저자: ChamathPalihapitiya · 외국인
요약 보기
엔비디아 CES 발표 중 $3,000 개인용 AI 컴퓨터(Project Digits)와 로보틱스·자율주행 전략을 논평. 추론(inference) 중심 시장 전환과 엣지 AI 수요 확대를 언급하며, AI 스타트업 증가로 경쟁이 치열해질 것이라고 봄. 본인은 전력·컴퓨트 리소그래피 등 인프라에 투자했다고 밝힘. 종목별 가격 방향 예측은 명시하지 않음.
다듬은 글
going consumer let's speak about it Nvidia made a huge announcement at CES this week they made a bunch of them one of them that was very interesting was this $3,000 personal AI computer for researchers it's named project digits and it's essentially like maybe Arduino would be a way to view that this like a personal device but it's strong enough to run llms on they're also going after physical AI like Robotics and self-driving as we said here on the award event a lot of folks on the panel were saying this year will be the year of Robotics and they stated that they're going to have driver assistant chips and maybe develop worlds for people to simulate which net net at the end of the day I think freeberg puts them on second it would put autonomous Partners on second or third base in terms of generating technology by implementing it into the chips and onto their stack I don't have much of an opinion on that to be honest similar to what I stated on the prediction show weo and Tesla will dominate this Market in my opinion additionally I believe it will push traditional vehicle oems to undergo a lot of consolidation the Intriguing thing in my opinion is that they really went all out and built a respectable robotics test bench that struck me as being rather intriguing so in terms of the long-term future I believe that supports what many intelligent people have said including what freeberg and Gavin also discussed that was cool in my opinion for robots the low-end PC confused me a little what the point of that is I do not comprehend I suppose it creates a crazy deep in Market where you can purchase a GPU then use it to contribute to a distributed Network and let certain distributed workloads run on it it's a toy a hobbyist contraption that functions similarly like a bridge in my opinion and we frequently see this in the field of Technology where some people build things like the initial PCS which let's face it were more like toys and hobby gadgets arduinos and the first drones yes I suppose the purpose of the toy is to do what because everything tells us that we are approaching the boundaries of training when you try to do inference however that is an llx yes but the point is that it's not let me begin yes indeed thus there are two types of AI in the modern world inference and training and as of right now we believe that certain training has reached its limit the market is currently moving towards inference for what purpose will you utilize this modified personal computer if you decide to purchase it I suspect a test time compute use case which is a use case for inference however it's unclear to me why that's a better option than all of the AI accelerators and tensors that are currently widely being made public whether it's Amazon disclosing their activities Google disclosing their activities or a long list of startups declaring their activities I was simply perplexed I'm not really sure what the point is freeberg what are your thoughts on this if the market develops as they believe the robotics issue was intriguing therefore freeberg we may be discussing two or three distinct things here which of these two the $3,000 desktop GPU that you connect to your PC and use to play locally do you find most intriguing is that encouraging in your opinion if you were to guess where would that go we've discussed Machine Vision models graph neural Nets and other technologies that are used for weather forecasting so I believe the wager he's making is that it's not simply llms that are predicting text genome language models of this type are now being developed in an effort to forecast genomic output for biotech applications additionally there will be robotic response and real-time Machine Vision for example at ohal we are working on this project and attempting to determine the best runtime environment for these kinds of systems that will use robotic response systems and Machine Vision additionally a large number of these Industrial uses are developing imagine operating a robot in a warehouse that robot in the warehouse sending data to the cloud while it waits for a model to run and receive a response is that really what you want it's likely that you want that at the Network's Edge something local is what you want furthermore I don't believe he has a strong opinion about the kinds of models and Industrial applications that will be used hush money strapped his smart monocle over his forehead his wager however is that the models are sufficiently good good and that the chips are now sufficiently good to be able to use Machine Vision to realize realtime responses using realtime input and then react rapidly with the local model running whatever that model is to drive some output in the industrial setting and that there will be a lot of these kinds of applications whether they are for developing new research models operating robots in warehouses or generating predictions for biotech research alternatively this PC might be strapped to the rear of a car tractor lawnmower humanoid robot or any other type of device thus I believe that freeberg explain to the audience the advantages of having the computer at the edge for those who might not be aware it's more faster to run the model locally if you're ingesting a lot of data and then need to run a lot of data in a model i i i i i i i for example Tesla's self-driving technology does not broadcast video footage from your vehicle to a computer located thousands of miles away then leave it up to the server to choose how to operate your vehicle the vehicle is using its model to determine what to do in relation to the visual imagery it's local because you don't have to wait for the internet to send and receive data because all of that data can be handled in the automobile there is no lag time for you the 60 and 100 millisecond reaction times are not available to you losing your phone connection and then being unsure of what to do is not something you have that's right in the data centers servers may break or the connection may drop as you wait for one to come online it's all local therefore you don't need to be an expert in Hardware if you strap this in video computer which is just plug andplay it can be used to run research applications locally or strapped onto the back of a humanoid robot whether it replaces the Apple Macintosh Pro Studio Gadget or something else entirely I believe there will be some pretty intriguing use cases yes Mac Mini 4 however many have noted that the $3,000 price tag of the Mac Mini 4 actually eliminates many Macs from the market it's insane simply put I am unable to use an operating system in the conventional sense Sam I recall that when you and I first began investing in companies two of the main motivators were free storage free bandwidth and cloud computing which greatly aided in the capacity to launch a product efficiently and rapidly among other things what effect will AI have on the numerous startups that are starting to emerge in 2024 and 2025 examine your crystal ball and predict how they will expand their footprint how will the startup scene be accelerated by this in fact I believe that there will be a Cambrian boom in the invention and development of various things and while some of them will be brilliant ideas others will be dumb ones however I believe it will make our work increasingly difficult particularly when we are just starting out with Investments there will simply be a lot of people with identical IDE ideas at the same time who can execute swiftly and do things at hither to unattainable speeds and it will get increasingly difficult you know yes additionally I've been considering whether or not you invest in rival companies which is something I never did before do you wager on an index a whole category or something else that interests you what's the strategy at seed and preed you know because when I have an idea I think wow that's really cool after that I go outside and see that 30 people are working on it and it wasn't always like that additionally I believe that we have actually unlocked a tool it provides people the opportunity to do things that would have been prohibitively expensive or makes them believe gosh I could be an entrepreneur and I can try this and I could do this as a result I'm witnessing folks do a variety of things # Summit some of the AI stuff is merely a feature for entrepreneurs right now it's just table Stakes it's essentially a conversation or something and it doesn't really matter however you can see that individuals are reimagining games and even your kitchen appliances among other things therefore I do believe that it will be really challenging additionally I frequently start hype Cycles if you persevere my personal hype cycle could come true though it takes a long time I therefore made investments in power and compute lithography which are essentially the foundations of all of this until it begins to stab
댓글 0