DeepSeek And ASI-1 Mini: A Closer Look At AI Computing Optimization
AI is advancing faster than ever—that much is clear. But what's often overlooked is the knock on effect on computing power, which is struggling to keep up with demand. With models like DeepSeek and ASI-Mini 1 introducing smarter architectures, it might seem like we're on the verge of a solution. Yet, this opens up a bigger question—are we solving the compute crisis, or are we actually accelerating it?
The common denominator between DeepSeek and ASI-Mini 1 is their use of Mixture of Experts (MoE)—an architecture incorporating multiple expert sub-models. Rather than engaging the entire model for every request, MoE selectively activates specialised expert models, reducing computational strain while maintaining performance. This approach enhances compute efficiency, scalability, and specialisation, making AI systems more adaptable and resource-conscious. This breakthrough has highlighted the growing importance of MoE in AI development.
While both models employ MoE, ASI-Mini 1, built by Fetch.AI, takes this even further by incorporating Mixture of Agents (MoA) or Mixture of Models (MoM). As an example, MoA allows multiple autonomous AI agents to collaborate, optimising resource use and making AI more adaptable. Not only excelling in expansion, but also becoming the world's first Web3 large language model.
Optimised compute usage should, in theory, reduce overall computing demand. However, it's not that simple. Jevons Paradox suggests that efficiency gains often lead to greater adoption, ultimately driving demand even higher. DeepSeek's ability to deliver high-performance AI at lower costs is a prime example—by making AI more accessible, it fuels greater investment in AI projects, intensifying the need for infrastructure. As a result, the focus shifts toward ensuring solutions are not only cost-efficient, but also scalable and adaptable to sustain AI's rapid growth.
Both LLMs and AI Agents are intensifying this demand, requiring substantial computing power for training, inference, and real-time decision-making. LLMs, particularly the latest iterations with billions of parameters are computationally expensive not just during training but also where they process massive datasets and in inference, where generating responses at scale remains resource-intensive.
AI Agents, operating in dynamic environments, introduce continuous workloads, constantly analysing incoming data and making autonomous decisions in real time. This sustained computational demand places additional strain on infrastructure, requiring consistent access to high-performance compute resources.
As highlighted in Aethir's analysis, GPUs remain the foundation of AI infrastructure, yet their high costs, supply chain constraints, and availability pose significant challenges for businesses scaling AI operations. This surge in AI adoption makes high-performance, cost-efficient, and scalable infrastructure an imperative, particularly as businesses seek flexible, transparent, and globally distributed compute solutions to maintain a competitive edge.
The market isn't just seeing incremental advancements. What we're experiencing is an infrastructural shift, where companies must rethink how they build, deploy, and sustain AI systems. That's the new status quo.
One of the biggest shifts is the broadening of AI applications which are no longer limited to research labs or enterprise automation, AI is embedding itself into consumer products, financial systems, and real-time decision-making engines. AI agents, once a niche concept, are now being deployed in autonomous trading, customer interactions, creative fields, and decentralised networks, all of which require constant, real-time compute power.
At the same time, we're witnessing an evolution in how AI infrastructure is funded and scaled. SingularityNET's $53M investment in AI infrastructure reflects a broader trend: businesses aren't just developing better models—they're strategising around compute access itself. The scarcity of GPUs, the need for decentralised compute solutions, and the rising costs of cloud AI infrastructure are becoming as critical as AI model improvements themselves.
