Agree. I took a look at these last few months, did a write-up: https://languageops.com/blog/ai-safety-pdoom-local-vs-fronti... and I don't know if I agree or not on outlawing completely, but I think an age restriction *at least* like for alcohol, firearms and driving would be not unwise.
It is not feasible. They never made much of an inroad against torrents and that is a much easier target than abliterated models. As the linked website shows; the process to abliterate a model can be as simple as
If you think closed source software/binaries only is bad, wait until you see how awful the state of the art is with a clear-as-mud bucket of matrix weights.
We know it's possible to train an LLM to secretly respond to certain trigger phrases, and last I checked these could only be detected with the assistance of whoever chose those phrases.
The trigger condition for such backdoors is not something anyone can do a systematic brute-force check for, for the same reason we had to invent LLMs in order to do natural language processing: combinatorial explosion.
Passing around open weight models from known sources is already asking you to trust those sources; because of how difficult this is to do correctly even without deliberately inserting such things, we still don't know if China has already put such trigger conditions into their models despite headlines such as these: https://venturebeat.com/security/deepseek-injects-50-more-se...
Regardless of if it was deliberate or not, we don't know if we caught all of these misbehaviours. We don't know how to.
And note, I'm not saying "and therefore you should trust the Big Name Models". If open weight models score 2/100 in this context, closed ones score 1/100.
600b-a27b doesn’t sound enticing. Also with the higher number of active parameters compared to GLM 5.3 flash and DeepSeek V4/4.1 flash, I don’t see how they want to be more efficient at inference.
They see all the traffic in cleartext. Plus you have to trust them not to maliciously alter your traffic. As a US company, their options may be limited if they are coerced by their government to do so.
Suggestion: Every time someone uses nitter, the post in question is copied over to Mastodon and subsequent accesses just redirect to the copied post. That way you can interact with the copy on Mastodon.
I've been using the Safari MCP in a "grocery agent" tool I built. It takes the weekly menu my wife puts together, turns it into a shopping list, and then uses Safari to add things to the shopping cart on Walmart's website. It's pretty solid, I like it more than the Chrome dev protocol I had been using before Safari.
I've actually never had it trip any bot detection, captchas, or anything like that, and haven't built anything into it for evading detection. If I had to guess, it's because I make sure to manually sign in to my Walmart account before it begins adding things to the cart.
Not on macOS but have my own browser extension basically doing the same with Firefox, for use with agents.
It's using a clone of my real Firefox profile, and can "ping" me if it encounters a page that needs my human input, like a login page (as I don't give it access to my password manager) or captcha page. Although sometimes it "bypasses" the captcha page by itself, even though I've instructed it not to...
Unfortunately, not well suited for stealth use cases for personal automation. Here's a comparison to an mcp tool that allows driving your live safari via osascript:
Given that Safari needs the most debugging and automation tooling for Safari was hardest to come by, this is a welcome change. I wonder if it works with pages on a connected iPhone though? Anyone tried it? Because debugging on iOS Safari is the most maddening. (No, responsive design mode doesn't cut it, iOS Safari has its own special bugs that don't manifest in desktop Safari merely emulating a phone viewport.)
reply