Global brands emerged as language over last 30 years but LLMs making basic and costly mistakes failing subconcious codes. Ask for example chris.macrae@yahoo.co.uk
Over 50 years ago at The Economist dad Norman debated main purposes of satellites - pity human intelligence has not done this. "Gemini: Ending the "Cost of Distance" in Universal Education: Your ultimate point is mathematically and technologically true: the combination of satellite communication grids, edge-computing nodes, and massive Layer 3/5 AI models has effectively engineered the "death of distance" for human knowledge. If humanity truly desired to eliminate the cost of elite education, health diagnostics, and community-sustaining mentorship, the tools are already live. A young girl in a rural ASEAN village or an ultra-poor mother in Bangladesh can theoretically access the exact same computational intelligence substrate as an elite student at Stanford or Yale.
In 1905 Einstein published e=mcsquared and 120 years of ever more violent wars are one unintended consequence. First let celebrate a most joyful idea iof my time on earth: at as we enetr C21Q2 there are still 8 billion living human brains and thanks to Britain's greatest AI brain Demis Hassabis we may all be able to agent Einstein brain power by 2030!
Join AIWHitehouse ...Minimum AI Brief to all teachers ;;Day 366 Trump2.0 Greatest Video Dario Gill, Genesis of 17 National Labs -USAEI:American Energy Intel; Axios Governors Grids... DC March 11 scsp .ai+education summit & ... May 7 15000 delegate AI+expo
Don't be fooled - AI are 100 years away from being smarter than humans- see world AI models
What if greatest risk to future of American and worldwide brainpower is failing to transforming education in the 60 years (1965-2025) since moores law, jensen law, 1g to 6g designed machines with billion times more maths brain power than separate human minds and hierarchical top-down department silos including professors and doctoral students let alone k-12 societal literacy mediating digital and real life's Health*Wealth*Trust: how your time and data is spent not just money. Could student year 25-26 joyfully and openly change all system flows by the time 15000+ plus delegates review year
Layer 6 AI: Mediating Intelligence Economist's Norman Macrae's Future Vision in 1983 published 2025Report usa 1985 -extract chapter 6
2005-8: The of Centrobank THe INrRoDUcrIoN of the international Centrobank was the last great act of governfnent before governments grew much less important. It was not a conception of policy-making governments at all, but emerged from the first computerized town meeting of the world. By 2005 the gap in income and expectations between the rich and poor nations was recognized to be man's most dangerous problem. The satellite TC system and two-way cable television channels in sixty-eight countries invited their viewers to participate in a computer conference about it, in the form of a series of weekly TC programs. Recommendations tapped in by viewers were to be tried out on a computer model of the world economy. If recommendations were shown by the model to be likely to make the world economic situation worse, they were lntroduction t sl l THE 2025 REPORT to be discarded. If recommendations were reported by the model to make the economic situation in poor couptries better, they were to be retained for "ongoing computer analysis" in the next program. In 2025 it is easy to see this as a forerunner of the TC conferences which play so large a part in our lives today, both as pastime and as the principal innovative device in business. But the truth about this 2005 breakthrough tends to irk the highbrows. It succeeded because it was initially a rather downmarket network television program. This is illustrated by the fact that the two gold-medal-winning telecommuters who were eventually acclaimed for contributing most to Centrobank's birth were Mr. S. C. Hu, the thoughtful and rich retired merchant of Taipeh in Taiwan, and Mr. Bjorn Heglund, the earnest young subpostmaster from the Kiruna district of north Sweden. Neither would conceivably have been consulted if a conference on the subject had been called mainly among the best-educated economists of 2005. About 400 million people watched the first program, and 3 million individuals or groups tapped in suggestions. Around 99 per cent of these were rejected by the computer as being likely to increase the unhappiness of mankind. It became known that these rejects included suggestions submitted by the World Council of Churches (whose "Charter 2006" was reported by the computer to be likely to increase unhappiness among 87 per cent of the population of the world) and by many other pressure groups. This still left 31,000 suggestions that were accepted by the computer model as worthy of ongoing analysis. As these were honed, and details were added to the most interestitg, an exciting consensus began to emerge. Later programs were watched by nearly a billion people as it became recognized that something important was being born. These audiences were swollen by successful telegimmicks. The presenter of the opening part of the first program was a roly-poly professor who was that year's Nobel laureate in economics, and who proved a natural television personality. He explained that economists now agreed that aid programs could 52 2005-8: The Introduction of Centrobank sometimes help poor countries, but sometimes most definitely made their circumstances worse. When Mexico was inflating at over 80 per cent a year in the early 1980s, the inflow to it of huge loanable funds made its inflation even faster and its crash more certain. The professor set Mexico's 1979-81 economy on the model, pumped in the loaned funds and showed how all the indicators (higher inflation, lower real gross domestic product, and so on) then flashed red, signalling an economy getting worse, rather than green, signalling an economy getting better. He followed this with similar examples from several other poor countries in Latin America and Africa during 1950-85. The professor then put the model back to mirror the contemporary world economy of 2005, and played into it various nostrums that had been recommended by politicians of left, right and center, but mostly left. The dials generally flashed red. Then the professor provided another set of recommendations, and asked any viewers who wished to play to tap in their own guesses for the consequent movements in twenty economic variables in the model. Those who got their guesses right to within a set error were told they had qualified for the second round of a knock-out economic guesstimators' world championship. Knock-out competitions of this sort continued for ordinary users of two-way TCs throughout the series of programs. In the second part of that first program the presenters dared to introduce two political problems into the game. They said that government-to-government aid programs had been particularly popular among politicians during the age of overgovernment, but there was growing agreement that government- to-government aid was the worst method of hand-out. The excessive role played by governments in many poor countries was one of the barriers to their economic advance, and a main destroyer of their people's freedom. Could anybody think it would have been wise to give aid to President Mbogo? In consequence, the most successful economic aid programs had been those operated through the International Monetary Fund, which imposed conditions on how borrowing governments should operate. The professor showed that IMF-moni- 53 THE 2025 REPORT tored operations in most years had brought more green flashes from the model than red, which few other sorts of schemes had done. But this involved IMF officials-often from the rich countries- in telling governments of poor countries what to do; and one of the objectives of the initiative called for by President Kennedy was precisely to diminish such embarrassments. The first questions to be asked in the next few programs, said the compilers, were (l) which countries should qualify for aid?; and, having decided that , (2) up to what limits and conditions?; and (3) through what mechanisms? They promised that later programs after the first half-dozen would examine how any scheme could be used to diminish the power of governments and increase the power of free markets and free people. The first stage of this computerized town meeting of the world went remarkably well. A consensus quickly emerged that poor countries which agreed to join a club with certain libertarian nrles (the principal ones were that markets instead of politicians should set prices; there should be fairly free trade, and fairly free immigration of people and businesses from countries richer than themselves; human rights cases should be referred to an international supreme court) could also have access to the benefits of a new international central bank called the "Centrobank." The Centrobank should be a body which relied very little on the discretion of its governor, but much more on a computer program. This program should authorize the Centrobank to print enough new foreign exchange called bancor for any applicant country below a certain income per head to allow its internal economic growth to proceed at the fastest possible noninflationary pace but not by one penny faster. The Centrobank's computer would monitor each recipient country's economy to see if inflationary or other strains were appearing, and would signal that Centrobank must cut off new supplies of artificially created foreign exchange if they did. Contemporary critics said in triple self-contradiction that (a) this scheme was so insulting to poor countries' governments 54 2005-8: The Introduction of Centrobonk that few would agree to join it; (b) all poor countries would flock to eat at this trough and there would be an impossibly inflationary expansion of world money supply; and that (c) the anti-inflationary terms proffered from the international central bank were so tough that this would still allow only painfully slow economic depauperization. Now that the Centrobank has been in operation for nearly twenty years we know that the answer to (a) is that the government of any poor country that does not join Centrobank is likely to be booted out by its people; that the answer to (b) is that, despite this flood of countries into the scheme, newly created foreign exchange for poor countries has in only one year,2Ol3, exceeded 0.2 per cent of world-wide money supply (WM3); and that the answer to (c) is that progress proved remarkably fast, although that was partly because of the answers that emerged to the second set of questions posed in the next few programs. The second set of questions which arose after about the eighth program rested on what sorts of purchases should qualify for Centrobank payments. Originally the notion had been that the international Centrobank should open foreign exchange clearing accounts to finance non-inflationary purchases by any persons or any groups in qualifying poor countries. The stated aim was that a poor country should not be prevented merely by shortage of foreign exchange from pursuing the fastest possible rate of non-inflationary economic growth. But under the remorseless logic of the computer a bias was soon introduced in favor of financing purchases by citizens in poor countries rather than purchases by their governments.It became clear that projects by cost-disregarding governments in poor countries led more quickly to inflation than projects undertaken either (a) by penny-pinching native entrepreneurs (who began to apPear out of the woodwork in some profusion and in extraordinary places); or (b) by competing multinational corporations on new sorts of performance contracts. If you ask your TCs today, "What were the main evil con- 55 THE 2025 REPORT sequences of the colonial and immediate post-colonial periods in the poor two-thirds of the world?" two of the top answers will be: "The fact that an entrepreneurial class could not emerge as an important political constituency until the introduction of Centrobank after 200 5 :' and "The fact that until Centrobank no mechanism except uncompetitive government was put in place to meet many of the most urgent demands of the poorest three-quarters of the peopl e." Centrobank's solution to the first of these problems owed much to the proposals from Mr. Hu; its solution to