepSWE scores.

Not cool, Mistral.

Artificial Analysis is overrated, but yes, Mistral is playing fast and loose with their claims.

With permanent DST we won't get the stories twice a year.

It takes one hell of a toll to try to do verbal exams 1:1 on students to evaluate them.

Nobody said "verbal exams 1:1".

The only thing that is required is that they be separated from their computers and their phones.

I guarantee you that they gave themselves the right to review your input for any purpose and inform law enforcement if they feel the need. I haven't read the agreement, but I'd still bet money.

You're very good at ignoring things, like the rest of the story.

Initiative, Clearinghouse... What? Most of us don't even want to know whether those are programs or products, and certainly don't know now.

IBM is pretty bad at naming in general, their names really tell weird stories. And I say that as someone who used to work for them doing support for TME10, another beautifully named product.

On VMware fusion? So on top of an inferior OS and also inside of virtualization?

Your answer to Apple using twelve or more gigabytes of users' disk space is they should print and delete some documents? Do you work for Adobe, or HP?

The difference is that if you pay people to monkey around with things, you get a non-revocable perpetual license to the result. With license rental, you get a perpetual bill. It's the same rent trap as individuals get stuck in but with less excuses.

Yes, and the reason is that they stupidly painted themselves into a corner and never thought about an exit strategy. Had they been thinking about the possibility that they might one day need to migrate from the beginning, it would be much cheaper to do so now.

That's the thing about painting yourself into a corner, a little better planning at the start can always avoid it with little to no additional effort.

This is only true when the data is insufficiently characterized to begin with. It should be obvious which parts are confidential. You flag records as such ahead of time specifically for purposes like this.