FDE Episode :9 An analysis of AI news and why time compression is becoming a global equalizer

Hi everyone. Welcome to the code Ninja spaces and today's episode we are covering a mix of hard infrastructure, some research and culture shift that quietly signals that how fast the world is changing. So let's start with the news one of the day civilization talks with it now compute. So basically what's happening? There's an organization like EVOK AI that track large scale organizations training data and the trend is clearly that AI compute is compounding really fast, like extremely fast. So I have a very. We often hear that compute is doubling after every few months. So the question here from you Umar, is that if compute is compounding this fast, what changes fast? The model, the product or the science behind it?

Everything. So one of the things that we are seeing is that the whole infrastructure so the amount of time it used to take to train pre train a model is now cut to half now. Similarly at the inference the whole memory infrastructure that we have now is far more than for example 6 months back. So if you see this AI is going to improve by at least 20 to 30x only in 2026. What it means is better cloud code. So for example I hear a lot that engineers are not able to design complete applications. There's something missing if you're using models to for example for better UI or ux there are things that that require human input. I think these things you would see. I mean I was with the demo with Ali on Hyper, we were using cloud board at the last mile and on top of Hyper and what it just blew my mind how he built those Kanban boards and all that stuff, all those things and generated code for it and just the need of UI and UX and front end engineers and backend and software. What I've seen firsthand is like what models can do. So so this all is possible because of that compute infrastructure that's powering all this. And what it means is not just GPUs or these things, it means energy very large data center. So I think in 2026 you would see models performance would improve to a point where some people would feel that AGI is either it's in the models now or probably a couple of months away. So for some people this year will be very scary.

Yeah, so our second use is also about the models the prompt. So the news is basically repeating the prompt improves the model performance. So what happened? Basically the Google researchers published result that showing that repeating the input prompt without increasing the output tokens can improve performance in non reasoning runs across multiple models and benchmark it's mean no extra Output just better results. So the question here is basically like what's in your opinion, what does this teach us that how LLM works?

So first of all, I mean I would just reiterate it like so I mean if you give models the same prompt twice, like for example, if you are saying okay, can you do this or whatever you've asked for a model, just repeating the prompt prompt makes the model better. So that's a hack. I mean if someone is listening and they're trying to make model work better. So think about it like you would give the same prompt and model's performance would better in the second question. Now there are two, three things that we need to understand here. Some of the people find it hard is it possible because they're coming from the era of determinism where they ask computer to do a certain thing and it exactly did what it was supposed to do. One is like no, I mean it means just talking to a computer and it would give different results. It's not randomness, it's how neural words works, it's how human brain works.

Right?

That's at a macro. We need to get used to of this non determinism in the models. But the second thing I think the more important is that models are not just dependent on the context. It depends on the conditioning which is like it depends on allocation of attention. Right? If model feeds like that we are giving way too much attention to a particular thing, they would respond better. Right? And in the prompt you can basically you're changing things at the inference during the. I mean the better your question would be, the better your attention mechanism, the better your conditioning would be. Model would perform better. It's perfectly fine for two people talking to a same model. And one person gets a lot out of model, others can't. So that's what it is. But yeah, I mean it's a fun fact too. Like you asked the same question twice and models behave differently.

So our third use is again related to the compute, but this time really different. The news is basically using the air itself as compute. So what's happened? Basically the researcher from Duke with the collaborators including mits described a system called wise. The idea is basically to encode model based directly into the radio frequency signals. So that's something really interesting like using physical mind bending, not just digital shifts. So in your opinion, why should people need to care about Ars in computer?

First of all, it's something that they're trying to experiment with and they have some very good results to share. So I mean historically the Information used to trans. I mean, I mean if you're using radio signals, so basically all what are these are, These are waves, right? They're being used to encode signals and information, right? But now we have another kind of information. The information is not all fact or something. Information is in two forms, weights. Right. Imagine if someone can outsource their brain, just the weights of their brains and if you can transfer those brains as signal, right? So you could be using someone's brains using radio signals. So what it. I mean, and forget about. It's not about using Claude or using any model. No, that's what you already use, right? That's not what it means. It means actually transmitting the weights into the air. That's why it's called using air as a compute. So historically when you say that when you're using air, you're using air just as a medium to transfer information. Now you're using air as a computer chip to perform compute. Just think about it. I'll say it again, you're using models, you're using model weights as an analog signal, right? And then if you are transferring model weights as an analog signal, you can perform computer number one. You can perform a very heavy compute on the edge, but you can also use AIR to perform computations. So that is mind bending. Let's see where it goes.

So our port news is also related about drones. So drones that recharge from the power line. So this is something really interesting as well. What's basically happening, there's a company called Walt Air is developing drones that can pouch on the power lines and recharge continuously in sections without returning to the base. So what's the main question here is that when report like from drones to the boards effectively have infinite range, what's going to be changing and what's happening?

So the news is like drones can autonomously take power from the power lines, right? So one of the major challenge of drones is like you have to swap the battery after a certain flight hours. That's a challenge in tough terrains. That's a challenge for very long operations. So what currently this company is experimenting is how to build a continuous infrastructure of drones where they can just drop power from the power lines on their way to their missions. Right. There's no battery swap infrastructure, there's no physics that says it's not possible. But it's not that simpler too. So for example, I was listening to Elon podcast and they were building this huge power infrastructure for Xai, although he said power lines were going next to our facility, but they couldn't draw power from those power lines. The reason for that was like it would take a year to build that infrastructure to draw power from those power lines. So what's better for them was like they built their own infrastructure from gas and all that stuff to generate power rather than just drawing power from the power line. So it won't be that easier. But yeah, it looks cool. The problem is there that battery swapping. I've spoken to a lot of companies who are building drones that they talk this a lot. The continuous operations of drones are really a challenge.