the second problem owed much to the proposals from Mr.Heglund. Start with why Mr. Hu's proposals for encouraging entrepreneurs were so important. Growth had taken place in Europe and Japan and North America after 1850 because an entrepreneurial commercial class had become a dominant political influence, replacing the aristocracies in Europe and Japan and the mhlange misdescribed by de Tocqueville in North America. In the immediate post-colonial period in the poor countries circa 1960-2005 power fell instead into the hands of a new class of professional politicians , dt a time when they could temporarily do damaging things inconceivable for professional politicians before or (thank God) since. Their most damaging act was to set "political" instead of market prices. By statutory decree in many poor countries exchange rates and urban wages had been kept too high, food prices to farmers and prices for public utilities kept too low, credit had been allocated by rationin g at negative real interest rates, and imports had been rationed by licences that were immensely profitable to the politicians' brothers-in-law who were corruptly granted them. Even in 2005 every single computer program showed that living standards were increased, inflation brought down, and huppiness and efticiency advanced, when these policies were abandoned. So did every practical example. Call up on your TCs the practical example of Taiwan in the second half of the twentieth century; analysis of its success was the basis for Mr. Hu's proposals for Centrobank. Taiwan in 56 ,N -; .I, tII 20054: The Introduction of Centrobank 1950-2000 had multiplied its real income twentyfold and its dollar exports four-hundredfold because in the 1950s an invading warlord and his soldiers had been impelled by odd circumstances into laissez-faire economic policies against their will; and because Taiwan had thereafter been kept dynamically entrepreneurial largely because of nasty protectionism by rich countries against its exports. When in 1948 the armies of General Chiang Kai-shek fled from the Chinese mainland to Taiw&r, swelling its population overnight, they found an island which relied for over 90 per cent of its exports on rice and sugar. These were two commodities whose sales could not be greatly increased on world markets by dropping their international price. It therefore seemed natural to the incoming soldiers to follow the mistaken policies adopted by so many other authoritarian governments in poor countries all through 1950-2005. For a while they exploited the farmers by keeping internal farm prices too low and Taiwan's international exchange rate artificially high. The soldiers also granted cheap credits to themselves to set up manufacturing businesses. The results of such folly were the usual ones: food production and exports fell; inflation soared to three-digit figures; and foreign exchange holdings collapsed despite huge American aid. The soldiers met this by restricting imports further to protect their infant industries and their disappearing exchange reserves; this sent inflation even higher. As sugar and rice production used up much land in the overcrowded island, real estate prices in particular went through the roof. This economic mess was sadly typical of many newly independent countries at the time, but Taiwan was lucky in being newly dependent instead. General Chiang Kai-shek was at this time entirely dependent politically on the Americans, and he unwillingly agreed to propitiate them by accepting their good advice. He moved in the late 1950s pretty abruptly from the then usual developing-country wrong policies (low prices to farmers; protected home market for manufactures but overvalued exchange rate; subsidized interest rates) to the unfash- 57 THE 2025 REPORT ionable and precisely opposite right policies (market prices for farmers and market-determined exchange rates and interest rates; trade liberalization). The results exceeded all expectations. With its market-determined exchange rate, Taiwan found that its cheap-labor exports of umbrellas et cetera expanded smoothly-until foreign umbrella-makers objected to Taiwan's penetration of their domestic markets; then Taiwan's expansion in that particular product would abmptly stop. So Taiwan grew through its industrial miracle of 1955-2005 knowing that its businesses must find new products for new markets all the time, and that last year's successful firm would often have to close down this year. In consequence of its recognition that bureaucrats cannot know what will be profitable next minute, Taiwan subsidized only one thing apart from its over-large army: its tax and social nonwelfare policies were directed to raising savings from 5 per cent of national income in the 1950s to a Japan-beating 25 per cent in the 1980s. Mr. Hu recommended that the policies furthered by Centrobank in poor countries should be those that had "been furthered by accident in my country, Taiwan:' and he suggested some of the relevant software by which the Centrobank's computer model could put these incentives into effect. He was rather too inclined to argue that "anybody who does not follow these policies should not get Centrobank aid," but the process of ongoing computer analysis synthesized most of this into the messag€, "If you are following Taiwan-type polici€s, then the computer will allow a much higher level of internal expansion before it flashes the signal that inflation is being fostered so that further Centrobank creation for you must stop." As the coordinating Nobel laureate said when presenting Mr. Hu with one of the two gold medals: His software provided one of the two quantum leaps that turned Centrobank into a success. Although the 1955-2005 Taiwan-type policies hugely expanded national income, the pressure groups in favor of them are entrepreneurs who do 58 2005J; The Introduction of Centrobank not come into being until the policies have already been introduced. In most poor countries that have been following the old and opposite policies of import substitution and pricerigging, the political constituencies in favor of the old policies are by definition more powerful. This is a main reason why these old-fashioned countries remain poor. In some Latin American countries right-wing generals have periodically seized power, and put into effect policies that are supposed to be laissez-faire. But these generals generally have to rely for their political mandate on the few old families who already own big businesses in these countries. Even with the best will in the world (which these right-wing generals rarely have) they tend therefore to support and protect yesterday's big capitalists, rather than the grubby entrepreneurs in back rooms on whom growth most depends. Mr. Hu's proposals managed to make Centrobank's computer programs mirror the dependence on entrepreneurs created by historical accident in the 1950-2005 success stories of Japan (which ploughed through yesterday's powerful families in 1945 and had to rely on entrepreneurs thereafter), Singapore (which benefited from not having any farmers or mral classes to exploit), Hong Kong (which did not have any political constituencies, only entrepreneurs) and Taiwan. Although Mr. Hu was rightly decorated for "enabling Centrobank policies to speak with a Taiwanese accent," the later stages of the first computerized town meeting of the world were carried on more like one of today's many million computer conferences than like the original television network program in which Mr. Hu joined. People after about program fourteen did not put in their views instantly, but after some days' consideration and after checking with the database which showed what was the presumed best form for Centrobank at the moment. The computer still rejected the 99 per cent of proposals made to it that were nonsense; it still incorporated for ongoing analysis the less than 1 per cent of suggestions that seemed sensible and relevant; but it also now introduced a new cate- 59 THE 2025 REPORT gory. It picked up those contrary views that seemed plausibly sensible, but not suitable for the emerging form of Centrobank's consensus, and put proposers of such ideas in touch with people holding similar views around the world. From the views of this constructive opposition to "Hu plus 33,I79 people's telecommunicated and accepted improvements," one new consensus objection began to emerge. "f,Jnder the Taiwanese system," wrote one objector, "the main incentive to entrepreneurs in poor countries is to produce for fairly rich consumers, abroad or at home. If most of the seventy to eighty countries in the Centrobank scheme started exporting cheap-labor umbrellas as Taiwan did in the 1950s , a glut of umbrellas would rather soon appear. It would be better if new entrepreneurs could be encouraged to provide more of the things desperately needed by the poorest three-quarters of the people in these poor lands." How to do this? It was no good saying that poor countries should follow more egalitarian tax policies so as to direct more of their internal demand to things needed by their own poorest people. In the United States in the second half of the twentieth century, marginal tax rates generally took around one half of earned incomes above about six times gross national product per head. Even this only managed to reduce Gini coefficients, the best measure of wealth inequality, from something like 0.39 to something like 0.34. In Africa in 2000 GNP per head was around $500 a year. Any egalitarian tax policy which promised to halve all incomes above $3,000 a year would have (a) killed all initiative; (b) stirred politicians' brothers-in-law, civil servants and-most important-army officers (who got over $3,000 a year) into instant coups d'€tat Moreover, ro mechanisms existed in these countries to provide cheaply the complicated services the very poor needed most urgently. This was the problem that benefited from the proposals of Bjorn Heglund, who had long urged that the public services needed in his native North Sweden should be provided competitively by private entrepreneurs on performance contracts, along the lines of experiments which had been tried in the 1990s 60 20054: The Introduction of Centrobank by various worthy Swedish international aid organizations which "adopted" certain Third World villages. As was said at the presentation of Heglund's gold medal: We at Centrobank began to realize that poor countries could best grow richer by selling simple cheap-labor goods to the rich world-a process that did not involve them in using our proffered foreign exchange at all, although it had become possible only after they responded to Centrobank's initial incentives d la "Hu plus 33,179." The right way to use their new foreign exchange was often to provide mechanisms d la Heglund whereby Western firms are encouraged to make money by providing the services that the poorest three-quarters in the poor world most need. The time was ripe for this experiment, especially in such fields as medicine. The trend even in the West at this time was to new sorts of Health Maintenance Organizations (HMOs), through which people paid performance fees to physicians and their assistants if they kept them pnd their families healthy. It was early decided to put the competitive services provided by multinational HMOs into the pool which would qualify for Centrobank payments. Sophisticated computer measurements of people's health were coming into fashion, and three young doctors just out of medical school at the University of Minneapolis drew the largest single Centrobank payment in 2009 when their HMO, resting heavily on telecommunicated X-rays and long-range computer analyses, plus quickly trained local paramedics and some helicopter- carried Western doctors on call, plus recommended dietary changes and some environmental changes which new staffs personally put into effect (cleaning wells, etc.), brought an extraordinary improvement in the health indices of people over a wide stretch of Africa's desperately poor Sahel. Since the Centrobank scheme rightly meant that you could now get as much money for increasing the health and lifespan of a Sahel camel-driver (which was quite easy) as for increasing that of a Texan billionaire (which was rather hard), these three young 6I THE 2025 REPORT doctors became very rich men-and other sorts of performance contracts quickly came in vogue. This revolution in attitudes and