But, how will companies sustain this level of growth? Even with MoE and its extensions reducing computational inefficiencies, the demand isn't shrinking—it's accelerating. Companies that once focused solely on AI capabilities now must navigate compute economics just as carefully. Those who fail to plan for infrastructure growth risk being left behind.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Tom's Guide
3 hours ago
- Tom's Guide
These are the 10 prompts in DeepSeek to help you create a budget
Creating a home budget doesn't have to always involve an Excel spreadsheet and coffee to keep you awake during the process. Instead, AI bots can guide you through a budget and make it much easier to track your income and expenses. I should know — I avoid the topic as much as possible. For years, I calculated most of my expenses in my head and hoped it all worked out. Roughly about 10 years ago, I started relying on apps like Mint to help me and even the tools provided by my bank. Recently, I tried using DeepSeek to guide me through the process. It's amazing how these 10 prompts can help you figure out where all of the money is going. Here are the best ones to try. Before diving into the details of a home budget, you can start by asking DeepSeek which home budget works best for you. Fortunately, this prompt provides a wealth of information, no pun intended. DeepSeek presents several options for home budgets including the popular Zero-Based Budget' that helps you track income, expenses and savings down to the last dollar. In that home budget, you put money into savings first, then allocate money to other things. This prompt really narrows things down, because it means you'll need to budget more aggressively. DeepSeek recommends several budgets but two of them are worth calling out. The first one is a percentage system called an Aggressive Savings plan. 50% goes to needs, 20% to savings, 10% to debt, and 20% to wants. DeepSeek also recommended the zero-based budget, which prioritizes savings even more — but maybe a little too much for me. I decided to try the Aggressive Savings plan because it looked practical and achievable to me. I was impressed with all of the detail DeepSeek provides here, listing a full table of expenditures with suggestions for how much to set aside for groceries, gas, and credit card debt. Using this budget means I could save $192 per week — reaching my $10,000 per year goal. So far the prompts and suggestions have all been generic — they could apply to anyone. I decided to make it more personal. I asked about customizing for my income level and DeepSeek then guided me through the entire budget process. The AI asked for my annual income and expenses including car payments and student loans. DeepSeek also asked where I sometimes slip up — say, by overpaying for Uber Eats. I inputted all of this info. (Truth be told — I made up some details. I am not yet comfortable adding actual data to an AI bot yet.) I noticed my final home budget required saving about $833 per month, which seemed a little high. I asked DeepSeek to adjust my budget to save a little less, and suddenly it all seemed more practical. I was allocating more money to eating out and gas/groceries without having to sacrifice so many things that I'd end up failing. I was happy with the results so far. I knew at this point that a home budget would not be useful if it was just contained within a chatbot. I asked DeepSeek to generate an Excel file and was a little surprised when it generated a CSV file instead. (I'm sure there are copyright issues with providing an Excel file.) No matter — I imported the file into Google Sheets using the comma-separated values. A home budget is not just about the numbers and creating a spreadsheet. What works the best is when you have an action plan each month that helps you adhere to the budget. I was surprised, though, when DeepSeek offered some unusual suggestions to help. The bot said I could consider doing a side gig to help generate more income, which seemed smart. Another surprise is that the bot suggested a temporary pause on some charitable donations. Mint is a popular home finance app and I've used it many times. Ironically, I had never actually important a spreadsheet before and DeepSeek noted that is not even possible. Instead, the chatbot suggested inputting all of the info manually, which seemed clunky. And guess what? The bot actually used that word. DeepSeek said 'If this feels clunky…' to try some competing apps that do support imports. DeepSeek also offered to show screenshots of Mint to help me see where to input the budget info. This process took several minutes, though. And, they were not real screenshots — instead, DeepSeek created text-based guidelines. I asked DeepSeek for more guidance, and the bot covered quite a few practical tips for managing my money — one involved using a cash-based budget where you put money into envelopes for different spending categories. Another tip was to wait 24 hours before spending X amount of money (say $5) if it is an impulse buy. Lastly, I asked DeepSeek to save my home budget and all of the tips. It turns out, DeepSeek can't save anything as a text file, so the bot suggested I copy all of the text and paste it into Google Docs or another word processor. That worked just fine for me! Get instant access to breaking news, the hottest reviews, great deals and helpful tips.
