Big Technology Podcast
00:42:20 · Alex Kantrowitz, Ranjan Roy
[00:42:20] S0: could solve loneliness, and we need to be more well. And this could solve [00:42:25] S1: all those solved the data center problem, the loneliness epidemic, wellness and longevity in health. [00:42:34] S0: All I'm saying is try it once you listen, try once if your community still votes to put a moratorium on your data center, then I'll apologize. [00:42:42] S1: I'm in. You're right. I'm in. I'm sold. I like it.
00:42:42 · Ranjan Roy, Alex Kantrowitz
[00:42:42] S1: I'm in. You're right. I'm in. I'm sold. I like it. [00:42:45] S0: Listen, let's take a break. Because, even if there might be some financial up and down to the technological process, progress continues unabated for now. We'll talk about that. And then we'll also discuss the role of prediction markets in this election. We'll be back right after this. And we're back here on Big Technology Podcast Friday edition with Ranjan Roy of Margins. Alright, Ranjan. I don't know how deep we can get into this, but it's very interesting. METR, right, which is the the organization that monitors autonomous AI did an analysis on Anthropic r and d. Let me read it, to you, so you can react here. This is from the information Anthropic r and d speed may have doubled due to AI researchers find AI use within Anthropic may have doubled the company's pace of AI development. A new report from AI research nonprofit meter says the rate that Anthropic's models are improving as of this October is roughly twice twice the rate at which they improve between March 2024 and October 2025. METR estimated around 80% of the acceleration in Anthropic research pace is due to internal use of AI within Anthropic. This is, of course, after Anthropic has given METR employee like access, though Anthropic did restrict a lot of what meter could publish because of what it called proprietary information and meter complained about that. Are we not fully taking seriously how fast this technology is getting better? This is a little bit crazy.
00:44:12 · Ranjan Roy
[00:44:12] S1: I don't know. I'm sorry. Like, I know I've gotten... It's just every action taken in every one of these cycles just makes me more skeptical and cynical even though I know this technology is incredibly powerful, and I'm sure the unreleased models are... Can be dangerous. Even in this example, again, so much information was redacted, and there was a backlash against... From what... One of Anthropic's cofounders who was trying to kind of, like, laud their work that was being done and how they're getting ahead of it. And again, as he said, Sam McCandlish, he was like, we've redacted a lot and I accept this makes it harder for others to check meters work, but we determined that sharing exactly what is driving the speed up isn't the right thing to do. So they go out saying there is this significant acceleration that they're... We're entering scary uncharted territory once again, coincidentally right as they're trying to IPO into a very dangerous market. But then we will not tell you exactly what any of this is or how it works or what it means because... And again, using security as the blanket to actually try to like cover this, it's still, I don't know. I just just have an honest open conversation if this is truly terrifying. And I just want to note, and I loved when Alex Stamos was on a couple of weeks ago. Well, I didn't love the fact that I left that conversation more terrified than I've ever been, but the dangers and the risks are bad actors using AI not agents going rogue. And I still... All these meter reports, all these Anthropic security kind of, like, reporting, it's always around agents going rogue and painting this picture of that rather than, like, how does North Korea not get access to these things?
00:46:08 · Alex Kantrowitz
[00:46:08] S0: Alright. I'll just have one thing to say here, which is if I were to push back on you, so I'll just give it a shot. I would say that these companies have been saying that the releases are going to accelerate. Year after year, I remember being at the Anthropic Dev Day, quote, quote with Claude last year, which is crazy. That was just last year. And they said the releases are going to start to come faster, and the technology is going to get better, faster. They kept their word, the releases are come faster, and technology has gotten better, much more quickly. And so, meet are going in there and saying that they've almost doubled the rate of progress, to me is not a stretch. And, you know, as they've been able to make these leaps, we've seen, like, new products become, available or possible because of it. Like, I'm thinking about dots or muse, right, which was not possible last year. So I am curious, what we'll see after this batch of improvement shows up. And I I lean towards believing
00:47:14 · Ranjan Roy, Alex Kantrowitz
[00:47:14] S1: that. Well, no. Okay. Here's the thing again. I don't not believe it. I I was actually thinking about this the other day that, like, part of me finds it ridiculous because when, like, Dario and whoever else was saying the threats to humanity back with like the GPT one days, they clearly saw where things were going and were right, were completely right about where things were going and how this technology would evolve. To me, still the gap though is these meter reports and this kind of like, this kind of work is not actually useful at all in this entire conversation. It's not gonna help solve any of these problems. It's not raising any awareness to a point where we can do anything about it. So that's the part that bothers me. Again, I'm discounting, again dude, video editing blows my mind. What has happened in the last six weeks? Like, it is crazy what is available right now and how much better it's gotten. [00:48:18] S0: Wait. Say what you've what you've seen.
00:48:18 · Alex Kantrowitz, Ranjan Roy
[00:48:18] S0: Wait. Say what you've what you've seen. [00:48:20] S1: I wanna hear So now you can lit... Like, motion graphics on video, you can just ask AI to do. And Claude has already done it, and I'm sure it's gonna be everywhere very soon. Like the most complex thing, even prompt based video editing of like being like, through this, eliminate all the jump cuts, improve the audit, you just prompt it now. And this did not exist like two months ago. All of that, so yes, the pace of innovation continues and that is heavy compute complex type stuff and it's working. So so to me, it's less about is the is the is the model front... The frontier evolving at a pace that is still dramatic. To me, that's not the question. It's are you giving us some useful information about what to do about it or just kind of maintaining your position as like the god in all of this ahead of your IPO that is in a bit of trouble. I know I shouldn't look at it just like that, but I can't help it. [00:49:24] S0: No. I hear you. I hear you completely. Alright. Before we leave, give us your thoughts on the... Tell tell everybody about the prediction market debate that has sort of spiraled over the past week and what your thoughts are.