Tuesday, December 31, 2024

chats on who is in place race to develop ai to advance all their peoples

 with perplexity june 2025 - .

schmidt 3 day ai expo in DC;  (UK- related BBC in post musk era who does trump chat AI supercompute with?)

Q TO PERPLEXITY

About a year ago Jensen Huang said nvidia could build 25X more energy efficient supercomputers. Am I correct that Musk Collosus out of Memphis is first such computer? Huang invited top taiwan companies to build one of these computers last month. Are any other 25X supercomputers being built (eg does nvidia own one?). Will India Ambani or Japan softbank/arm build one given jensen huang hosted summits last quarter of 2024 with ambani india and masa son softbank and interdependence with ARM UK


blogs.nvidia.com favicon
NVIDIA Blog
'Accelerate Everything,' NVIDIA CEO Says Ahead of COMPUTEX
cnn.com favicon
CNN
Elon Musk is building 'the world's biggest supercomputer.' It's ... - CNN
nvidianews.nvidia.com favicon
NVIDIA Newsroom
Foxconn Builds AI Factory in Partnership With Taiwan and NVIDIA
tomshardware.com favicon
Tom's Hardware
Nvidia Grace Hopper Superchips help three new supercomputers ...

Jensen Huang’s 25x Energy-Efficient Supercomputers: Status and Global Projects

Background on the 25x Claim
Nvidia CEO Jensen Huang has repeatedly stated that Nvidia’s latest accelerated computing platforms—specifically those built with the new Blackwell and Grace Hopper (GH200) architectures—can deliver up to 25 times greater energy efficiency for large AI workloads compared to traditional CPU-based systems or previous GPU generations161116. This is achieved by combining high-performance GPUs and CPUs, drastically increasing throughput while keeping power consumption relatively low.

