Yeah and this was RPi first step into RISC-V. It seems to have been quite successful? So perhaps in a couple of generations from now we will see a pure RISC-V chip from them?
I would have loved more nitty gritty details also. But to conclude that they "have no idea what they are doing" seems to be an uncharitable take...
The described system seems like a useful test bed, and I did not see anything that claimed this covers all the tests such a system should have.
These days you could probably build a pretty performant numpy like using shared memory with Arrow format and IPC for control.
Though it would be considerably more complex and not at all easier than FFI...
There is topological naming available about USB ports from the Linux kernel. So one can find what device was connected to which hubs, and which root hub that hub was on, etc.
See lsusb -t, and the device paths exposed in udev. Old blogpost of mine about it:
https://www.jonnor.com/tag/usb/
Yep, the advertising packets can be used for things like this. And frequently are in IoT settings. Typically by using the AD type Manufacturer Specific (0xFF). BLE 5.x makes it possible to have many more bytes of advertisement data, including longer packets and chains of packets. Otherwise the mechanisms stayed the same as in 4.x. Which is good, as BLE 4.x only hardware/software is still around...
RISC-V is doing great in microcontrollers. From 10 cent Cortex-M0 competitors by WCH etc, to dual core WiFi/BLE enabled devices by Espressif etc, competing with Cortex M33. On Embedded Linux class devices the ARM Cortex Axx reign supreme, until one hits X86 industrial PC territory. I think these trends will continue, but RISC-V will likely get stronger in all of these as time goes on.
I think that the best way to address this potential ARC overfitting, would be to create more benchmarks - that are similar in concept, focusing on fluid intelligence, but from another angle than ARC.
Of course it is quite costly and also requires some "marketing" to actually get it established.
I believe Stack Overflow and Quora were screwed because chatbot interfaces became the new entrypoint. _If_ Instagram/TikTok/YouTube manage to stay the place people go habitually, then I think they will benefit, not be harmed. But there are probably many ways to step wrong here. There is a bit of an innovators dilemma for the platforms that currently relies on human produced content. If I were them, I would fund 2-3 new/separate efforts to be able to experiment while not immediately killing the current golden goose.
Why would it devalue the platform? The more time people spend watching, the more attention that can be monetized. More content, or higher quality content, or content produced more cheap does not detract deon that?
The algorithms will pick winners anyway, and can shape things into the familiar power law of popularity, if that is the desired mode. People parasocial needs can be exploited by human creators to differentiate from AI generated content.
Modern social media is like a casino, the house always wins.
Which companies would, if they alone become radically more effective, actually be able to radically change impact on their industry? The vast majority of companies are in value chains, where practical value creation depends also on the suppliers and customers of the company. Often there are multiple suppliers and customers, and sometimes there are multiple levels of relationships on each side... That places constrains/inefficiencies on how quickly and how much one can achieve.
This kind of dataset is useful also with a non-learned algorithm. Testing and benchmarking on realistic data is the best way to validate any solution to problems that are non-trivial and have a very diverse input space.
One could build a dataset out of Github data, by analyzing KiCAD files. They are likely to be "completed" projects, so one would have to rip out all the traces and start from that. And the parts are also placed (in a way which makes routing feasible for a human), which is a large part of routing. So one could have another task setting where the parts also have to be placed, and then routed.
Such a dataset would likely represent simple to medium cases, as open designs are usually on the lower side of the complexity compared to industry. And it would be hard to automatically infer important constraints such as differential pair matching. That would require manual annotation, most likely. But if there indeed are no open datasets, then I think this would be a worthwhile contribution to the field.
How is the stem support these days? In particular I would like to create some songs with vocals (my lyrics), then be able to remove the vocal track and replace it with my own.
I do not think the leveraging is going to end there. I suspect this will be used to justify/secure power generation investments, possibly even nuclear. Likely via one or more of the OpenAI/Altman adjacent power startups.
Doesn't a head chef in a restaurant context delegate a lot to other people for the cooking? And of course they use many tools also. And pre-prepared parts also, often from external suppliers.
We have powerful preprocessing blocks for images: Strong computer vision capabilities predates LLMs by several years. Image classification, segmentation, object detection, etc. All differential and trainable in same way as LLMs, including jointly.
To the best of my knowledge, no team has shown really high scores by adding in a image preprocessing block?
Funnily enough a lot of the challenges in this particular case is related to PyTorch and CUDA being native code (mostly in C++). Of course combined with the fact that pip is not really adequate as a native/C++ code package manager.
Perhaps if C++ had a decent standardized package manager, the Python package system reuse that? ;p
This might be said in jest. But does everything have to be for world domination? Is the guy not allowed to have actual hobby projects? That go just where he fancies, including potentially nowhere at all really...
That is a hilarious bodymod. And considering that is is possible to get rough brain activity indicators in a non-invasive manner, something like it could actually be made.
This happens to me a lot. Almost once per session now, and not even when things are complicated. The model also thinks it has done the changes. So it seems a UI/state bug, not on the model side.
Happily using gevent for our backend (IoT+ML) since 2019. Was very glad when I saw it is still being well supported by recent SQLAlchemy and pscycopg releases.
Uber market cap makes places it in the top100 in the world, whereas Lyft is around 1/25th of Uber in market cap, and not even top1000. I would consider that a success...
That is basically as much winner-takes-it-all one can realistically get in a global market. Cases where the top is just 5x the runner up would still be very winner oriented.
Hope it stays long enough to build up serious electricity generation, storage and distribution. Cause that has a lot of productive uses, and has historically been underdeveloped (in favor of fossile fuels). Though there will likely be a squeeze before we get there...