Forbes
3 hours ago
- Forbes
The Ardent Belief That Artificial General Intelligence Will Bring Us Infinite Einsteins
In today's column, I examine an AI conjecture known as the infinite Einsteins. The deal is this. By attaining artificial general intelligence (AGI) and artificial superintelligence (ASI), the resulting AI will allegedly provide us with an infinite number of AI-based Einsteins. We could then have Einstein-level intelligence massively available 24/7. The possibilities seem incredible and enormously uplifting. Let's talk about it. This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here). There is a great deal of research going on to further advance AI. The general goal is to either reach artificial general intelligence (AGI) or maybe even the outstretched possibility of achieving artificial superintelligence (ASI). AGI is AI that is considered on par with human intellect and can seemingly match our intelligence. ASI is AI that has gone beyond human intellect and would be superior in many if not all feasible ways. The idea is that ASI would be able to run circles around humans by outthinking us at every turn. For more details on the nature of AI, AGI, and ASI, see my analysis at the link here. AI insiders are pretty much divided into two major camps right now about the impacts of reaching AGI or ASI. One camp consists of the AI doomers. They are predicting that AGI or ASI will seek to wipe out humanity. Some refer to this as 'P(doom),' which means the probability of doom, or that AI zonks us entirely, also known as the existential risk of AI. The other camp entails the so-called AI accelerationists. They tend to contend that advanced AI, namely AGI or ASI, is going to solve humanity's problems. Cure cancer, yes indeed. Overcome world hunger, absolutely. We will see immense economic gains, liberating people from the drudgery of daily toils. AI will work hand-in-hand with humans. This benevolent AI is not going to usurp humanity. AI of this kind will be the last invention humans have ever made, but that's good in the sense that AI will invent things we never could have envisioned. No one can say for sure which camp is right and which one is wrong. This is yet another polarizing aspect of our contemporary times. For my in-depth analysis of the two camps, see the link here. Let's momentarily put aside the attainment of ASI and focus solely on the potential achievement of AGI, just for the sake of this discussion. No worries -- I'll bring ASI back into the big picture in the concluding remarks. Imagine that we can arrive at AGI. Since AGI is going to be on par with human intelligence, and since Einstein was a human and made use of his human intelligence, the logical conclusion is that AGI would be able to exhibit the intelligence of the likes of Einstein. The AGI isn't especially simulating Einstein per se. Instead, the belief is that AGI would contain the equivalent of Einstein-level intelligence. If AGI can exhibit the intelligence of Einstein, the next logical assumption is that this Einstein-like capacity could be replicated within the AGI. We already know that even contemporary generative AI allows for the simulation of multiple personas, see my analysis at the link here. With sufficient hardware, the number of personas can be in the millions and billions, see my coverage at the link here. In short, we could have a vast number of Einstein-like intelligence instances running at the same time. You can quibble that this is not the same as saying that the number of Einsteins would be infinite. It would not be an infinite count. There would be some limits involved. Nonetheless, the infinite Einsteins as a catchphrase rolls off the tongue and sounds better than saying the millions or billions of Einsteins. We'll let the use of the word 'infinite' slide in this case and agree that it means a quite large number of instances. Einstein was known for his brilliance when it comes to physics. I brought this up because it is noteworthy that he wasn't known for expertise in biology, medicine, law, or other domains. When someone refers to Einstein, they are essentially emphasizing his expertise in physics, not in other realms. Would an AGI that provides infinite Einsteins then be considered to have heightened expertise solely in physics? To clarify, that would certainly be an amazing aspect. No doubt about it. On the other hand, those infinite Einsteins would presumably have little to offer in other realms such as biology, medicine, etc. Just trying to establish the likely boundaries involved. Imagine this disconcerting scenario. We have infinite Einsteins via AGI. People assume that those Einsteins are brilliant in all domains. The AGI via this capacity stipulates a seeming breakthrough in medicine. But the reality is that this is beyond the purview of the infinite Einsteins and turns out to be incorrect. We might be so enamored with the infinite Einsteins that we fall into the mental trap that anything the AGI emits via that capacity is aboveboard and completely meritorious. Some people are more generalized about the word 'Einstein' and tend to suggest it means being ingenious on an all-around basis. For them, the infinite Einsteins consist of innumerable AI-based geniuses of all kinds. How AGI would model this capacity is debatable. We don't yet know how AGI will work. An open question is whether other forms of intelligence such as emotional intelligence (EQ) get wrapped in the infinite Einsteins. Are we strictly considering book knowledge and straight-ahead intelligence, or shall we toss in all manner of intelligence including the kitchen sink? There is no debate about the clear fact that Einstein made mistakes and was not perfect. He was unsure at times of his theories and proposals. He made mistakes and had to correct his work. Historical reviews point out that he at first rejected quantum mechanics and vowed that God does not play dice, a diss against the budding field of quantum theory. A seemingly big miss. Would AGI that allows for infinite Einsteins be equally flawed in those Einsteins as per