HN user
Anyone have any idea what the architecture/vendors they are using for inference/compute?
Getting the compute to run inference for multi-trillion parameter models at any sort of scale and performance is daunting. There are a handful of vendors that have systems that can do this (~ Nvidia NVl-72 class) that pretty much only the frontier labs and hyperscalers effectively have access to.
Non-profits have treasury functions.. it is prudent that many of them have some sort of cash reserves.
That statement means nothing. You could say the exact same thing about Rails and have an equally defensible position. What about its architecture makes it better?
Been following the vesuvius challenge and to me this is nothing short of alien magic tech. Incredible work.
Apple was on the USB Implementers Forum that designed USB-C so.. I would say they could definitely be credited as a co-inventor of USB-C, they also introduced one of the first devices that used USB-C.
There are some studies that suggest human brain sizes have been shrinking over the last 20,000 years. The theory is that as civilization developed the demand for individual humans to be independently intelligent has weakened because we developed a "collective brain" and also self-domesticated to be more cooperative.
Yes. Absolutely. They will move more units and make more profit overall, and if they don't do it a competitor will.
We use headless browser providers because the companies we interact with don't and won't create a proper API for us to use. Lots of legacy web apps/portals. Saves thousands of man hours.
There are no true scotsman
Everything is obvious in hindsight, but the data (and theory) at the time was that this had a really good shot at being the next big thing in a world where >90% of drugs never make it past clinical trials. 10% probability of success * $200B in lifetime sales (assuming a Keytruda level smash hit) means an EV of ~$20B or more. Not a surprise more than a few companies wanted a shot at it.
If you don't value your time sure. Pen and paper is also basically free and even simpler with no setup time..
Soooo.. not to sound like a luddite but to me the best dedicated writing device for me has been just pen and a notebook or a typewriter.
There are surprisingly many "portable" typewriter options out there (including electronic ones).
How so? Most of these companies will take a hit but will be fine Alphabet, Amazon, Google, etc can write off their entire investments in AI and will be a-OK. The pure AI companies will obviously be dead.
More like poorly structured loans and incentives.
What do you mean? Pandoras box has already been opened. Even if OpenAI disappears, there will be another one to take its marketshare. The tech is too useful to die
I wouldn't blame that incident on Railway.. you can delete your prod database on AWS just as easily with their API.
Been a customer with them for over a year now, small incidents here and there but never anything this major.
Not strange, Google has never had a proper support team unless you are an "Enterprise" level customer.
They run a decent amount of their own compute/bare metal server for customer workloads. But likely still had some critical dependencies on GCP.
Absolutely dumb take. There are plenty of very bright and talented people that would have made excellent teachers but chose different career paths because - surprise surprise - the pay is better.
Yep, the COP goes down as the temperature goes up, and at a certain point it's not worthwhile increasing the temperature.
Stefan-Boltzmann law means radiative heat transfer in space is approx. to the 4th power of the hot side of your radiator. Typical space based radiators operate around 350K. If you can increase the hot side of the refrigeration cycle by 4x (1400K) you increase heat transfer by 256x. Create a radiator design that can operate at this temp (multi-stage Brayton loops, heat pumps, possible liquid metal final stage) with a large enough surface area and now a datacenter in space seems possible.
It's a difficult engineering challenge but physically possible, and Elon is no stranger to engineering challenges.
Some numbers: assume an emissivity of 0.85, assume no absorption from the sun, assume heat rejected from both sides of a panel, a 1m^2 panel will reject 1.45kW/m^2 @ 350K.
At 900K its 62 kW/m^2. Not a trivial amount of heat.
Moved object storage from AWS to CloudFlare and have been pretty happy. No problems with performance so far. Bills were 90% cheaper too (free bandwidth)
Agreed. The market should decide if beef consumption is viable. Ultimately energy is the basis all food production. Cheap and plentiful energy solves the food production and distribution problem, then its just matter of preferences.
Well.. farming equipment are high 6 figures 7 pieces of business equipment (the lifetime operating costs are definitely in the 7 figures.) These are owned and operated by people who I would expect to do this type of research and critical thinking. These aren't normie consumers buying everyday appliances or electronics.
However.. farmers are a weird bunch and they are blinded by brand loyalty or will only buy from an "American" company which ironically allowed JD to stomp all over them because of their dominant market position.
Bootstrapping an electronics supply chain on another planet seems harder than building the dyson swarm itself.
Probably not that flexible without some big cancellation fee. It can take 4-6 months to go from bare wafer to packaged chips.
Investors are not part of the market?
I speculate that they are hitting the reticle limit for models not much bigger than this. Judging by the size of the chip in their demonstrator for a 8B model I'm sure they know this already.
To scale this up means splitting up large models into multiple chips (layer or tensor parallelism). And that gets quite complicated quite quickly and you'll need really high bandwdith/low latency interconnects.
Still a REALLY interesting approach with a ton of potential despite the unstated challenges.
Apple has products in their lineup where they refresh and keep the name. Example: Mac Studio is the same every refresh.