Join our daily and weekly newsletters for the most recent updates and specific content of the industry AI's business. learn more
The Ai world was sheed last week when DeepseekAi named his Chinese, the latest language model Desertk-R1 similar to matching American major American American systems of the cost. The announcement has been freedom Deallow Deal This placed the nearly $ 200 billion from NVidi Market value and heat discussions about AI Development Future.
The statement suggested out quickly that exexseek would be Breakified basically The economies of the progressed building systems are achieving with only $ 6 million what had to perform billions. This explanation put down through Glen Silicon, where similar companies Openi, Anthropic and Google Only deposit investigations in computer infrastructure to maintain the technology edge.
But in the middle of masonry and breath headless ends, Dario AomiEstablishment of Athorpic property and one of the designers of the branch groups on the back of large language modules today, a detailed analysis which offers a more enthusiastic perspective on deep achievements. His A blog post Cutteria to deliver a number of essential times about what deep deep it will be for the future of the future.
These are the four major breeders from Andei study that redesigning is that the redesign is that the redesign is that our understanding of a deep inspiration.
1. The '$ 6 million statement lose a crucial context
Desertek has been reported Development Costs must be seen through a wider lens, depending on anonode. He is connected to the popularity explanation:
“It's not a doilseek for $ 6 million how much you will only spend companies billions for us. 'I can only talk for a naming of $ 10s. Training (I will not give a correct number). Also, there were 3.5 sonet trained in any way that included a larger or more expensive model. “
This layering this whiskey moves on a shape of deep cost efficiency. When that think Sandet Training had been made to 9-12 months ago and still a deep model on many tasks, achievement reflects further according to natural breasts.
The time and context is also important. After historic shifts of reduction in developing AI – what is the deep cost structure like the proper cost of the loop.
2. Deepseek-v3, not R1, is the true technical performance
As Markets and Media were a key The model r1 deepsekAndei indicates that the company had the more important insurance at the company.
“Deepsek-v3 The actual innovation was and what should be taken to note a monthly monthly monthly (certainly). As a wealthy model, it seems that it seems to come close to state performance models of the US strengths, and how meantally impaired. “
The side is different between V3 and R1 is essential to understand very technology. V3 commanded major engineering generations, especially in managing the model '”Full-worth cache“And pushing the limits of the A combination of experts Mode (moe).
This scene helps to explain why a wonderful market response to R1 has been damaged. R1 was satisfactory of viability of non-job statutory learning v3 – a degree of companies currently adopted their models.
3. Full corporate investment reveals a different picture
The most prominent side of an amodei study may be concerned with deep total investment in AI development.
“There was a statement – we cannot be sure that it is true – that deep was 50,000 Chips of HopairI am steering inside a factor ~ 2 —- 3x of what's large US companies. The 50,000 chips will cost that to the order of ~ $ 1B. Therefore, the total cost of Damplek is not company (as different from the cost of spending or train individual model) from the US AI Labs. “
The spills is very much making updates on the statement of deep socien efficiency. Although the Company may have achieved a dramatic results with a separate model training, the investment seems to be raised in development which is likely to be wrecked.
The difference between Model Training and Corporate Investment covers the importance of major resources in organizing AI. He suggests although it can continue to be improved, competition competition in AI continues to stay in a large capital investment.
4. The current 'Right' Crossover
Andei describes the current time in developing AI development as presented to the fleet.
“So we are in 'crossesover' crossover 'crossover' crossing, it is a number of companies can provide a number of companies,” wrote. “This sweeply swells be true because everyone moves beyond the tale of the story of these models.”
This observation provides a vital context for understanding the state of the AI contest. The potential companies to complete this similar products in reasonable abilities represent temporary emotional expression rather than new quo status.
The impacts are important to develop the future development of the future. As companies still mate their modules, particularly in learning environmental confirmation, the field seems again in training and infrastructure. This proposes that deeply bedsted milestone would earn a positive milestone, it was not modified in the positive development long-term AI.
The true cost of the building AI: What does a Amodei analysis appears
AModi's detailed study on Dalweek Sessions will cut through weeks of construction systems that building AI practice systems. Her blog blooch both shock and their enthusiasm and shows how a $ 6 million training fee is responding within AI training procession.
Mars and media markets will be towards simple statements, and the story of the Chinese company highlights development costs which are confirmed. But still anonode breach of a more complex reality: total deposit investment, especially it describes $ 1 billion in computer hardware. .
This tim is of expenditure between our US's charges and the SEA US development marks and calling “a temporary window where several companies can accomplish a similar window. His full audit proposes that this window closen as potential to bodies will be enabled inevings. The range is likely to return groups with the deepest facilities.
Building the reasons why it is still caution, and a careful survey is showing why heated its real cost ordering full of investment. The polite drop of deep achievement may be found at the end more important than the original emergency raised in the markets.
Source link