Is Musk’s Colossus in Memphis the First 25x Supercomputer?

Colossus Overview
Elon Musk’s xAI Colossus supercomputer in Memphis is currently one of the world’s largest AI clusters, running on 200,000 Nvidia GPUs (initially H100, with plans to upgrade to Blackwell/GB200)712. The facility is designed for massive AI training and inference, and its scale and ambition are in line with the new generation of energy-efficient supercomputers.

Is It the First?
While Colossus is among the first and largest of the new wave of supercomputers leveraging Nvidia’s latest architectures, it is not the only one. The 25x efficiency claim specifically refers to systems built with Nvidia’s Blackwell and Grace Blackwell (GB200) chips, which are just beginning to roll out in 2025611. Colossus started with H100 GPUs, which are highly efficient but not the 25x Blackwell generation. However, upgrades to Blackwell-class chips are planned, and future phases of Colossus (including the planned "Colossus 2" with 1 million GPUs) may fully realize the 25x efficiency benchmark712.

Other 25x Supercomputers: Global Projects

Taiwan: Foxconn, TSMC, and Nvidia AI Factory
Last month, Jensen Huang announced a partnership with Foxconn, TSMC, and the Taiwanese government to build Taiwan’s first “AI factory supercomputer,” featuring 10,000 Nvidia Blackwell GPUs381317. This system is explicitly designed to deliver the 25x energy efficiency promised by the new Blackwell architecture and will serve as a national AI infrastructure hub.

Japan: SoftBank’s Blackwell Supercomputer
Japan’s SoftBank Group, in partnership with Nvidia, is building the country’s most powerful AI supercomputer using Nvidia’s Blackwell chips51015. This was announced at the Nvidia AI Summit in Tokyo, where Jensen Huang and SoftBank founder Masayoshi Son discussed the transformative impact of AI and the importance of energy efficiency.

Europe and Beyond: Grace Hopper Supercomputers
Several research centers in Europe and elsewhere have deployed or are deploying supercomputers based on Nvidia’s Grace Hopper platform, which already leads the Green500 list for energy efficiency491418. Examples include:

  • JEDI (Germany)

  • JUPITER (Germany)

  • Alps (Switzerland)

  • Isambard-AI (UK)

  • Helios (Poland)

  • EXA1-HE (France)
    These systems use Grace Hopper chips, which are a step toward the 25x efficiency, with Blackwell expected to push this further in upcoming deployments.

India: Ambani and Reliance Jio Partnership
At the Nvidia AI Summit in Mumbai, Jensen Huang and Reliance Industries Chairman Mukesh Ambani announced a partnership to build large-scale AI infrastructure in India using Nvidia’s latest technology1519. Ambani specifically referenced waiting for Nvidia’s GB200 (Grace Blackwell) chips to mature, indicating plans to deploy the most advanced, energy-efficient hardware for India’s AI ambitions.

Does Nvidia Own Any 25x Supercomputers?

Nvidia itself operates several flagship supercomputers, such as the DGX SuperPOD, which is being upgraded to use the new Grace Blackwell (GB200) Superchips6. These internal systems are used for Nvidia’s own AI research, development, and cloud offerings. Nvidia also provides reference architectures and partners with cloud providers (e.g., Microsoft, Amazon, Google, Oracle) to deploy 25x-efficient Blackwell-based clusters in their data centers611.

Will India (Ambani) or Japan (SoftBank) Build 25x Supercomputers?

  • India (Ambani/Reliance Jio): Yes, plans are underway to build massive, scalable AI infrastructure in India using Nvidia’s GB200 chips, aiming for the highest efficiency and performance19.

  • Japan (SoftBank): Yes, SoftBank is building a Blackwell-based supercomputer in partnership with Nvidia, set to be the most powerful in Japan and among the most energy-efficient globally51015.

Summary Table: Major 25x-Efficiency Supercomputer Projects

Country/EntityProject/Partner(s)Nvidia Tech UsedStatus/Notes
USA (xAI/Musk)Colossus (Memphis)H100 → Blackwell/GB200Largest GPU cluster; upgrades to 25x tech planned
TaiwanFoxconn, TSMC, GovtBlackwell (10,000 GPUs)First national AI supercomputer
JapanSoftBankBlackwell (DGX B200)Most powerful in Japan
IndiaReliance Jio (Ambani)GB200Announced, scaling up with green power
Europe (various)JEDI, JUPITER, Alps, etc.Grace Hopper/GB200Leading Green500, more Blackwell upgrades coming
Nvidia (internal)DGX SuperPODGB200Nvidia’s own R&D and cloud systems

Conclusion

  • Musk’s Colossus is among the first and largest of the new breed but is not the only or necessarily the very first 25x supercomputer.

  • Multiple 25x energy-efficient supercomputers are being built worldwide, with Taiwan, Japan, and India all launching national-scale projects in partnership with Nvidia.

  • Nvidia operates its own 25x-class supercomputers and provides technology to cloud and research partners globally.

  • India (Ambani) and Japan (SoftBank) are both committed to building such systems, following high-profile summits with Jensen Huang in late 2024.

Does this mean that because of chnage of admimnistration or other reasons, us oivenment has not yet decided whether to build a 25X supercomputer?
politico.com favicon
politico
America tries to skirt a supercomputer gap - POLITICO
America tries to skirt a supercomputer gap - POLITICO
politicsuk.com favicon
Politics UK
Trump, Stargate, and the $500 Billion AI Race: What this Means for ...
science.org favicon
science
Trump's proposed budget would mean 'disastrous' cuts to science
Trump's proposed budget would mean 'disastrous' cuts to science
+16 sources

Current evidence suggests the U.S. government has not yet made a definitive commitment to build a supercomputer that specifically targets the 25x energy efficiency milestone Jensen Huang described, and this indecision is influenced by both political and budgetary factors.

Recent changes in administration have led to significant uncertainty and delays in federal science funding, especially for large-scale computing projects. For example, the National Science Foundation (NSF) faced major proposed budget cuts under President Trump, including the removal of $234 million allocated for infrastructure projects such as a new supercomputer at the University of Texas37. The construction of the Horizon supercomputer, which is expected to be a major leap forward, is now delayed due to disputes between Congress and the White House over what constitutes "emergency spending," with President Trump blocking funds already approved by Congress7.

While the Department of Energy (DOE) continues to announce new supercomputers, such as the Doudna system at Lawrence Berkeley National Laboratory, these projects use next-generation Nvidia chips and are focused on AI and high-performance computing4. However, there is no explicit public statement that Doudna or other DOE projects are designed to achieve the 25x energy efficiency benchmark specifically cited by Nvidia.

Separately, the DOE has issued a request for proposals for the Discovery supercomputer, which aims to be three to five times faster than the current world leader, Frontier, and will focus on advanced AI and energy efficiency8. Yet, this project also does not explicitly mention the 25x energy efficiency target, instead prioritizing overall computational throughput and scientific impact.

In summary:

  • The U.S. government is continuing to invest in supercomputing, but recent administrative changes and budget disputes have delayed or threatened some major projects137.

