Ive always wondered where the inflection point lies between on the one hand trying to train the model on all kinds of data such as Wikipedia/encyclopedia, versus in the system prompt pointing to your local versions of those data sources, perhaps even through a search like api/tool.
Is there already some research or experimentation done into this area?
The training gives you a very lossy version of the original data (the smaller the model, the lossier it is; very small models will ultimately output gibberish and word salad that only loosely makes some sort of sense) but it's the right format for generalization. So you actually want both, they're highly complementary.
Here you are comparing a decent bluetooth speaker to a pretty good wireless active speaker to a hifi setup. I think the original comment about audiophiles is them wasting money on upgrading the hifi setup with all kinds of audio cabling, bi-wiring, etc.
That would be similar to upgrading to that one tiny bit sharper lens which otherwise has the same aperture etc.
Yes that's more accurate. And it's about measurability. Even with that tiny bit sharper lens, you can probably point to an actual measurable difference in the photos. Whether that makes them "better" remains subjective.
Audio is a weird world where everyone lives in their own experience and the externally measurable things often don't really translate to the visceral experience. So everyone kinda comes up with their own tribal knowledge that's often more superstition than science, and a lot of people just tend to assume they need "the best" in lossless files and analog whatever and gold-plated this and that.
Because newer batteries are not degrading as fast due to better thermal and load management. Because newer cars use newer chemistries that are less prone to degradation.
Moreover, just like some cars are good enough for people now, the cars with some degraded batteries will be good enough for some second hand buyers.
On AMD the L3 cache is partitioned between the 2 chiplets.
So for 9950X3D half of the cores use a small L3 cache.
For applications that use all 16 cores, the cases where X3D2 provides a great benefit will be much more frequent than for a hypothetical CPU where the same cache increase would have been applied to a unified L3 cache.
The threads that happen to be scheduled on the 2nd chiplet will have a 3 times bigger L3 cache, which can enhance their performance a lot and many applications may have synchronization points where they wait for the slowest thread to finish a task, so the speed of the slowest thread may have a lot of influence on the performance.
Agree. The article's 2nd para notes "AMD relies on its driver software to make sure that software that benefits from the extra cache is run on the V-Cache-enabled CPU cores, which usually works well but is occasionally error-prone." - in regard to the older, mixed-cache-size chips.
> I'm curious to see...
Yeah - though I don't expect current-day Ars Technica will bother digging that deep. It could take some very specialized benchmarks to show such large gains.
Some of their writers, who are quite excellent, still do. Others just seem to regurgitate press releases with very little useful investigation.
How critical of the lazy writers I am may seem outsized, but I grew up reading and learning from the much better version of Ars -one I used to subscribe to.
Here in the Netherlands ford sales seem to have completely consumed by Kia sales. Around me houses that typically had Fords now have Kia’s, Toyota, Tesla or small Volvo like EX30/40.
After the huge hits of the focus and to some extend Mondeo, the Kuga has sold subpar. There were only a few new ones around here. Now you see some new EV Ford Explorer SUV and just a tiny account of the big old Explorer. (Yes, the traditional Explorer suv counts as big here.)
In the mean time there is an explosion of BYD, Volvo, Skoda Enyaq, etc happening. Mostly driven by which model has the most beneficial tax package for lease.
I own a Plugin one, I completely understand why. It's "meh", plus all the recalls because Ford cheaped out on the battery production and Samsung (the battery cells) can't do inventory management. For the US audience: it's the Escape (they are identical in all but numbering).
IIRC from some discord threads, some games already perform better on Linux than on Windows. We are getting there. The only moat left is kernel anti cheat for games like Battlefield. I’m just fine if those stay on windows actually.
I also like the Optica! It somehow has a lot of space vibes from Freelancer and FireFly. Shame of the large toy like duct indeed. But I suspect it works!
I would love to have a Java compiler with the capabilities of the .net compiler. To make incremental builds to aid code completion including type information, looking past simple syntactical errors, fixing them, and continuing compilation.
Currently, this is “magic” embedded in eclipse, IntelliJ, and maybe a bit in the vscode plugin. Imagine having a Java LSP running that can provide all this information while typing.
.net has had this for ages. From a language design I think that is wonderful.
It also slows VS code a lot, is not properly documented, and still relies on concatenating strings together, due to Source Generators interaction with attributes.
In Italy there are enough chargers that charge for both kWh and time connected. kWh for what you use and connected to discourage being connected all the time.
Is there already some research or experimentation done into this area?
reply