Anthropic partners with SpaceX for compute
A summary of things I read this week
AI is on the rise, but with LLMs comes the need for compute. This week, Anthropic partnered with SpaceX to use its Colossus 1 data center. It’s quite a big deal, as Anthropic was struggling to find computing resources to meet its increasing demand. Colossus 1 provides them with 220,000 NVIDIA GPUs. They also increased the Claude limits because of this news. It’s gonna be fun to see the competition between OpenAI and Anthropic for compute. It also comes as a surprise that Elon has always called Anthropic Misanthropic, but then he posted about talking to them and making sure they do not flip Elon’s evil detector.
Elon also announced that SpaceX and xAI are merging to become SpaceXAI, and that they have already moved on to their own Grok training at the Colossus 2 data center.
I am definitely excited about this (and more Claude credits). I think we need to build more computing power and more data centers to enable humanity to thrive.
Someone posted another angle on why Elon is leasing Colossus 1 to Anthropic. It’s a pretty good analysis.
Amazon has launched its supply chain services to everyone. Amazon is known for its logistics and supply chain, which reliably delivers packages around the world. Given their expertise in this, it makes total sense for them to launch something like this. This is going to change the world of supply chain and logistics, as I expect many small and medium businesses to use them without having to worry about handling all this.
Amazon disrupted internet infrastructure by building AWS, and now it’s disrupting the supply chain business. They already have “Fulfilled by Amazon” for 3rd-party sellers on Amazon, but this is a step forward that lets anyone use their services.
As I am reading a lot more about energy these days, I read this article on Core Memory (this one is not paywalled) about Terraform Industries
They are trying to produce cheap synthetic natural gas from the atmosphere by extracting Carbon from CO2 in the air and energy from the Sun. Natural gas (mostly composed of Methane) is probably the cleanest form of fossil fuel (Methane is still a greenhouse gas, which, if leaked, can definitely harm the atmosphere, but it’s cleaner than coal or oil).
Getting CO2 out of the atmosphere is definitely a good thing, as it is a greenhouse gas, but I still don’t like the idea of producing Methane. Maybe in the short term, this might be the right thing to do, but I am very pro-solar, wind, and nuclear energy, which are the cleaner forms of energy.
Ok, this one is actually scary and fascinating. A company called Cortical Labs put 200,000 real human brain cells onto a silicon chip and trained them to play Doom. They were able to do it in one week. This is CRAZY.
This is called biological computing. This company has a new product called CL1, which was used to do this. They provide a cloud service that lets you deploy your code to living neurons. No lab required.
Here’s a video of human cells playing DOOM.
Incidentally, I bumped into another article about the implications of this. But this article kind of says what we all have been thinking: are LLMs conscious? They may not, but putting them together with living cells like this might be the way to add consciousness. The cells can provide feedback to the LLM and receive a response. This area needs to be closely monitored over the next 2-3 years.
Interesting X posts
This is somewhat old, but I bumped into it again. Still relevant
Miscellaneus
US Govt declassifying UFO & UAP findings - https://www.war.gov/UFO
A different perspective on the current state of the internet - https://matduggan.com/the-intolerable-hypocrisy-of-cyberlibertarianism/
a16z on AI is not eating our jobs (which I think is right)