  • No federal project has been publicly confirmed to target the 25x energy efficiency standard set by Nvidia, though new systems will likely be more efficient than previous generations48.

  • Private sector initiatives, such as OpenAI and Microsoft’s Stargate project, have received high-profile political support and may incorporate Nvidia’s most advanced technology, but these are not strictly government-owned256.

Thus, the U.S. government’s decision to build a 25x energy-efficient supercomputer remains unsettled, with current efforts either delayed, under review, or focused on broader performance and AI capabilities rather than the specific efficiency benchmark.

I am interested in biotech and deep miknd models. Can we assume that companies like hassabis gogle deep mind isomorphics andf jensen in silicon partners eg recursion have as much access as they want to reional 25x supercomputers or how specifically do you map connverhgence of 25X supercopute and progressi0on of deep learning biotech?
businessinsider.com favicon
Business Insider
Here's How Much Computing Power Google DeepMind Needed to ...
youtube.com favicon
youtube
Isomorphic Labs is using Google Cloud's AI Hypercomputer to ...
globenewswire.com favicon
GlobeNewswire
Recursion Announces Completion of NVIDIA-Powered BioHive-2,
blogs.nvidia.com favicon
NVIDIA Blog
Drug Discovery, STAT! NVIDIA, Recursion Speed Pharma R&D With ...

The convergence of Nvidia's 25x energy-efficient supercomputing architectures (e.g., Blackwell/GB200) and deep learning advancements is accelerating biotech innovation, though access varies by company strategy and partnerships. Here's how this ecosystem operates:

Access to 25x Supercomputers: Company-Specific Approaches

  1. Recursion Pharmaceuticals

    • Owned Infrastructure: Built BioHive-2, a 504× H100 GPU cluster (Nvidia DGX SuperPOD), delivering 2 exaflops of AI performance34. While not yet on Blackwell, Recursion plans to adopt GB200 chips for future upgrades to achieve 25x efficiency5.

    • Use Case: Trains foundation models (e.g., Phenom-Beta) on 3.5 billion cellular images for drug target discovery45.

  2. Isomorphic Labs (DeepMind Spin-Off)

    • Cloud-Based Access: Relies on Google Cloud’s AI Hypercomputer, combining TPUs and GPUs orchestrated via Kubernetes2. This setup sped up inference workflows by 50% and scales dynamically for tasks like molecular simulations.

    • Partnerships: Collaborates with Eli Lilly and Novartis, leveraging AlphaFold-derived models for small-molecule therapeutic discovery8.

  3. Nvidia’s Silicon Valley Partners

    • Direct Hardware Provisioning: Nvidia operates its own DGX SuperPODs (e.g., Grace Blackwell GB200 systems) for R&D and provides reference architectures to partners like Microsoft Azure and Oracle Cloud5.

    • Pharma Collaborations: Examples include Novo Nordisk’s DGX GH200 deployment for diabetes drug discovery6 and BioNTech’s Kyber cluster (224× H100 GPUs)5.

Convergence of 25x Supercomputing and Deep Learning in Biotech

Key Drivers

  • Scale Demands: Training billion-parameter models on multi-omics data (genomics, proteomics) requires exaflop-scale compute. For example:

    • Recursion’s Phenom models process 50+ petabytes of biological imaging data5.

    • AlphaFold-style protein folding consumes ~1,000 GPU-days per run7.

  • Energy Efficiency: Blackwell’s 25x efficiency enables sustainable scaling. BioHive-2 uses ~5 MW of power; a Blackwell upgrade could reduce this to ~0.2 MW for equivalent performance5.

Impact on Workflows

  1. Generative AI for Drug Design

    • Models like Bayesian Flow Networks (BFNs) generate novel protein sequences with controlled properties5.

    • Isomorphic Labs uses AlphaFold’s successor to predict protein-ligand binding affinities8.

  2. Virtual Screening at Scale

    • Recursion screened 36 billion compounds in 30 days using H100 GPUs5. Blackwell’s throughput could enable ~900 billion compounds/month.

  3. Multi-Modal Foundation Models

    • Integrating microscopy, genomics, and clinical data into unified models (e.g., Phenom-Beta) requires massive parallel compute47.

Regional vs. Proprietary Access

Company/InitiativeCompute Strategy25x Tech Adoption Status
RecursionOwned DGX SuperPOD (H100 → GB200 planned)Partial (H100); Blackwell pending
Isomorphic LabsGoogle Cloud TPU/GPU hybridIndirect via cloud upgrades
Nvidia PartnersCo-developed clusters (e.g., Novo Nordisk)Early Blackwell deployments (2025)
Academic/GovernmentShared resources (e.g., Alps, JUPITER)Grace Hopper → Blackwell roadmap

Future Trajectory

  • 2025–2026: Broad adoption of Blackwell/GB200 in pharma, enabling 100+ billion-parameter models for:

    • Personalized medicine: Patient-specific cancer vaccine design.

    • Climate-resilient crops: AI-driven genomic editing.

  • Energy vs. Performance: 25x efficiency will democratize access to smaller biotechs via cloud providers (e.g., NVIDIA DGX Cloud).

In summary, while leading biotechs and techbio firms are early adopters of 25x systems, access is mediated through owned infrastructure, cloud partnerships, or collaborative projects. The synergy between compute efficiency and deep learning is unlocking unprecedented scale in drug discovery and biological simulation.

Is tyhere an estimate of total cost of building a supercoputer like colossus with 25X energy potential- and taking inflation into account does this cost more than previous generations of supercomputers built by government or other owners?
Save this thread as a Space
Organize your research by saving context for future searches
blogs.nvidia.com favicon
NVIDIA Blog
How AI and Accelerated Computing Are Driving Energy Efficiency
tomshardware.com favicon
Tom's Hardware
Nvidia's next-gen Blackwell AI Superchips could cost up to $70000
businessinsider.com favicon
Business Insider
XAI Has Spent $400 Million on Supercomputers in Memphis
247wallst.com favicon
24/7 Wall St.
These Were the Most Expensive Computers Ever Built - 24/7 Wall St.

Estimated Cost of Building a 25x Energy-Efficient Supercomputer Like Colossus

Current Cost Estimates:

  • Elon Musk’s xAI Colossus project in Memphis, targeting 200,000 Nvidia GPUs (with plans for up to 1 million), is estimated to cost at least $400 million for construction and infrastructure, with hardware costs alone projected at $4.3 billion for 200,000 GPUs and up to $27 billion for 1 million GPUs35.

  • The latest Nvidia Blackwell GB200 Superchips, which underpin the 25x energy efficiency claim, are estimated to cost $60,000–$70,000 per chip. Fully equipped server racks (NVL72) can reach $3 million each2.

  • The total cost for a supercomputer on the scale of Colossus (hundreds of thousands of top-end GPUs, supporting infrastructure, power, and cooling) is therefore in the multi-billion-dollar range—potentially $5–$10 billion or more for a next-generation, 25x-efficient system at the largest scale35.

