After the end of AI Action in Paris, anthropic founder and the CEO Dario Amodei called an event “lost opportunity.” He also added that “great and urgency is required for several articles that are given the speed when technology continues” in a statement issued on Tuesday.
Ai company held an event that focused on the Paris Engineering with French launches, and Techcrunch had the opportunity to negotiate Amodei stage. When he explained his thinking line and protected the third way that there was no pure hope or pure criticism on Ai Innovation topics and governance, respectively.
“I used to be Neuroscientist, where I looked inside the real brain to heal. -Amodei tells techcrunch.
“But it is race. It is a racing between the strong, fastest of us and wonderfully?
Since the first AI concentrative conference in Bletsley in the UK, the conversation tone around AI the treatment has changed dramatically. Partly because of the current Geopolitical Current country.
“I’m not here this morning I spoke about AWA safety over the past few years,” American President JD Vonce at A ac Action Tuesday. “I’m here to talk about AI chance.”
It is interesting that the Amodei tries to avoid this conflict between security and opportunity. In fact, believing to focus on the safety Is the opportunity.
“At the original conference, the UK Bletsley Summit, with many discussions about a variety of accidents. “If there is, making this kind of measurement helped us to better understand our models, the only one, helps us to produce better models.”
And every time Amodei emphasizes safety, she also likes to remind everyone that anthropic is still focused on building AI and built.
“I don’t want to do anything to reduce the promise. We provide models every day where people can build it and they are definitely not to stop doing so,” he said.
“When people talk about danger, I get offensive, and say: ‘Well, no one has done good work really how this technology can be.
Deepseeek ‘training’ expenses are not good ‘
When the conversation adjusted to the latest Chinese LLM-Maker DeepseeEek, Amodii has reduced technical achievements and said that he felt as if he felt that the response was “Anorganic response.”
“Reality, the way I react was very small. “The model released in December was in this kind of regular currency to reduce the cost we have seen in our models and other models.”
What was noticed that the model was not leaving the “three or four labs” based on the US installed on Google, Openai and Anthropic as other front labs stimulating the envelope by new model.
“And that was the story of caring for me geolipitor to me. I never want that authorized governments to control this technology,” he said.
As for increased training expenses, he chased the idea that the deep V3 training was 100x cheap 100x compared to the cost of training in US “I think [it] It is not just good and is not based on facts, “he said.
The next coming models have consultation
When Amodii did not announce any new model at a weekend event, he made some of the coming company – and yes, including some strength of thinking.
“We often focus on trying to make your own partnering models. We worry about making enough power, that the models are intelligent, and worry about safety,” said Amodii.
One of the problems trying to solve anthropic is a conindrum of choosing the model. If you have ChatGPT plus account, for example, it can be hard to know which model to choose from the model selection pop-up with your next message.
The same is true of developers using a large model model (LLM) API for their applications. They want to estimate objects between accuracy, answers and expenses.
“We have been confused in the idea that there are ordinary models and there are models to consult and that they are different from each other,” said Amodii. “When I speak to you, you don’t have a second brain and one of them respond right away and you like, the other waits too long.”
According to him, according to installation, there must be a smooth transformation between the former trained models like Claudude 3.5 Sonnet or GPT-4O and service delivery models and that can produce O1’s O1 or R1 Deepseek’s R1.
“We think this should be there as part of a single continuous business. “We must have a smooth change from that for previously trained models – rather than ‘the IA and this thing b,'” add.
As large AI-like AI companies continue to remove better models, Amodei believes that will open up the best opportunities that interfere with the world’s largest businesses throughout the industry.
“We also work on specific pharma companies to use Clouds to write clinic studies, and they have been able to reduce the time to write a clinic report from three days,” says Amodii.
“Except for biomedical, in the formal, financial, insurance, manufacturing, software, things that will be available – and we want to help you, we want to support it all,” concludes. .
Read our full coverage of the Intelligence action conference in Paris.
Source link
