Re: Completely unnecessary
Haven't seen much evidence of the former lately
3546 publicly visible posts • joined 19 Jun 2009
The best computer scientists in the world have built a neat toy, but they've already used all of the training data available to them, both legally and illegally.
When OpenAI runs out of money and still doesn't have a product that real people want to pay actual money for, what happens?
Some go to even more extreme measures...
I recall reading a car magazine published in the mid-late 70s, in which they had a group test of medium sized four door sedans from the era, including a handful of Japanese models and some British ones. The Datsun 180B/Bluebird could have been one of them. It struck me that the info box for every single vehicle in the test included a list of the defects the test example had off the showroom floor, which definitely isn't a thing these days!
A 1970s Datsun was a product of its time. However, the Japanese automakers really sorted their shit out in the 1960s and by the late 70s were generally class leading in terms of value and reliability.
Tesla has been speedrunning the mistakes that the traditional automakers (you know, the ones that are only "worth" a fraction as much as Tesla) made more than 50 years ago.
From my perspective, the demonstration flights of crew dragon were far more interesting, and media coverage of those would also have been more in line with the public interest cycle of the Apollo and Shuttle programs. This wasn't exactly an Apollo 13 moment...
I am a bit cynical but I don't think I'm too cynical?
It was interesting to see many of the new channels here in Australia covering the return live. I'm fairly sure that hasn't happened before. There was nothing special or out of the ordinary about the mission, other than the fact that a couple of its passengers, who are otherwise veteran professional astronauts, had spent longer than they expected in orbit.
Presumably, most regular media outlets simply didn't understand how routine it was and got sucked into the "rescue" circus by this administration's PR department.
That's the problem with all "smart" TVs - a few years down the track support ends and then the apps either stop working or simply get deleted. It's a scam that's been going on for more than ten years now and I don't understand why consumers still give it any attention. Or are there a significant number of people who buy a new TV every three years?
Given their past shenanigans, I wouldn't trust a Samsung TV to actually have obeyed an instruction to disable Wi-Fi, as opposed to silently searching for a nearby open access point and spying on me.
No, you don't need datacentre products, you need a Quadro. The current high end is effectively an RTX 4090 with 48 GB of VRAM and none of the CUs disabled. Its replacement is effectively an RTX 5090 with 96 GB of VRAM and none of the CUs disabled, and is apparently days from an official announcement.
Tom's Hardware maintains comparison charts which can be useful, although they exclude the absolute low end parts. For example the 1440p ultra non-RT chart shows that almost everything modern outperforms an RX 580. Either an AMD RX 7600, Intel Arc B570 or Nvidia Geforce 4060 will outperform what you have by a factor of about 2.
If you're doing generative AI work, the VRAM becomes important and, per PCPartPicker, the least expensive cards with at least 16GB appear to be the Intel A770 and the AMD 7600 XT. I have no idea whether they're very good at it, though, and I'd simply avoid the Intel A series.
I asked a similar question about the "neural processing units" being added to CPUs these days. The answer was to do with precision. LLMs don't need high mathematical precision, so they perform really well if you can trade off precision for more OPS. Traditional CPUs don't do that very well, but GPUs do.
Generative AI models are just complicated mathematical predictive models. Calling unwanted outputs "hallucinations" is an attempt to deflect (from the fact that "hallucinations" constitute 100% of their output) and to anthropomorphise them (so that people believe they're capable of "understanding" the idea of "truth" in the first place).
And we probably learned from the best...
The sheer amount of information that people have given up today is something that Hoover could only have dreamed of. Now that the tech bros are kowtowing to or even working with the executive (and, presumably, also the legislature, now apparently merely a vassal of the executive), the possibilities for abuse are endless.
though this time, it's not proprietary company information but state secrets of the US of AAnd not just the state's secrets. By way of example, the previous (conservative) Australian government developed a habit of leaking the personal information of private citizens as a way to threaten and suppress them. We'll no doubt see that kind of thing on a whole new level from this American regime.