mrshadowgoose

No. This is just the start. It's going to get way worse before it gets better.

Hey dude, I don't get the sense you're being malicious, but you might want to consider the following...

I'm a Home Assistant user because I don't like my devices reaching out to random cloud services more than they need to. I suspect quite a number of fellow HA users feel similarly.

Although your intent was probably good, this feature certainly falls into the "more than they need to" bucket and feels hella icky. And yes, I'm very familiar with HIBP and cryptography in general. This simply isn't an opt-out feature I expect in a privacy positioned home automation product.

Proactive public sentiment management to lessen the chance of reactive policies.

In Canada, it's currently vogue to hate on all things American. (I'm making this observation as a Canadian resident)

Nope. This is exactly the same way they fucked up when they had the model quality lead with Gemini 2.5 pro.

They made using it so hard/unclear that myself and every other dev I knew gave up.

It's clear that they still haven't done any organizational self-reflection.

This was my conclusion when I saw the feature for the first time.

One of the open problems at the frontier, as we climb the levels of abstraction towards longer task horizons, is "what's the next step that makes the most sense?". This feedback seems like a frictionless way to gather those yes/no feedback signals in a natural manner to feed into future training runs.

Similarly, when talking about "human intelligence", I sometimes find it useful to use "bunch of atoms bouncing around".

It's tragic to see, and they already made this very same mistake with Gemini 2.5.

They made their model so hard for people to drop into workflows, that people just... didn't.

Lack of adoption caused them to lose out on usage-based training data needed to smooth out their model's rough edges and progress the frontier.

On the off chance there are Google execs going through this thread:

Google, if you've actually managed to catch up again, please don't fuck this up (again).

You made Gemini 2.5 Pro so difficult to use that myself and everyone else I know (who even bothered to try) just gave up and used something else. If you make this hard to access, you're going to miss out on rich usage-based training data that you need to progress your capability frontier. Again.

"lighter" is the revolutionary bit.

A huge impediment to VR uptake is having to go through the process of strapping a brick to one's face.

This isn't quite the "just putting on a pair of glasses" experience, but it's much closer to that than the former.

I frequently have speculative conversations about what (good-leaning) post-AGI societies might look like.

When asked about how one might attain social mobility in such a society, where the intrinsic value of our intelligence and ability to do labor is gone, I unironically respond with "be hot".

The arms race is an intrinsic game theoretical property of a multi-adversarial-actor scenario involving exponential growth of a universally potent technology. It's almost certainly winner-take-all, on a global scale, which behooves everyone to participate.

And no, I don't think it will end well.

This is not an impulse reaction, as I've thought deeply about this for many years, and I'm quite resolute in holding the following view:

Quibbling over the academic nature/definition of "true intelligence" is a terrifically useless endeavor from the lens of evaluating "practical impact to the world".

Regardless of whether or not one academically disagrees that these systems are intelligent, they are clearly already capable of permanently displacing a portion of human-based economic value. A small portion currently, but it's very clear that most computer-bound domains are imminently at-risk.

And that will profoundly affect the world, regardless of whether or not they're "truly intelligent" as per your personal definition.

Look up "mechanistic interpretability" in the context of LLMs. The next token prediction machinery is just a foundation for a higher order learned structure that appears to encode specific concepts, regardless of input language.

The analogy to humans is that the human brain is "just atoms bouncing around", but there's unquestionably something "more" going on that just that.

In areas of applicability where GenAI to now considered to be good/great, the pattern I've observed is a frighteningly rapid progression from "abjectly terrible", to "middling and mostly more trouble than it's worth" and then "good/great".

Agentic coding and image generation are probably the best examples of this.

There's no guarantee that this trend will be seen with virtual assistants, but if it does, we're on the brink of incredible. Dedicated personal assistants, that actually know and understand one's context and work well together, are unarguably hugely appreciated by those who can afford them.

The conclusion of this blog post is a bit hysterical. The intent of this steg is excruciatingly clear (identifying usage by Chinese firms that may be conducting model distillation). It's unclear on how this "punishes normal developers" in any shape or form.