the imperfections of the modeled human? This is an important point. Envision that we are making use of AGI and the infinite Einsteins to explore new frontiers in physics. Will we know if those Einsteins are making mistakes? Perhaps they opt to reject ideas that are worthy of pursuit. It seems doubtful that we would seek to pursue those rejected ideas simply due to the powerful assumption that all those Einsteins can't be wrong. Further compounding the issue would be the matter of AI hallucinations. You've undoubtedly heard or read about so-called AI hallucinations. The precept is that sometimes AI generates confabulations, false statements that appear to be true. A troubling facet is that we aren't yet sure when this occurs, nor how to prevent it, and ferreting out AI hallucinations can be problematic (see my extensive explanation at the link here). There is a double-whammy about those infinite Einsteins. By themselves, they presumably would at times make mistakes and be imperfect. They also would be subject to AI hallucinations. The danger is that we would be relying on the aura of those infinite Einsteins as though they are perfect and unquestionably right. Would all the AGI-devised infinite Einsteins be in utter agreement with each other? You might claim that the infinite Einsteins would of necessity be of a like mind and ergo would all agree with each other. They lean in the same direction. Anything they say would be an expression of the collective wisdom of the infinite Einsteins. The downside there is that if the infinite Einsteins are acting like lemmings, this seems to increase the chances of any mistakes being given an overabundance of confidence. Think of it this way. The infinite Einsteins tell us in unison that there's a hidden particle at the core of all physics. Any human trying to disagree is facing quite a gauntlet since an infinite set of Einsteins has made a firm declaration. A healthy principle of science is supposed to be the use of scientific discourse and debate. Would those infinite Einsteins be dogmatic or be willing to engage in open-ended human debates and inquiries? Let's consider that maybe the infinite Einsteins would not be in utter agreement with each other. Perhaps they would among themselves engage in scientific debate. This might be preferred since it could lead to creative ideas and thinking outside the box. The dilemma is what do we do when the infinite Einsteins tell us they cannot agree? Do we have them vote and based on a tally decide that whatever is stated seems to be majority-favored? How might the intellectual battle amid infinite Einsteins be suitably settled? A belief that there will be infinite Einsteins ought to open our eyes to the possibility that there would also be infinite Issac Newton's, Aristotle's, and so on. There isn't any particular reason to restrict AGI to just Einsteins. Things might seem to get out of hand. All these infinite geniuses become mired in disagreements across the board. Who are we to believe? Maybe the Einsteins are convinced by some other personas that up is down and down is up. Endless arguments could consume tons of precious computing cycles. We must also acknowledge that evil doers of historical note could also be part of the infinite series. There could be infinite Genghis Khan's, Joseph Stalin's, and the like. Might they undercut the infinite Einsteins? Efforts to try and ensure that AI aligns with contemporary human values is a vital consideration and numerous pathways are currently being explored, see my discussion at the link here. The hope is that we can stave off the infinite evildoers within AGI. Einstein had grave concerns about the use of atomic weapons. It was his handiwork that aided in the development of the atomic bomb. He found himself mired in concern at what he had helped bring to fruition. An AGI with infinite Einsteins might discover the most wonderful of new inventions. The odds are that those discoveries could be used for the good of humankind or to harm humankind. It is a quandary whether we want those infinite Einsteins widely and in an unfettered way to share what they uncover. Here's the deal. Would we restrict access to the infinite Einsteins so that evildoers could not use the capacity to devise destructive possibilities? That's a lot harder to say than it is to put into implementation. For my coverage of the challenges facing AI safety and security, see the link here. Governments would certainly jockey to use the infinite Einsteins for purposes of gaining geo-political power. A nation that wanted to get on the map as a superpower could readily launch into the top sphere by having the infinite Einsteins provide them with a discovery that they alone would be aware of and exploit. The national and international ramifications would be of great consequence, see my discussion at the link here. I promised at the start of this discussion to eventually bring artificial superintelligence into the matter at hand. The reason that ASI deserves a carve-out is that anything we have to say about ASI is purely blue sky. AGI is at least based on exhibiting intelligence of the kind that we already know and see. True ASI is something that extends beyond our mental reach since it is superintelligence. Let's assume that ASI would not only imbue infinite Einsteins, but it would also go far beyond Einstein-level thinking to super Einstein thresholds. With AGI we might have a solid chance of controlling the infinite Einsteins. Maybe. In the case of ASI, all bets are off. The ASI would be able to run circles around us. Whatever the ASI decides to do with the infinite super Einsteins is likely beyond our control. Congrats, you've now been introduced to the infinite Einsteins conjecture. Let's end for now with a famous quote from Einstein. Einstein made this remark: 'Two things are infinite: the universe and human stupidity, and I'm not sure about the universe.' This highlights whether we could suitably harness an AGI containing infinite Einsteins depends upon human acumen and human stupidity. Hopefully, our better half prevails.
