"Diamond?" asked Jack.
"Yep," said Rina. "The only thing with higher thermal conductivity than diamond is graphene. But with graphene, that only applies when the heat is traveling parallel to the graphene sheet surface, but not perpendicular to it. So I combined the two to get the best result."
Rina did sothing with her virtual interface, and a semi-transparent cube appeared between them.
The cube had what he assud were one thousand layers, each a separate silicone wafer. Across each wafer, and between the wafers, were thin sheets highlighted in red connected to tubes that passed through the wafers, highlighted in yellow.
"The red is where I put graphene, and the yellow is where I used diamond. The graphene is molecularly bonded to the diamond, so technically it’s all just one giant molecule. When it’s running, we have to pump hundreds of gallons of water per minute through the core to carry away the heat."
"This is amazing work, Rina. You’ve achieved a thousand tis increase in compute power for only a four tis increase in electricity."
"And I’ve already upgraded my server sphere," said Madison. "It only has 25 million servers now, but I have 250 tis more cores than I did before. And each server can handle models that are a thousand tis larger than what the old Cerebras servers could handle."
"And once I get the micro-core architecture working," said Rina, "the core count will be closer to a trillion cores per server, but each server will be at least ten tis more efficient, and maybe as much as one hundred tis more efficient."
Jack shook his head. "It’s amazing how much you all have accomplished in such a short period of ti. And I’m not just talking about physical world ti. Even with the extra soul space ti you’ve had, what you’ve accomplished is amazing."
The rest of the girls started filtering in, and soon their eting started.
Besides Madison, Rina, Nora, Samantha, Katie, and Naoko, he had invited Lorie, who had taken it upon herself to drag Isabella into the eting.
He started the eting by looking at Nora. "Nora? Any progress on Lars’ cancer cure?"
"Yes, killing his cancer turned out to be easy. The problem is, my treatnt kills his cancer so fast it causes tumor lysis syndro, and his kidneys fail shortly after. So right now I’m trying to figure out how to moderate the process so his kidneys can keep up, but I think I’m close to a solution."
"That’s great to hear. Any progress on the nanotech front?"
"Yes. I was able to create a small nano-cell aluminum battery using only a nanite solution. Unfortunately, it’s just a proof-of-concept at this point because it doesn’t scale. I still have energy transport and heat dissipation problems to solve."
"Still. That’s years ahead of anything anyone else has achieved."
She nodded. "It is! Rina’s latest upgrade to the molecular modeling software has really helped. I might even have a fully nanotech battery fabricator ready in ti for small-scale production in a few months. Maybe even sooner at the rate things are going."
"Let’s consider putting the batteries on the market after we deal with the first generation of our new VR headset."
Then turning to Rina, he asked, "Anything to share besides the molecular modeler upgrade and your new giga-cube?"
"I was able to incorporate optical interconnects in the giga-cube, so inter-core bandwidth is ten tis what it is on a Cerebras. And I’m getting closer to a fully photonic processor."
"Great work, Rina," he said before turning to Madison.
"I haven’t solved AGI yet," said Madison. "But I think I have sothing that might be just as useful."
"Oh?"
"Rember that general human knowledge database I told you about and how I was trying to use it to clean up all the training data scraped from the internet?"
"Yes, did you have any success?"
"Yes, I did. It’s even better than I had hoped. It’s not AGI, but it makes Yocto search seem like a toddler in comparison. It even says ’I don’t know’ if you ask it a question it cannot answer!"
"But can it do your math or composition howork?" asked Miranda.
"Mostly," said Madison. "It can solve most math and logic problems, but it’s not great at composition. It’s really good at sifting through docunts and answering questions, but not great at prose. It doesn’t have any creativity."
"What you’re describing sounds like what Yoctoly already offers," said Jack. "It being able to say ’I don’t know’ sounds great. Better than it hallucinating so bullshit. But what makes it so much better?"
"Existing AI models can’t summarize. Not really. They just shorten content, or regurgitate soone else’s summary. But my model can actually summarize. And even better, it will automatically identify conflicting opinions in content. Look, it’s hard to describe."
"It really is better," said Rina. "I was struggling to find existing research to accelerate my own work. Mainly because I had trouble asking the right questions. But with Madison’s new AI, I just described the general problem, and it suggested several fields of study I’d never even heard of."
She sighed. "It only took a couple more questions to find the papers I was looking for. Less than ten minutes with Madison’s new AI answered questions that days of searching on Yoctoly couldn’t answer."
"If we ever make it available to the public, I suspect there will be a lot of turmoil in the scientific community," said Nora.
"Why?" asked Jack.
"Because it was trivially easy to find plagiarism, logically flawed argunts, faulty math, and outright lies in thousands of papers. I picked a well know paper that was recently proven to have invalid claims, then asked Madison’s AI to identify papers that cited the refuted paper."
She shook her head. "The AI found all the papers with citations and also identified which papers’ claims were invalidated because the cited paper was invalidated."
"Ah. If you can think to ask those questions, then so will other scientists and journalists," he said.
"Yes. It will cause a shitstorm. But I think it will be good in the long run. It will make it much easier for young scientists to filter out the bad papers and that will reduce the incentive to publish BS."
"It sounds like we should keep Madison’s new AI in-house for now and only make it available to the public once we are firmly established as an AI leader. Otherwise, we could get drowned in lawsuits."
The rest of the eting went quickly. Naoko was studying the nanotech components Nora was designing. Samantha was making progress in understanding subatomic particle math, and Miranda said her professor had approved the plan for decorating his mansion.
Isabella said the construction of his research institute was ahead of schedule and Lorie was waiting to get moved into his mansion before applying to a chef school.
"Okay, that’s it for this eting. Just rember, Saturday is beach day, and we’ll spend the whole day at normal ti instead of accelerated ti."
"Yay, jet skis!" said Naoko.
Reviews
All reviews (0)