The Latest Research and Perspectives Shaping AI Today

Every day brings something new in AI, and sorting the noise from what actually matters can feel like a massive undertaking. We channeled our expertise as builders, investors, and operators to help you navigate the trends and breakthroughs shaping the future. Through market analysis, insights from our portfolio, and conversations with top researchers, our goal is to give you a clear view of AI today and where it’s headed tomorrow.

portfolio
pioneers

Exploring the latest research and innovations from across our portfolio.

Research Highlights

Skild: Building a Robotics Foundation Model

What does it look like to build toward AGI for the real world? Skild released their 2024 research results for the development of Skild Brain, taking us behind the scenes of their journey toward “any task, any robot, one brain”. 

Anthropic: Persona Vectors and AI Character Traits

Why did Bing’s chatbot adopt the persona “Sydney”? Anthropic’s new paper introduces persona vectors—patterns in a model’s neural activity that reveal and help control traits like sycophancy, hallucination, or “evil.” Their preventative steering technique offers a way to monitor and curb these personality shifts to keep models aligned with human values.

Mistral AI: Inside the Development of the Company’s First Reasoning Model

Magistral was created to “think things through,” while incorporating expertise from different professional domains, providing transparent reasoning, and offering multilingual flexibility. And they’ve documented both their successful and unsuccessful experiments in a new paper.

Product Milestones

Learn more

Goodfire Launches Paint with Ember

Traditionally, interactions between creators and image models centered around a prompt box. Goodfire has replaced this with a canvas that plugs directly into the “brain” of the model and uses its understanding of visual features to edit and create images in new ways.

Learn more

Exa: Introducing Exa Fast

We believe the future of search will be invisible infrastructure embedded in every app and powering AI in the background. Exa Fast is built for that vision. In internal benchmarks, it showed p50 latencies under 425ms, supported by a fully custom engine (no wrappers, no scraped SERPs).

Learn more

EvenUp: Transforming Hours of Legal Work Into Minutes

In the legal industry, outcomes often hinge on large amounts of unstructured data. EvenUp’s latest releases—AI drafts, Smart Workflows, and Medical Bill Summary—are the latest examples of how targeted AI can close the justice cap for personal injury, triple drafting output, and reduce average settlement times. 

Learn more

LMArena: The Video Model Leaderboard

After just one week, over 14,000 community votes have shaped the first public leaderboard for text-to-video and image-to-video models on LMArena. From Veo 3 to Kling 2.1 Master and Pika 2.2, the rankings are updated live and grounded in real user feedback.

Lightspeed Lens:
Research and Commentary

Apple’s AI Reasoning Paper: Good Questions, Few Answers
7/2/2025
Learn more

Apple’s AI Reasoning Paper: Good Questions, Few Answers

You may have seen a paper out of Apple last month with the title, “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity”, which caused a bit of a stir.

Visual
Breakdowns

  • 2/26/2025

Market Map: Where We're Making Our AI Investments

Market Map: Where We're Making Our AI Investments

Research Worth
Watching

Jonas Hübotter: 1000 Expert Models, 100x Faster, For Free

Play now
5/31/2025
  • Nnamdi Iregbulem
  • Jonas Hübotter

Daniel Kang: AI Hackers Are Coming Sooner Than You Think

Play now
5/2/2025
  • Nnamdi Iregbulem
  • Daniel Kang

Satish Chandra: Meet Google’s Agent-Based Systems that Automatically Reproduce and Fix Software Bugs

Play now
6/25/2025
  • Nnamdi Iregbulem
  • Satish Chandra

Andy Zhou: Language Agent Tree Search & Autonomous Red Teaming

Play now
4/8/2025
  • Nnamdi Iregbulem
  • Andy Zhou

Charlie Snell: Optimally Scaling Test-Time Compute & Predicting Emergence

Play now
3/11/2025
  • Nnamdi Iregbulem
  • Charlie Snell

Bill Dally: Revisiting Our Conversation With NVIDIA’s Chief Scientist and Senior Vice President for Research

Play now
1/16/2025
  • Michael Mignano
  • Bill Dally

Dr. Olga Russakovski: What the Next Generation of AI Talent is Learning

Play now
12/19/2024
  • Michael Mignano
  • Dr. Olga Russakovsky
  • Daniel Kang: AI Hackers Are Coming Sooner Than You Think

    • 5/2/2025
    • Nnamdi Iregbulem
    • Daniel Kang
    Play now
  • Satish Chandra: Meet Google’s Agent-Based Systems that Automatically Reproduce and Fix Software Bugs

    • 6/25/2025
    • Nnamdi Iregbulem
    • Satish Chandra
    Play now
  • Andy Zhou: Language Agent Tree Search & Autonomous Red Teaming

    • 4/8/2025
    • Nnamdi Iregbulem
    • Andy Zhou
    Play now
  • Charlie Snell: Optimally Scaling Test-Time Compute & Predicting Emergence

    • 3/11/2025
    • Nnamdi Iregbulem
    • Charlie Snell
    Play now
  • Bill Dally: Revisiting Our Conversation With NVIDIA’s Chief Scientist and Senior Vice President for Research

    • 1/16/2025
    • Michael Mignano
    • Bill Dally
    Play now
  • Dr. Olga Russakovski: What the Next Generation of AI Talent is Learning

    • 12/19/2024
    • Michael Mignano
    • Dr. Olga Russakovsky
    Play now
  • Jonas Hübotter: 1000 Expert Models, 100x Faster, For Free

    • 5/31/2025
    • Nnamdi Iregbulem
    • Jonas Hübotter
    Play now
The AI Digest The AI Digest The AI Digest

The AI Digest

Each month on LinkedIn, we help you cut through the noise and highlight the AI updates you really need to know. It’s your place to catch up on AI research, market trends, and insights that matter most to founders, investors and builders.