Over 3,000 participants, including me, last week I went to Las Vegas Humanx AI conference.
The running theme during a three-day conference was trust-miantly how one can build trusted results from powerful, but by nature probabilistic technology.
Remember that it was the same week Apple He retreated After starting the AI function as a consequence of concerns about accuracy. And at Humanx, it will state that “only 6% of AI projects get into production” relaxed on the screens – statistics with AWS testserving as a blunt reminder of experiments and challenges in implementing artificial intelligence.
Nevertheless, Over $ 100 billion Invested in the AI sector in 2024, an increase of 80% in comparison with 2023, based on the common Humanx i Crunchbase You report developed for a conference.
Conference mixed panels and products are launched at many stages in a large exhibition hall with branches of sessions and answers and versions of products in smaller rooms. In extensive spaces there have been seats in the front room, capsules and application for meeting and making connections.
There was a lot of content, but perhaps my perspective in the most aroused, just over two years after launching OpenaiChatgPT has created a generative foremost current AI from some leading model firms.
Openai moat
Openaiprice $ 157 billion, I still had to reply the query Kate Rooney On CNBC“What is your moat?”
Kevin WeilThe company’s product director in the last 10 months has noticed that 12-month advantage days have disappeared, and the current reality of three to 6 months’ advantage is still “really valuable”.
And while in previous programming cycles “Database was a database”, Weil summed up Zeitgeist with “There is a new model every two months, [that] He can do something that computers have never been able to do. “
Despite this, the numbers of OPENAI are huge. Weil said that 3 million programmers use the API interface, over 400 million people use CHATGPT every week, and over 2 million firms use corporate products.
Anthropic on Claude code
The most significant event of the conference was the conversation between Alex HeathDeputy editor in The VergeAND Mike KriegerCPO Anthropicabout building a model company and its application building plan. Claude Code, launched a week ago, reached 100,000 users per week.
Krieger said he contacted the leading clients of API Antropic before launching, because he puts him in direct competition with clients, AnyoneCursor producer, Windsurf Z Codeum AND GirubCopilot.
He pressed the must have first-page products on the market, because “you just can’t get such feedback if you’re just a provider of API,” he said.
This learning of the first batch will go on to the model, “ensuring equal opportunities, transparent and then feeling it.”
“We hope that we will all be able to move after sometimes closer to admissions,” said Krieger.
In a more philosophical note, he admitted that he joined the Anthropic for the role he could play in “conducting the future of human interaction-Ai.”
“If these are just chat fields and chat bots for a year, everyone failed,” said Krieger.
Mistral, Open Source and smaller models
France Mistral you have It differs from anthropic and OpenAi as the builder of the Open Source model to create decentralized artificial intelligence so as to not dominate several firms. Arthur MenschCEO and co -founder Mistral, said that there is a huge demand for Open Source for people with data management and sovereign needs.
“What we introduce to our Open Source models is a platform for implementation, to create agents, manage data, to manage feedback, which can be implemented in a completely isolated way,” he said.
Thanks to the leading smaller models, Mistral is lively in robotic applications. “The implementation of a small vision model for action on a specific equipment will be extremely important in the coming years and we introduce a software stack to it.”
The company cooperates with Helsing on his drones and cooperates with robotics at Bay Area.
Mistral began as an enterprise company. However, when you have API interfaces, you are near having a product, said Mensch. Mistrala’s consumer product, Le Chat, was launched last month.
Next conference
Next yr, Humanx goes to San Francisco, because most of the AI investments are focused in the bay. And with a projection that just about 30% of firms presenting themselves in Humanx They are probably the goals of the acquisition, the conference may look completely different in a yr.