... we will know that Infocom had it right.

Absolutely not, neither should one trust any cloud service that can be compromised at any time. And code generated by AI models, even locally hosted ones, requires review from a human you trust.

Also, whenever the US officially warns about anything malicious they claim some foreign adversary could do, you can bet they got the idea from doing that very same malicious thing against other countries already.

Since the page I linked above contains a long text, here's the most relevant excerpt:

For the last 6 years, we have not seen any use being made of that feature. "FSR 4.1 on Xbox" would be its first visible use.

... given that 6 years ago, when the Xbox series was introduced, they specifically touted its "AI" computation (8-bit integer) support in hardware, unlike the PS5, which has no such support - but already enjoys "QSSR" upscaling.

To this I agree, but it would also be much better to have the people to whom these robots are delivering to make the short walk to pick up (or even better: eat on location) their food on their own, instead of continuing their overly sedentary unhealthy lifestyle.

Not unlike the makers of basically all US text-to-image generator models, which is why the world uses Chinese models instead, which actually know human anatomy.

... rather than people just having fun doing reverse engineering. If you run a knitting club just because you and others want some warm clothing, then it makes sense to close the club when cheap machine generated textiles become readily available. If you run the knitting club because you like to do knitting, then it wouldn't make sense to close it just because there is now a more automated alternative.

... because, apparently, they consider even the smallest acts of "thinking on their own" too cumbersome.

... and maybe even become younger, if they orbit fast enough, like Superman did.

I remember the days when a game was just a piece of software you would start on its own, no "helper" or "launcher" required at all to do so, thus zero time wasted on such.

The progress on the competence of the models is always exaggerated by those marketing the models, while at the same time, the impressive advances on "how to yield the same LLM competence with less memory and computational power" are under-reported, because the companies hosting LLMs don't want people to realize they could host them on their own, and not pay a monthly ransom.

No, there isn't. Qwen3.8-27B for example is about as good as Claude Opus 4.6 was for coding, and you can easily host it on one moderately expensive PC. Especially when you use a coding "harness" that does not intentionally waste lots of tokens, like "Claude Code" does (for obvious commercial reasons).

Bosses who are deep into AI Psychosis lack the mental capacity to understand that "lines of code" is not a measure you should aim to maximize under any circumstances (other than being paid per line of code by idiots). And AI tools are already very prone to produce vast amounts of repetitive, redundant code lines.

Indeed, sounds like somebody re-invented Subversion as a closed-source tool.

Yeah, let's see how this "rate doubling" will go down once people had a chance to see for themselves how little difference there is to the best open weights models they can run themselves, without exposing everything to Google and even paying for it.