So our fifth news is something that's relatable with me as well. So there is news is that AI generating generated music becomes normal listening behavior. That's something some few days ago someone behind me was making as well. So the report is basically that there's a survey that suggests that 50 to 60% of listeners from the age group of 18 to 44 now spend around two to three or every week listening to AI generated music. So this is not only a report primarily or trend that come up across multiple like thought process as well that if culture becomes synthetic, what happens to originality?

So just think about like 50 to 60% of the listeners from 18 to 44 are saying that they are consuming two to three hours per week of AI music. I don't know what it means for the culture, for the overall industry, that music. I have consumed some of the tunes and they're very good. But as you, as you. So normally when you consume the music, you try to associate that with someone who is singing it or is creating that music somehow that, I mean, as you are, the tune is very good, but you somehow are finding it very hard to who to relate to because there is no one to relate to. So that's a challenge that in terms of culture and that the mixing is so good. At one time you feel like that probably it's an Argentinian music and also. And it turns into a Portuguese and then to an Indian or a Pakistan. So you find it very hard to understand like what it is, what's going on. But it's really good. So yeah, I mean, I think I don't know what it means for the music industry. It's. I don't know. There's so many questions about AI. We really don't know.

I mean, should we start to create AI generated music? For me it's fun. So our 6News is related to the Elon. We really talk about him open and his platform as well. So explains algorithm transparency, Elon stated that explain to open source its recommendation algorithm on our recurring biases. So what's basically happening is that they open source their whole code and algo related to the recommendation engine. So the question and your take here is I want to ask that why does open sourcing our social algorithm matters now? Is it freedom of speech or something?

No, no, it's not freedom of speech. So the stated goal is that this study shows that it perhaps would increase user time on X by 20%. I'm not sure how they calculated it but it's like the claim is that 20% more user time is basically the goal here. But I'm not sure the authenticity of this news. Elon talked about open sourcing X algorithm, but I think this year I don't think it's on the cards. I mean how would it improve freedom of expression? How would open sourcing axe algorithm would drive user engagement? I'm not sure about this. How would it help other businesses to bring to build capabilities very specific for X And probably that could be one of the channel, one of the thing that is like X is X needs to innovate on two fronts. Number one, so think about LinkedIn. LinkedIn has two major revenue engines. One is hiring. So LinkedIn hiring platform for recruiters, talent recruit and second is Sales Navigator for sales teams. So these two basically areas are the biggest drivers of LinkedIn. Now no such infrastructure exists for X. So currently no one uses X for hiring because the underlying infrastructure is not there or no, I mean people should use, must use. I mean but as a dedicated sales team are not using X, they're using sales navigator. Although X is far more valuable right in terms of connecting with the sea level. So if you really see these two things or other such kind of components, Elon had a plan to build those, but probably outsourcing these kind of things maybe allow the innovators to build cool products like these around X and maybe that's, that's how you can bring more user base and let people spend more time on X. And that would be a good thing too because I've seen, I use LinkedIn a lot, it's great but it the people I used I tried to connect years on LinkedIn I just connected with them on X in seconds. Right, I've got to reply and all that, whatever. I mean so X is the platform to connect with the decision makers but it's not wired for sales or hiring infrastructure. So I think that is something that I think probably these open sourcing will.

Lead to maybe so obvious the 7th news is related to Amazon Amazon agents buying from the competitors. So there's very interesting news. Amazon previously tested on agentic features called Buy for me that allow you allow AI agents to purchase from the third party websites on behalf of user. So what's your take on it that are we outsourcing the consumption Amazon trying to become?

So there are two things happening with Amazon. Number one is Amazon is not somehow they feel like they're missing on AI race because people are going to ChatGPT and doing comparisons there and taking the buying decisions there. So after Google one of the biggest search was within Amazon. So for example let's say if you want to buy shoes you would go and ask Amazon search to look for products and all that. But now those that kind of search is happening inside ChatGPT and models. So that is giving so that's basically taking traffic away from Amazon internal search. So Amazon is trying to build universal shopping assistants, right? So for example, what can be done by the way, this biopharm is not for internal Amazon website. They're experimenting in for third party websites. So but yeah, I mean I think there's a lot of value here by the way. So I have seen few agents working as a shopping assistant. There are a few products that there's no no fun shopping those products but you need those. So for example, if you're looking to buy a good part for your automatic machine in the home, let's say, right, there's no fun buying those and you want to pro that all you want is outsource the star. Just like for example you ask someone, let's say an office buyer look, I mean I need this, my machine is faulty. Go shop this and come back. And then he searches for that and come. There's lots of Internet shopping areas where you don't want you to shop. Not everything is buying from these clothing lines and buying shoes. It's not fun. So all those parts of shopping can be outsourced to agents who would just look at take the spec sheet and just going to the Internet and it won't be search. So think about people think of it more of like talking to Google and web crawling. It's not that it's basically actually searching those products, virtually negotiating all that stuff and getting a good price and then eventually buying it on your behalf. So that's what buy for meeting is.