Welcome back to quick,
The Chinese company he deepseek has taken the world from the storm. News of the model’s skills briefly shaved $ 600 billion of market lid from the Nvidia chip giant, which was then thrown back and recovered almost half of it. With its latest R-1 model (free and open source) at the same time with O1 O1 (200 dollars a month and closed for paid users), the starts of the app based as perplexity have already begun use the model in their products. Elsewhere the companies are rethink their training strategies, I was told many fundamental of many of this week.
Although the model received allocations from the prominent leaders of him as CEO of Openai Sam Altman, some remain skeptical in space that Deepseek was likely to train his models with fewer resources, making it cheaper. Openai’s leading research official, Mark Chen, said on X: “I think the outside response has been somewhat overloaded, especially in cost stories.”
The model is also full of security risks, according to a report by the AI Chatterbox Labs rating company, and can produce hate speech, misinformation and clear sexual content. Researchers at the Cella Cyber Intelligence Company were able to bypass protective measures to use the model to create malware.
Now let’s get into the titles.
Tell me the money
President Donald Trump’s announcement for $500 billion that Stargate Infrastructure ProjectJapan -based softbank -based softbank and MGX -based investment firms, resulted in great drama with technology leaders criticizing megadeal. Elon Musk called the initiative “false”, claiming the companies involved “actually have no money”. He called Sam Altman, “a rogue”. Microsoft Director General Satya Nadella commented: “All I know is that I am good for my 80 billion dollars,” assigned to the placement of data centers. Not wanting to be left behind, Meta followed Stargate’s announcement, saying he plans to dive $ 60 billion to improve his models.
The people of him
A New Start called Palona Launched by a team of former Meta and Google students is building a sales agent of he Simulates emotional intelligence. CEO Maria Zhang said the company uses a model to oversee another and keep track of conversation. It claims this also ensures that the system it is correct but it is also convincing, humorous and has personality. The company has raised $ 10 million in seed funds.
The one -week deal
Clay, which uses artificial intelligence to filter sales of sales, has collected $ 40 million in a $ 1.3 billion rating. Among its customer base of 5,000 enterprises are dresses such as Anthropic and Openai, which uses clay to try sales tactics by surfaces useful data points.
Leak
The power of the Deepseek model and its price are already changing the way the American one runs their businesses. It is a cheap, convincing alternative to task offers such as Openai, Jesse Zhang, CEO and Decagon, which builds agents for customer service, said Forbes. The new Deepseek model is likely Strengthen the American giants like Openai and Anthropic to reassess their prices.
Eiso Kant, CTO and co -founder of Poolside, the unicorn he builds for software engineering, he said stale This deepseek force is in its engineering ability to do more with less.
“What Deepseek is telling the world is that when you put a strong accent By making your account training, You can do a lot, “he said.” There are extraordinary things that you can continue to squeeze from these nvidia chips to make them much more efficient. “
Others are less impressed. Director General of Writer May Habib said stale She is not surprised that Deepseek’s models, trained with a significantly smaller budget, are able to match the most intelligent Models in the SH.BA in October, the writer launched a model that was trained with only $ 700,000 when costing 4.6 million dollars for Openai to build a model with similar skills. The company used synthetic data to reduce its training costs.
“Even before Deepseek’s model explodes on stage, we have said that These models are comfortable. They are scattered more and more, ”Habib said.
Read the full story in stale.
Index of he
CompassionA Palo-Alto-based company that offers data sourced from millions of human experts to train leaders of it by companies such as Openai, Anthropic and Meta, announced today that it became profitable in 2024. CEO Jonathan Siddhaarth told me yes build up an “Agi Infrastructure Company”, Providing data to make models more intelligent and perform advanced tasks such as coding and reason with complex guidance. Turing, estimated $ 1.1 billion in 2021It also evaluates the performance of his models and builds applications in sectors such as health care and retail.
$ 300 million
Repeated annual revenues that Turing recorded in 2024, rising 3x from 2023.
4 million
Coders, scientists and professionals contribute to what Siddhaarth claims is “the world’s largest cloud of human intelligence”.
joke
This company is using it to better detect and remove the sexual abuse material of children generated by it from the Internet.
- inattention
- Typical
- beehive
- Meta
Check if you got it here.
Model behavior
When Google introduced summaries generated by the one who sat at the forefront of the links, the immediate score was hallucinated inaccurate responses that told people to eat rocks and add glue to the pizza. But one drawn longer, lesser -known effect is now playing: a Reduce traffic on individual websites, which has begun to damage companies like Kajak, Yelp and Tripadvisor, stale reported.