Urgent.News

the world's headlines, one feed

Editions

AI

As Berkshire goes on a $20B buying spree, these AI-picked stocks are up +204%

As Berkshire goes on a $20B buying spree, these AI-picked stocks are up +204%

In a dramatic reversal, Berkshire Hathaway made a $20 billion buying spree in Q2 2026, marking its first foray into risk assets since 2023. The conglomerate turned net buyer with $23.5 billion in equity purchases, dwarfing its $3.7 billion in sales. Warren Buffett's firm is now backed by $710 billion in liquid assets and another $13.5 billion in undisclosed Q2 purchases.

AI-powered ProPicks subscribers have reaped massive rewards, with a staggering +204.31% return since November 2023, outperforming the S&P 500 by +121.89%. Among their high-value picks are Alphabet (Google's parent), Diodes, Gartner, and Mativ Holdings, all of which have delivered double-digit gains. ProPicks AI's sophisticated algorithm scrutinizes thousands of equities, processing 15 years of financial data and 150 quantitative models to identify high-conviction stocks.

Subscribers receive monthly lists of picks, complete with the rationale behind each inclusion and removal, allowing for data-driven investment decisions.

Written by urgent.news from Investing.com's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.

Read the original at investing.com →

More in AI

Mistral AI Regional Endpoints Bring EU and US Inference Controls to Enterprise Deployments

Mistral AI has introduced regional inference endpoints for Europe and the United States, giving API customers a documented way to select where model inference is processed.

  • Mistral AI introduces regional inference endpoints in Europe and US for enterprise deployments.
  • Two dedicated API base URLs provided: api.eu.mistral.ai for Europe, api.us.mistral.ai for US.
  • Regional processing applies to data involved in model execution, excluding control plane elements.

Compression Is Prediction — and It Explains Why LLMs Actually Work

Here's something that blew my mind recently: compression and language modeling are, at their core, trying to solve the exact same problem.

  • LLMs function as advanced compression algorithms predicting next data in sequence
  • Training minimizes cross-entropy loss, reducing bits to encode training data
  • Improved compression methods enhance LLM performance and expand capabilities