Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
This week, Google has AI news from Microsoft, Openai and Antropic, it is covered in the following news section. Most of the product updates are built on top of these companies on the “basic” models. These are great AI models that are trained once, it can perform all kinds of different tasks. Today’s big language models are trained to predict the next word in a sentence, but after performing many language tuitions, to meet the questions to answer questions such as virtual encyclopedias, after performing many language tasks.
However, there are still some advantages to develop narrower adapted foundation models for certain areas. For example, Google Deepmind’s Alphafold 3 is a foundation model for biology. Cannot write poetry. However, the structure of proteins and can predict interactions between the two proteins or any protein and any small molecule. This is very useful for tasks such as medicinal design. UK self-controlling car start, founded basic models It can handle the car and drive a very different side of driving identification facilities, and for example, accelerating and braking. Robotics company planted physical intelligence Basic models for robotics This can help perform any different tasks without any additional education of any robot.
For businesses, it is much easier to see a way for Roi from more narrow foundation models from the generalist LLMS. Swiss army knife is excellent. But you probably wouldn’t want to use it to make a transaction. In today’s eyes, I want to introduce you in AI PunchA silicone valley company built an institutional model that facilitates something sitting in the center of work decisions: to make accurate forecasts.
Time, saving information and money
Normally, it requires painkilling work by data scientists to give predictions, weeks or even months of information. Machine learning and deep learning – the lower branch of the bench near today’s AI has been applied to predictive analysts for years. However, these models should generally be prepared in a specific database to make only one special forecast in a specific context and use them without showing accurate forecasts. Great technology companies and main retailers often have the regions of the information you need to cultivate such predictive AI models. But there are no many small enterprises.
Kumo’s new RFM model can manage all kinds of different forecasts, which are suitable for customers and are affordable for customers on the other hand. Credit Default risk will need a chance to be read in 24 hours from the hospital from the hospital, and it can manage all of these different predictions, and almost without any additional exercise. “You point to this in the foundation model, this is how and forecasts, and forecasts,” Jury Leskovec, “Kumo President Stanford University computer scientist, told me using a computer scientist, customer Churn model. He said that a client can better fix the model according to its information and can get a 10% improvement in accuracy of their predictions.
Using graphe neuron networks to find key correlations
Kumo’s model applies to graphic neuron networks that can cod in Leskovec’s ways in the network of the network and to understand how the information in different tables changes in different tables. (RFM, the name of the Kumo’s model, the relational stock model.) Focus on a graphic structure with a graphic structure with a graphical structure with the same type of transformer architecture, and even important, important, predictive data to pay attention to a precise forecast. Open information on the foundation, as well as open information, as well as Leskovec has openly presented a large amount of synthetic information.
The time stamp along the tables of a user is correct, Kumon’s model can make a highly accurate prediction. Kumo’s Benchmark tests, without any subtle adjustment, some traditional machine learning methods perform better, a model is better than anything that has been prepared for a model or more than anything special training for this position. With additional delicate arranging, the traditional way of traditional way for a task is better than the Neyron network, for some tasks or some tasks. And critically, the METAN LLAM 3.2 B, when compared to the use of a large language model and is compared to create forecasts based on a desire, the RFM improved. (Kumo’s Benchmark results are not independent and not confirmed.)
KumorFM’s consequences can also be construed more than many hand-engine models that are building information analysts. Because human data analysts are sometimes predictors, a client visits an ad after 10 PM, but if they see an ad after 10 PM, you may have more likely to take a certain product. “Today’s models can only explain through the signals that you arise. However, we made this decision because of these reports, we made this decision because of these events.”
Kumo, including Sequoia Capital, has used $ 37 million to finance $ 37 million to date, and now used a team of 50 people around 50 people. His models have been used by companies including food delivery application so far NineReddit and the food chain of the UK Sainsbury, among others.
For enterprises fighting the cost of removing the value of the data and the construction of predictive models, Kumo’s approach can make an important effective progress. (Amazon Web services offers a foundation model called Chronos to predict the things that occur in a timely sequence, but require delicate arrangements to achieve accurate results. Data monitoring software company Datadog It also offers a similar baseline model called Toto. This is the rest of this week.
This is more of the news here.
Jeremy Kahn
jeremy.kahn@fortune.com
@jeremyakahn
Before coming to the news, the latest luck list of women’s most powerful female women and Amd CEO Lisa Su, Deputy Chairman Meng Wanzhou, an anthopic president Daniela Amodani and Teaching Mirhines Labent and CEO Mira Murati and CEO Mira Murati. You can view the list here. New York Times’ CEO Meredith has a great interview with Levien FortuneThe publisher said that the AI sees as an opportunity and threat and why he sued Openai. You can check this here.
This story was first displayed Fortune.com
