squeegeeninja

We can send information at the speed of light. It's quite easy, actually.

If (and that's a bit of an ask to be honest) I understand this correctly, the weights will look very different from what trained weights look like. Very sparse, not uniformly random.

I mean, this is a cool demo, but that's about it. It shows that you can build a model that holds secrets which are genuinely encrypted until the model gets the correct input. But it's also trivially detectable, so the framing seems a bit misleading.

This is such a bad video. Someone else's video, cut over multiple different times, a horrible slurred AI voice, shaky filmed screen with a guy talking over… Each layer adds nothing, but somehow this is the version worthy of sharing?

The phrasing makes it sound like you are using them through a chat interface. Have you tried something like Claude Code? The real value only starts materialising once it has sufficient access to your environment.

It's a function of scarcity. Generation (or "writing" as it used to be known back in the day) is now cheap, so judgement and taste are the new bottleneck and therefore the difference between slop and effort.

I don't know why, but in my head Firefly as a whole occupies roughly the same spot as a boring Farscape episode.

Stargate is just one of those rare shows that (apart from, uh, certain episodes) aged incredibly well. I really miss optimistic television where people are competent at their jobs.

Even the whole "science person solves the problem under pressure after getting an insight" thing suddenly seems a lot more realistic now that LLMs have actually become good.

The real story here is that people are selling modded Pis as higher-memory versions. And this check actually caught them + is circumventable. The restriction is somewhat annoying to a small subset of enthusiasts but seems to catch genuine fraud.

Depending on the model, thingino might help.

https://thingino.com/

Then there's a difficult question to answer: Who should get to use AI and who shouldn't?

The data centers aren't there for no reason, they track demand (and a large helping of hype). It's the same question that arises with cars. A single person driving a car is relatively unproblematic. It's when individual usage adds up that it becomes problematic. And what did decades of car opposition do? A lot less than what happened when a better way to do it (read EVs) came up.

The same argument has been used against most data center operations from the beginning. You can criticise specific bad actors and bad practices (like Musk and his blatantly illegal gas turbines) without morally condemning an entire field of technology through the lens of religious environmentalism. There is nothing about the technology that prevents data centers from running 100% renewably.

This feels more like unrelated political dissatisfaction packaged to hitch a ride on the AI discourse train. The experience of struggling for solutions only seems to disappear if you lack ambition. Keeping the amount of thought you put into problems constant only requires broadening your horizon to new problems.

The $199/$299 prices were never real. Those were on carrier-subsidized plans.

I'm not sure you can engineer your way out of the fundamental problem: You have a lever with both arms converging into a fragile display fold. Sand and dust can't not be brutal on that.

It seems like we are watching the newest form of the script kiddie evolve in real time (no offense intended towards the author, everyone starts out young). It's going to be interesting whether this evolves into its own thing or just ends up being the same old dynamics but with a new set of tools.

"feeling" implies experience. Functional emotions are learned text generation modes, nothing more. Our emotional states influence our writing, so modelling our emotional states is necessary for efficiently predicting/emulating our writing. Functional emotions are the model's inference of a fictitious author's emotional state in that situation.

Firefox has been blocked entirely for at least a year now, in my experience.

Aaron Maté is a notorious war crimes and genocide denier. His takes on chemical weapons attacks by the Syrian regime (or Russia, for that matter) for example barely rise above the level of blatant, easily disprovable lies.

The same applies to Seymour Hersh. No matter what your opinions on the North Stream bombings and the perpetrator behind them are, his version of events is objectively false. Almost every detail mentioned in his article is very easy to disprove using just publicly available data. Only relying on a single anonymous source in order to make dubious claims has been a dominant motif in his work for at least a decade now.

The consensus among manufacturers (and auditors) is that R 155 does cover the security of vehicle backend services with its wording and intent. There are of course still active discussions about what exactly constitutes a backend service, e.g. whether a production planning system that provides data to a direct vehicle backend service should also be considered relevant under R155. But in general, this is something that manufacturers in Europe are aware of and are working towards.