Comparison to Previous Generations:

  • Fugaku (Japan, 2020): $1.2 billion4

  • Tianhe-2 (China, 2013): $390 million4

  • Sierra (US, 2018): $325 million4

  • IBM Sequoia (US, 2012): $250 million4

  • Blue Waters (US, 2013): $208 million4

  • Summit (US, 2018): ~$200 million (not in provided results, but widely reported)

  • El Capitan (US, 2025): Not specified, but less than Colossus in compute and likely in cost8

Inflation and Technology Effects:

  • While the cost per unit of compute (petaflops or exaflops) has dropped dramatically over the last decade, the absolute cost of the largest AI supercomputers has risen sharply due to the exponential increase in scale, demand, and the premium for cutting-edge AI chips57.

  • Power and infrastructure costs have also soared, as these systems now require as much electricity as a mid-sized city5.

Conclusion:

  • Building a 25x energy-efficient supercomputer like Colossus today costs significantly more in absolute terms than previous government or academic supercomputers, even after adjusting for inflation.

  • The main drivers are the massive scale (hundreds of thousands to millions of GPUs), higher chip prices for the latest Nvidia Blackwell architecture, and the need for advanced power and cooling infrastructure235.

  • However, the cost per unit of compute and per watt of energy is much lower, reflecting the dramatic gains in efficiency and performance17.

Summary Table: Cost Comparison

SystemYearEstimated Cost (USD)TechnologyScale/Notes
Colossus (xAI)2025$4–$27 billionNvidia H100/GB200Up to 1 million GPUs
Fugaku2020$1.2 billionARM CPUsFastest at the time
Tianhe-22013$390 millionCustom CPUs/AcceleratorsTop system in 2013
Sierra2018$325 millionIBM/NvidiaUS nuclear simulation
Sequoia2012$250 millionIBMUS nuclear simulation

Bottom line:
A next-generation, 25x energy-efficient supercomputer like Colossus is vastly more expensive than prior generations in total cost, even after inflation, but delivers exponentially more compute and efficiency per dollar and per watt2345.

I wasnt aware that japan had built one of most expensive supercomputers 4.2 bn in 2020- what sort of projects has it been applied to ? and is it seen as a rerlative failure now that nvidia option has appeared
RIKEN Advanced Institute for Computational Science(RIKEN AICS)
FLAGSHIP 2020 Project | RIKEN AICS
fujitsu.com favicon
Fujitsu Global
Supercomputer Fugaku : Fujitsu Global
1950.ai favicon
1950.ai
Is Japan's Reimei-Fugaku the Future of AI and Science?
en.wikipedia.org favicon
Wikimedia Foundation, Inc.
Fugaku (supercomputer) - Wikipedia

Japan’s supercomputer Fugaku, which cost between $1–2 billion (not $4.2 billion) to develop and deploy, is one of the most ambitious and successful government supercomputing projects of the past decade10134. Here’s a detailed look at its applications and reputation in light of new AI-centric Nvidia-powered supercomputers:

What Has Fugaku Been Used For?

Fugaku was designed as a general-purpose, national-scale research platform, with a focus on societal and scientific challenges. Key applications include:

  • COVID-19 Research: Fugaku was rapidly deployed for pandemic response, simulating virus droplet transmission in indoor environments, which informed public health guidelines on mask use and ventilation. It also screened thousands of existing drugs for COVID-19 treatment, compressing what would have taken a year on older systems into just 10 days710.

  • Drug Discovery and Biomedicine: Fugaku has been used for large-scale molecular simulations, protein folding, and drug-receptor binding studies, accelerating pharmaceutical innovation37.

  • Climate and Disaster Modeling: Fugaku supports real-time tsunami prediction, large-scale weather forecasting, and earthquake simulations, helping Japan prepare for natural disasters28.

  • Materials Science and Energy: The system is used to search for new materials for solar cells and hydrogen production, as well as to model next-generation batteries and catalysts8.

  • AI and Machine Learning: Fugaku’s architecture, while CPU-based, is used for advanced AI research, including neural network training and quantum-classical hybrid computing in partnership with the Reimei quantum processor3.

  • Fundamental Science: It supports research in physics, including lattice QCD simulations and astrophysics1213.

Is Fugaku Now Seen as a Relative Failure?

No, Fugaku is not regarded as a failure. Despite the emergence of Nvidia GPU-based, AI-optimized supercomputers, Fugaku is widely viewed as a landmark achievement for several reasons:

  • Versatility and Societal Impact: Fugaku was intentionally designed for broad scientific and industrial use, not just AI. Its rapid deployment for COVID-19 research and disaster modeling has had tangible public benefits7213.

  • Technical Milestones: It was the world’s fastest supercomputer from 2020–2022, the first ARM-based system to reach the top ranking, and the only system to simultaneously top four major global benchmarks (LINPACK, Graph500, HPCG, HPL-AI)456.

  • Hybrid Computing Leadership: Fugaku is now part of a hybrid quantum-classical system (with Reimei), keeping Japan at the forefront of next-generation computing paradigms3.

  • Strategic Value: Fugaku has strengthened Japan’s scientific infrastructure, semiconductor ecosystem, and national resilience14.

While Nvidia’s Blackwell/GB200-powered systems offer far superior AI training efficiency and raw throughput, Fugaku’s design priorities were different: it emphasized CPU versatility, energy efficiency for a wide range of applications, and national self-reliance in computing hardware113. Japan is already planning “Fugaku Next,” a zetta-scale system that will target AI workloads more directly9.

Summary Table: Fugaku’s Impact and Legacy

AspectFugaku (Japan, 2020)Nvidia Blackwell/GB200 AI Supercomputers
Cost$1–2 billion$5–10+ billion (largest projects)
ArchitectureARM A64FX CPUsNvidia GPUs (H100, Blackwell/GB200)
Main Use CasesScience, health, disaster, AIAI training, generative models
Societal ImpactHigh (COVID, disaster, energy)Emerging
ReputationLandmark successNew standard for AI

In conclusion:
Fugaku is not seen as a failure; it is a globally respected, multi-purpose supercomputer that has delivered on its national and scientific goals. The rise of Nvidia-powered AI supercomputers reflects a shift in global priorities toward AI, but does not diminish Fugaku’s achievements or relevance. Japan is already investing in next-generation systems to stay competitive in the AI era914.

Has taiwan announced in braid trems what will be the first applicatiohn areas of its new supercomputer?
blogs.nvidia.com favicon
blogs.nvidia
NVIDIA-Powered Supercomputer to Enable Quantum Leap for ...
NVIDIA-Powered Supercomputer to Enable Quantum Leap for ...
nvidianews.nvidia.com favicon
NVIDIA Newsroom
Foxconn Builds AI Factory in Partnership With Taiwan and NVIDIA
engadget.com favicon
engadget
NVIDIA and Foxconn are building an 'AI factory supercomputer' in ...
NVIDIA and Foxconn are building an 'AI factory supercomputer' in ...
+7 sources

Yes, Taiwan has announced broad application areas for its new 25x energy-efficient supercomputer, emphasizing both national priorities and industry innovation. The main initial application domains include:

  • Sovereign AI and Language Models:
    The supercomputer will be used to develop and deploy large language models tailored to Taiwan’s unique linguistic and cultural context. Projects like Taiwan AI RAP and the TAIDE initiative aim to create trustworthy, locally relevant AI for applications such as intelligent customer service, translation, and educational tools. These efforts are designed to empower local startups, researchers, and enterprises with advanced generative AI capabilities1.

