I think parent comment is referring to the other companies before Cloudflare who settled out of court with Sable, not Cloudflare who just destroyed them in court.
This is very common theme in a lot of chinese labs/research labs - they promise code until the paper is accepted and never really update the repo/share the code.
IIRC morris chang indicated US operated fabs would cost ~50% higher, which is not cost competitive, well not something buyers would sign for without something happening behind the scenes (i.e. US gov pushes Nvidia and Apple to use Intel foundries). If A16 is $100 from TSMC TW, it's $150 from TMSC US, presumably $50 to BOM is something Apple can afford, but most others might not. List of companies who are willing to source at 50% limited (unless incentives).
Some of that has to be offset by shipping, importing fees, etc. But I do think the end goal is to get Apple/Nvidia/etc back on US soil manufacturing wise.
Don't these chips still need to be sent to China for assembly by Foxconn? If anything this will increase costs even more and seems like import fees could potentially be even higher due to the current US-China trade war. Unless there is a plan to assemble everything in the US/Mexico as well. But then the costs would be way higher to assemble outside of China: certainly in US but probably also in Mexico, and I don't think they have any factories there.
$100/$150 would be the "shelf price" of the Apple chips if they were in a box for sale like an Intel/AMD one
I believe you that the cost of the delivered (roll of) plastic chips is 50% bigger in the US. Probably less but it might be (also need to include the logistic cost to send it back to assembly on iPhone, etc)
Apple (pre-)pays for stuff and probably doesn't have any orders where Qty is under 7 digits with these big vendors.
> With the M3, engineers at Michigan are the first to accomplish energy neutrality via indoor energy harvesting in a wireless system of its size. With a 1mm2 solar cell producing 20nW, the device can harvest enough energy under ambient light to run perpetually.
Are there any benchmarks which compare existing LLMs using langchain-style multi-step reasoning?
The new OpenAI model shows a big improvement on some benchmarks over GPT4 one-shot chain-of-thought, but what about vs systems doing something more similar to what presumably this is?
reply