TechCrunch
17 hours ago
- TechCrunch
Week in Review: Why Anthropic cut access to Windsurf
Welcome back to Week in Review! Got lots for you today, including why Windsurf lost access to Claude, ChatGPT's new features, WWDC 2025, Elon Musk's fight with Donald Trump, and lots more. Have a great weekend! Duh: During an interview at TC Sessions: AI 2025, Anthropic's co-founder had a perfectly reasonable explanation for why the company cut access to Windsurf: 'I think it would be odd for us to be selling Claude to OpenAI,' Chief Science Officer Jared Kaplan said, referring to rumors and reports that OpenAI, its largest competitor, is acquiring the AI coding assistant. Seems like a good reason to me! Everything is the same: Chinese lab DeepSeek released an updated version of its R1 reasoning AI model last week that performs well on a number of math and coding benchmarks. Now some AI researchers are speculating that at least some of the source data it trained on came from Google's Gemini family of AI. WWDC 2025: Apple's annual developers conference starts Monday. Beyond a newly designed operating system, here's what we're expecting to see at this year's event, including a dedicated gaming app and updates to Mac, Watch, TV, and more. This is TechCrunch's Week in Review, where we recap the week's biggest news. Want this delivered as a newsletter to your inbox every Saturday? Sign up here. News Image Credits:Thomas Fuller / SOPA Images / LightRocket / Getty Images Business in the front: ChatGPT is getting new features for business users, including connectors for Dropbox, Box, SharePoint, OneDrive, and Google Drive. This would let ChatGPT look for information across your own services to answer questions. Oh no: Indian grocery delivery startup KiranaPro was hacked, and all of its data was wiped. According to the company, it has 55,000 customers, with 30,000 to 35,000 active buyers across 50 cities, who collectively place 2,000 orders daily. Techcrunch event Save $200+ on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Save $200+ on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Boston, MA | REGISTER NOW Artsy people, rejoice! Photoshop is now coming to Android, so users of Google's operating system can gussy up their images, too. The app has a similar set of editing tools as the desktop version, including layering and masking. Let's try that again: Tesla filed new trademark applications for 'Tesla Robotaxi' after previous attempts to trademark the terms 'Robotaxi' and 'Cybercab' failed. Rolling in dough: Tech startup Anduril just picked up a $1 billion investment as part of a new $2.5 billion raise led by Founders Fund, which means Anduril has doubled its valuation to $30.5 billion. On the road again: When Toma's founders realized car dealerships were drowning in missed calls, they hit the road to see the problem firsthand. That summer road trip turned into a $17 million a16z-backed fundraise that helped Toma get its AI phone agents into more than 100 dealerships across the U.S. Fighting season: All gloves were off on Thursday as Elon Musk and President Trump took to their respective social networks to throw jabs at each other. Though it might be exciting to watch rich men squabble in public, the fallout between the world's richest person and a sitting U.S. president promises to have broader implications for the tech industry. Analysis Image Credits:BlackJack3D / Getty Images Money talks: Whether you use AI as a friend, a therapist, or even a girlfriend, chatbots are trained to keep you talking. For Big Tech companies, it's never been more competitive to attract users to their chatbot platforms — and keep them there.