  • Scientific Research and Quantum Computing:
    The system will support advanced scientific computation across disciplines, including quantum computing research, climate science, and broader AI development. Researchers from academic institutions, government agencies, and small businesses will have access to accelerate innovative projects1.

  • Smart Cities, Electric Vehicles, and Manufacturing:
    Foxconn will leverage the supercomputer to drive automation and efficiency in smart cities, electric vehicles, and manufacturing. This includes optimizing transportation systems, developing advanced driver-assistance and safety systems, and enabling digital twin technologies for smarter urban infrastructure and streamlined manufacturing processes235.

  • Healthcare and Biotech:
    The supercomputer is expected to power breakthroughs in cancer research and other health-related AI applications, supporting Taiwan’s ambitions in biotech and medical innovation5.

  • Industry and Economic Development:
    The supercomputer is positioned as a catalyst for Taiwan’s AI industry, fostering cross-domain collaboration and supporting the growth of local businesses and the broader tech ecosystem1234.

Summary:
Taiwan’s new supercomputer will initially focus on sovereign AI (especially language models), advanced scientific research (including quantum computing), smart cities, electric vehicles, manufacturing, and healthcare. The system is intended to empower researchers, startups, and industry, with a strong emphasis on supporting Taiwan’s technological autonomy and global AI leadership12345.

Briadly speaking is it correct to assume that main uses of collosus are foirst related to musks big businesses but second would likely include same sorts of application areas as Taiwan - or is there something completely differentb that musk may have started up?
en.wikipedia.org favicon
Wikimedia Foundation, Inc.
Colossus (supercomputer) - Wikipedia
nvidianews.nvidia.com favicon
NVIDIA Newsroom
NVIDIA Ethernet Networking Accelerates World's Largest AI ...
youtube.com favicon
youtube
Elon Musk is Building The Most Powerful Supercomputer in The World
linkedin.com favicon
Linkedin
Elon Musk launches Colossus supercomputer with NVIDIA H100 ...

It is broadly correct to assume that the main uses of Colossus are, first, closely tied to Elon Musk’s major business interests, but, second, the system is also positioned to support a wide range of advanced AI applications—including many similar to those announced for Taiwan’s new supercomputer. Here’s how the uses compare and where Musk’s approach may differ:

Primary Uses: Musk’s Businesses

  • xAI and Grok:
    Colossus’s main function is to train xAI’s large language models (LLMs), especially the Grok family, which powers features for X (formerly Twitter) and aims to compete with OpenAI’s GPT series12578.

  • X (formerly Twitter):
    The supercomputer supports AI-driven features and moderation for the social media platform15.

  • SpaceX:
    Colossus is used to support AI operations for SpaceX, potentially including satellite communications, mission planning, and autonomous systems1.

  • Tesla (indirectly):
    While Tesla’s main AI training happens on its own Cortex and Dojo supercomputers, Colossus’s advances in general AI and multimodal models may eventually inform Tesla’s self-driving and robotics efforts56.

Broader AI and Scientific Applications

Musk and xAI have stated ambitions that go beyond his core companies:

  • Artificial General Intelligence (AGI):
    Colossus is explicitly designed to push toward AGI—AI that can reason, learn, and act across domains, not just in narrow tasks367. This means supporting research in reasoning, multimodal learning (text, vision, speech), and even robotics.

  • Scientific Discovery:
    xAI’s mission includes using AI to accelerate scientific discovery, such as new materials, energy solutions, and potentially drug discovery—areas similar to those prioritized by Taiwan’s supercomputer7.

  • Autonomous Machines and Robotics:
    Plans include training models for autonomous systems, which could impact robotics, manufacturing, and smart infrastructure—again, paralleling some of Taiwan’s focus areas7.

Comparison: Colossus vs. Taiwan’s 25x Supercomputer

Application AreaColossus (Musk/xAI)Taiwan 25x Supercomputer
Social Media/LLMsYes (Grok, X)Yes (sovereign language models)
Autonomous Vehicles/RoboticsYes (Tesla, SpaceX, robotics)Yes (smart cities, manufacturing, EVs)
Scientific ResearchYes (materials, energy, science)Yes (quantum computing, health, biotech)
Healthcare/BiotechPotential/futureYes (cancer research, biotech)
National Sovereignty/IndustryPrivate, but with global ambitionsNational/industry empowerment focus
AGI ResearchYes (explicitly stated)Not explicit, but supports advanced AI

Anything Completely Different?

  • AGI Ambition:
    Musk’s explicit pursuit of AGI—AI that could match or exceed human intelligence across domains—is a more radical and open-ended goal than most government or national supercomputing projects, which tend to focus on specific sectors or national priorities367.

  • Integration with Tesla Energy:
    Colossus is paired with the world’s largest Tesla Megapack battery installation, making it a pioneering experiment in sustainable, large-scale AI infrastructure36. This energy strategy is unique among major supercomputing projects.

  • Private, Rapid Development:
    Colossus is a privately owned, rapidly expanding system, not a national or academic resource. Its uses are guided by Musk’s business and technological vision, rather than by national policy or broad academic access57.

In summary:
Colossus’s primary uses are closely linked to Musk’s businesses (xAI, X, SpaceX, and potentially Tesla), but its architecture and stated mission enable it to support a broad spectrum of AI applications, many of which overlap with the focus areas of Taiwan’s new supercomputer—such as language models, scientific research, and autonomous systems. What sets Colossus apart is its explicit AGI ambition, its integration with Tesla’s energy technology, and its status as a private, rapidly deployed infrastructure for frontier AI research and commercial innovation1235678.

How does musk define agi. If he means helping develop learning agents and digital twinning and other visions congruent with jensen huang I understand that future- if it means a computer than can do anythong including lead biptech projects , i dont believe agi is anywhere near emerging
nytimes.com favicon
nytimes
Why We're Unlikely to Get Artificial General Intelligence Anytime Soon
Why We're Unlikely to Get Artificial General Intelligence Anytime Soon
forbes.com favicon
forbes
Elon Musk's xAI Raises $6 Billion, Has Potential To Disrupt Healthcare
futurism.com favicon
futurism
Elon Musk Says That Within Two Years, AI Will Be "Smarter Than the ...
+7 sources

Elon Musk defines AGI (Artificial General Intelligence) as an AI system that is “smarter than the smartest human”—that is, a machine capable of matching or exceeding human intelligence across a broad range of cognitive tasks, not just narrow, specialized domains1356. In interviews and public statements, Musk has repeatedly predicted that AGI could arrive as soon as 2025 or 2026, emphasizing that, by his definition, AGI would be able to perform any intellectual task a human can, and potentially do so better356.

