• Alphawise
  • Posts
  • Google Gemini is #1 at $0.1 / 1M tokens

Google Gemini is #1 at $0.1 / 1M tokens

Meta releases its new MILS project - LLMs can see and hear without any training!

Author

Matthew McCann
February 06, 2025

A technical AI newsletter
written with an entrepreneurial spirit for builders

What is today’s beat?


FOLLOW US ON SOCIAL FOR MORE


Your FREE newsletter
 share or subscribe 
to show support

🧨 GOOGLE 🧨

Gemini 2.0 Expands, and is released CHEAP

Google has announced significant updates to its AI offerings, introducing new models within the Gemini 2.0 suite and revising its AI ethical guidelines.

  • Gemini 2.0 Flash General Availability
    Offering enhanced performance and higher rate limits, at $0.1/ 1M tokens.

  • Introduction of Flash-Lite
    Input: $0.075/1M tokens (text, image, video)
    Output: $0.30 / 1M tokens (text).

  • Gemini 2.0 Pro is #1 on many benchmark

  • Revised AI Ethical Guidelines
    Google has updated its AI principles, removing prior commitments
    to remove its ANTI WEAPONS PLEDGE.

🗞️ NEWSROOM 🗞️ 

What’s hot in tech right now?

The Ukraine war is accelerating the deployment of AI-powered drones and autonomous systems, testing their battlefield effectiveness in reconnaissance, targeting, and electronic warfare.

Boston Dynamics is collaborating with its former CEO to accelerate AI learning for Atlas, aiming to enhance the humanoid robot’s real-world adaptability and autonomy.

OpenAI introduces data residency in Europe, ensuring regional storage and processing for enterprises needing compliance with EU privacy regulations.

Microsoft and CoreWeave are founding partners of New Jersey’s AI hub, aiming to drive research, innovation, and workforce development in AI technologies.

⭐️ REPOs of the DAY ⭐️

A framework for evaluating language models on diverse NLP benchmarks, supporting various architectures and datasets to assess reasoning, knowledge, and generation capabilities.

MILS - a META research project

Meta project that claims LLMs can see and hear without any training. MILS is a framework for learning invariant representations using meta-learning, improving generalization across domain shifts in tasks like computer vision and NLP.

⭐️ BUILDER BYTES ⭐️

What’s hot for builders right now?

NVIDIA introduces RTX AI Garage, Blackwell architecture, and NIM blueprints to accelerate AI-powered applications on PCs, pushing real-time AI and generative models forward.

Techniques for reducing memory consumption in PyTorch models, including mixed precision, gradient checkpointing, and memory-efficient tensor operations, to enhance training scalability.

AWS integrates NeMo Guardrails into SageMaker JumpStart, enabling developers to enhance LLM safety, control, and response filtering for enterprise AI applications.

🤩 COMMUNITY 🤩

What’s the latest beat?

Conf Talk

Demystifying LLMs: A Guide to Human-in-the-Loop Distillation
Ines Montani explores practical methods to distill large language models into smaller, efficient components for real-world applications, emphasizing human-in-the-loop strategies.

Video

The Engineering Unlocks Behind DeepSeek | YC Decoded
This video delves into the technical innovations of DeepSeek, highlighting their open-source reasoning model, R1, and its comparable performance to leading AI models.

Short Video

What are the differences between DeepSeek models?
Quick 3 minute video to explain the different Deepseek models.

THANK YOU

👇️ we are 100% free, so please let us know what you think! 👇️ 

Looking to promote your company, product, service, or event?