Behold the decade of mid tech!
That is what I want to say every time someone asks me, “What about A.I.?” with the breathless anticipation of a boy who thinks this is the summer he finally gets to touch a boob. I’m far from a Luddite. It is precisely because I use new technology that I know mid when I see it.
Anthropic made lots of intriguing discoveries using this approach, not least of which is why LLMs are so terrible at basic mathematics. “Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95,” the MIT article explains.
But here’s the really funky bit. If you ask Claude how it got the correct answer of 95, it will apparently tell you, “I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95.” But that actually only reflects common answers in its training data as to how the sum might be completed, as opposed to what it actually did.
CoreWeave Is at the Center of the AI Revolution, and Its IPO looks Like a House of Cards
Today, CoreWeave began trading on the stock market, and it is looking like an inauspicious start for the AI sector. The company had initially hoped to sell shares for $47 to $55 a piece, but began trading at $39 despite being propped up by Nvidia with a massive $250 million order at $40 per share.
I work in natural gas and a provider of natural gas in TX has taken the official stance that they will not be offering gas to “data centers” (aka AI) because of the crazy amount of gas they use would dry up the well for every other type of customer so to speak. I can’t speak too much about it beyond that but I am worried that the amount of gas they need is irresponsibly high.
I’ve worked in data centers for decades, with absolutely nothing to do with AI, and that kind of loose wording gives me chills. The data centers I’ve been at generally relied on gas for backup power, to ensure continuity for essential services for healthcare providers, etc. I hope they are being more specific in their stance than that.
In my world they’re being referred to as data centers so far, but the gas usage is obviously power generation instead of on-demand backup power. Which is a giveaway that its not really a typical data center.
I suspect they mean “new” datacenters. The demand for inane numberchrunching has created an artificial bubble of power guzzling AI-specific datacenters that are more similar to cryptofarms than regular datacenters.
Would not be surprised if they start regulating how much power and water can a datacenter consume before it falls into a different category.