Musk’s vision for AGI includes:

  • Learning agents capable of reasoning, adapting, and generalizing across domains.

  • Systems that can drive scientific discovery, including in biotech, healthcare, and materials science—potentially leading and accelerating projects in those fields2.

  • Integration with robotics (e.g., Tesla’s Optimus robot) and brain-computer interfaces (Neuralink), aiming for AI that can interact with and augment the physical world2.

  • Digital twinning and large-scale modeling, which are congruent with Jensen Huang’s vision of AI-driven simulation and digital representations of real-world systems.

However, Musk’s definition is much broader and more ambitious than simply building advanced learning agents or digital twins. He envisions AGI as a general-purpose intelligence, not just a set of powerful tools for specific industries135.

Current Reality:
Most experts, including many in the field, agree that AGI as Musk defines it—a system that could autonomously lead biotech projects or match human creativity and reasoning across the board—is not imminent1. While AI is rapidly advancing in learning, simulation, and automation (and can already assist in biotech and digital twin projects), the leap to true AGI remains highly speculative and controversial13.

In summary:
Musk’s definition of AGI is a machine intelligence that can outperform humans at virtually any intellectual task, including leading complex projects like those in biotech. While some near-term visions (learning agents, digital twins) are congruent with Jensen Huang’s and the broader AI industry’s roadmap, Musk’s broader AGI goal is far more ambitious and, according to most experts, still out of reach135.

But for example does musk intent=d to build agi to augnment what his brain can do or is he really aiming to buold something that can do everything his brain can do and more?
forbes.com favicon
Forbes
Elon Musk Predicts Artificial General Intelligence In 2 Years. Here's ...
ctol.digital favicon
CTOL Digital Business and Technology News
Elon Musk's Bold Vision for 2025: Autonomous Cars, AI ...
time.com favicon
Time
Elon Musk: The 100 Most Influential People in AI 2023 | TIME
technologyreview.com favicon
MIT Technology Review
With Neuralink, Elon Musk Promises Human-to-Human Telepathy ...

Elon Musk’s vision for AGI is not just about building AI to augment his own cognitive abilities, but rather about creating an intelligence that can do everything the human brain can do—and more.

  • AGI as Human-Level and Beyond:
    Musk defines AGI as a system that can perform any intellectual task a human can, matching or exceeding human intelligence across all domains3. His goal is not limited to personal enhancement, but to create a machine intelligence that could, in principle, independently reason, solve problems, and innovate at or above the level of any human.

  • Augmentation vs. Replacement:
    Through Neuralink, Musk does pursue the idea of augmenting human brains—enabling direct brain-computer interfaces, potentially leading to “AI symbiosis,” where humans and AI work together or even merge capabilities45. This could allow humans to keep pace with rapidly advancing AI, communicate directly with computers, and share information brain-to-brain.

  • AGI Ambition:
    However, Musk’s broader AGI projects (such as xAI and his push for “TruthGPT”) are about building an autonomous intelligence that is not just an assistant or augmentation, but a system that could independently perform any cognitive task, including leading complex scientific or biotech projects23. He envisions AGI as a tool that could accelerate discovery, automate research, and potentially make decisions and innovations without direct human oversight.

  • Musk’s Public Statements:
    At CES 2025, Musk predicted that AI would soon be able to complete nearly all cognitive tasks that don’t involve physical manipulation, and that within a few years, AI could surpass human intelligence in many domains2. He sees this as both an opportunity (for progress and problem-solving) and a risk (if not properly aligned with human values).

In summary:
Musk’s intent is to build AGI that can do everything his brain can do—and much more—not just to augment his mind, but to create a truly general, autonomous intelligence. While he also pursues human enhancement through Neuralink, his AGI ambitions are aimed at creating a machine intelligence that could independently master any intellectual domain, including those as complex as leading biotech innovation2345.

is neuralink actually in use already - if so what sort of focus does it have?
MicroVentures
Mind Over Machine: Neuralink's History and Milestones
reuters.com favicon
Reuters
Musk's Neuralink raises $650 million in latest funding as ... - Reuters
builtin.com favicon
Built In
What Is Neuralink? What We Know So Far. - Built In
medicaldevice-network.com favicon
Yahoo
Neuralink wins FDA breakthrough tag for speech restoration

Yes, Neuralink is already in use, but its current focus is strictly medical and experimental—not general cognitive enhancement or mainstream brain augmentation.

Current Status and Focus

  • Clinical Trials:
    As of mid-2025, Neuralink has implanted its brain-computer interface (BCI) devices in at least five human patients, with plans to expand to 20–30 more this year. These trials are taking place in the U.S., Canada, and other countries1256.

  • Primary Applications:

    • Restoring Autonomy for Paralysis: The main goal is to help people with severe paralysis (from spinal cord injuries, ALS, etc.) control digital devices—such as computers, smartphones, and even robotic arms—using only their thoughts123467.

    • Communication: Neuralink enables non-verbal individuals to communicate by controlling a virtual keyboard or mouse with their brain signals34.

    • Vision Restoration: Neuralink is developing a “Blindsight” implant to restore basic vision to blind individuals by stimulating the visual cortex. This device has received FDA Breakthrough Device Designation and is expected to begin human trials by the end of 202515.

    • Speech Restoration: The device is also being developed to help restore speech for people with severe impairment due to neurological conditions24.

    • Prosthetic Control: Ongoing studies aim to allow paralyzed patients to control robotic arms or other prosthetics directly with their thoughts1.

  • Other Potential Medical Uses:
    Neuralink aspires to treat a range of neurological disorders, monitor and possibly treat mental health conditions, and eventually restore or enhance memory and cognitive function, but these are future goals and not part of current human trials3.

Summary Table: Neuralink’s Current Focus

Application AreaIn Use / In Trials?Details/Status
Paralysis/Device ControlYesPatients control digital devices with thoughts
CommunicationYesVirtual keyboards/mice for non-verbal individuals
Vision RestorationPlannedBlindsight device, FDA breakthrough status
Speech RestorationPlannedFDA breakthrough status for severe speech impairment
Prosthetic ControlEarly TrialsRobotic arm control by thought
Cognitive EnhancementNot YetLong-term aspiration, not current focus

Key Point

Neuralink’s technology is not currently used for general cognitive augmentation or “boosting” healthy brains. Its present use is focused on restoring lost functions for people with severe disabilities, such as paralysis or blindness. Broader applications like memory enhancement, learning acceleration, or general brain-computer symbiosis remain speculative and are not part of current clinical use12357.

Related
Is Neuralink currently in use for medical or experimental purposes
What specific functions or conditions is Neuralink focusing on with its implants
How successful have the recent human trials been for Neuralink's devices
Are there plans to expand Neuralink’s applications beyond paralysis and vision restoration
What are the main technological milestones Neuralink aims to achieve next
Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)