We produce an annual report that covers Google's overall environmental impact (of which Gemini models inference is a portion). See our 2025 report (covering 2024):
RESEARCH
-
Bad Model Good Context Outperforms Good Model Bad Context
By
–
bad model + good context > good model + bad context The Big Labs don't want you to know this
-
Mirage 2 Revealed: Major Leap Forward in World Model Technology
By
–
1/ Mirage 2 was revealed. Mirage 1 was released just a month ago!
— Chubby♨️ (@kimmonismus) 21 août 2025
The leap in quality is breathtaking.
"Where Mirage 1 revealed the raw potential of a GTA-style world model, Mirage 2 takes a giant step forward: a general-domain world model that empowers you to create, play,… pic.twitter.com/KsE7QmdDwx1/ Mirage 2 was revealed. Mirage 1 was released just a month ago! The leap in quality is breathtaking. "Where Mirage 1 revealed the raw potential of a GTA-style world model, Mirage 2 takes a giant step forward: a general-domain world model that empowers you to create, play,
-

RotBench: Evaluating MLLMs on Image Rotation Detection
By
–
RotBench Evaluating Multimodal Large Language Models on Identifying Image Rotation
-

DeepSeek V3.1: Game-Changer or Hype Analysis
By
–
Is DeepSeek V3.1 a game-changer or mostly hype? Let’s break it down 👇
— Louis-François Bouchard 🎥🤖 (@Whats_AI) 21 août 2025
What’s new:
• Still a similar MoE transformer (671B params, 37B active) with 128K context
• Merges DeepSeek V3 + R1 → one hybrid model with thinking (reasoning) and non-thinking (direct) modes — like… pic.twitter.com/WKMPGwGlQMIs DeepSeek V3.1 a game-changer or mostly hype? Let’s break it down What’s new: • Still a similar MoE transformer (671B params, 37B active) with 128K context • Merges DeepSeek V3 + R1 → one hybrid model with thinking (reasoning) and non-thinking (direct) modes — like
-
State of AI in Business 2025 Report Released
By
–
Corrected link: https://
mlq.ai/media/quarterl
y_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
… -

Questioning AI Business Report Methodology and Generalizability
By
–
Okay, got the report. I would read it yourself. I am not sure how generalizable the findings are based on the methodology (52 interviews, convenience sampled, failed apparently means no sustained P&L impact within six months but no coding explanation). https://
mlq.ai/media/quarterl
y_decks/v0.1_State_of_AI_in_Business_2025_Report.pd
… -
Pilot Failures Report Impact on Market Movement Analysis
By
–
I have no doubt pilot failures are high, but I think it is really hard to see how this report gives the kind of generalizable finding that would move markets.
-

Methodological Concerns in 2025 AI Business Report Analysis
By
–
Okay, got the report. I would read it yourself. I am not sure how generalizable the findings are based on the methodology (52 interviews, convenience sampled, failed apparently means no sustained P&L impact within six months but no coding explanation). https://
mlq.ai/media/quarterl
y_decks/v0.1_State_of_AI_in_Business_2025_Report.pd
… -
Project NANDA Report: Understanding 95% AI Pilot Failure Rate
By
–
Has anyone seen a copy of the Project NANDA report on 95% of AI pilots failing? It has been reported about widely, but I haven’t been able to actually read it, though I filled out a form to get access. I want to understand what they found and how they measured it. Any links?