TODAY IN AI
| The battle for top AI talent continues with a flood of DeepMind researchers headed to Meta and Microsoft. Source: Created with Gemini |
-
Google opens access to ‘fastest 2.5 model yet’: Google just ~released~ what might be the perfect model for developers who want to scale up their AI-powered apps without breaking the bank. With a cost of just $0.10/1M input tokens, Gemini 2.5 Flash-Lite is best for high-volume tasks where speed and low-latency are key — things like generating summaries, translating across languages, and extracting screenshots from videos.
-
OpenAI races to secure more chip capacity: OpenAI announced it’ll pay an unprecedented $30B per year for ~4.5 gigawatts~ of data capacity from cloud giant Oracle. That should give the ChatGPT-maker access to more than 1M AI chips by the end of the year, a crucial step to train GPT-6 and beyond. The startup faces stiff competition from xAI, which already has hundreds of thousands of GPUs online while looking for $12B to ~fuel even more~ chip spending.
-
Meta scoops DeepMind researchers behind IMO gold breakthrough: Just days after DeepMind turned in a record performance on the International Mathematical Olympiad, Meta has ~lured away~ three of the lab’s top researchers. While we don’t have many details, rumors ~suggest~ Meta recently offered $1.25B to one sought-after researcher in a different negotiation. And it doesn’t end there: Microsoft has also joined the industry’s cutthroat talent war, ~poaching~ around two dozen DeepMind researchers in recent months to beef up its Copilot offerings.**
