Musk Pledges SpaceX Shares to ‘Every Kid in America’ WebBloomberg TechnologyOct 19 Websites are closing their doors to AI agents WebThe Verge AIOct 17Popular
xAI OfficialWebLearn & DoResources People Podcasts
All NewsTradeVideosVoicesCommunityResearchPodcasts
Subvolts take AI benchmarks miss the mark; Prolific guests reveal how to truly measure AI's human impact. Dec 20, 2025
For you
Developers
Learn how to design and evaluate AI models that prioritize human experience and safety beyond technical metrics.
Business
Understand the limitations of current AI performance metrics and explore alternative evaluation methods for better product development.
Everyday users
Discover why AI models that score high on tests might still be unhelpful or unsafe for everyday use.
Hype note
3/5Notable

This episode offers a solid interview with researchers from Prolific, providing valuable insights into AI evaluation methods.

Machine Learning Street Talk
Machine Learning Street Talk Episode

Are AI Benchmarks Telling The Full Story? [SPONSORED] (Andrew Gordon and Nora Petrova - Prolific)

16 min Interview Hosts: Tim Scarfe
0:0016:04

Audio streams from Machine Learning Street Talk (MLST). Your place is remembered on this device.

Go to the podcast episode page
Open Machine Learning Street Talk in
Apple PodcastsOvercastPocket Casts
Follow the show
Apple PodcastsSpotifyOvercastPocket CastsAny Android app

From Subvolts

Summary and analysis by Subvolts editorial, AI-assisted.

Summary

Machine Learning Street Talk features Prolific researchers Andrew Gordon and Nora Petrova discussing AI benchmarks. They argue current tests often fail to capture the human experience, leading to models that perform well technically but poorly in real-world applications. The episode explores flaws in AI ranking systems and proposes a more humane approach to measuring AI helpfulness, safety, and relatability.

Explained simply

Imagine a race car that's super fast but terrible for driving to the grocery store. This episode explains how AI models can be like that race car. They might win technical tests but fail when real people try to use them, and the guests suggest better ways to check if AI is actually helpful and safe.

Why listen

This episode is for anyone interested in the practical application and evaluation of AI. It challenges the common reliance on technical benchmarks, which is relevant as frontier labs like OpenAI, Google Gemini, and Anthropic push model capabilities. Listeners will learn about the limitations of current AI rankings, the importance of human-centric evaluation, and the potential risks of prioritizing raw performance over user experience and safety, especially as AI integrates into sensitive areas.

On the mic
  • HostTim Scarfe
  • GuestAndrew GordonStaff Researcher in Behavioral Science, Prolific
  • GuestNora PetrovaAI Researcher, Prolific
About
xAIBenchmarksSafetyResearch
Key facts
  • Episode Title: Are AI Benchmarks Telling The Full Story? [SPONSORED]
  • Show: Machine Learning Street Talk
  • Length: 16 minutes
  • Guests: Andrew Gordon and Nora Petrova from Prolific
  • Mentioned Incident: Grok-3’s "Mecha Hitler"
Show notes from Machine Learning Street Talk (MLST)

Is a car that wins a Formula 1 race the best choice for your morning commute? Probably not. In this sponsored deep dive with Prolific, we explore why the same logic applies to Artificial Intelligence. While models are currently shattering records on technical exams, they often fail the most important test of all: **the human experience.**


Why High Benchmark Scores Don’t Mean Better AI


Joining us are **Andrew Gordon** (Staff Researcher in Behavioral Science) and **Nora Petrova** (AI Researcher) from **Prolific**. They reveal the hidden flaws in how we currently rank AI and introduce a more rigorous, "humane" way to measure whether these models are actually helpful, safe, and relatable for real people.


---


Key Insights in This Episode:


* *The F1 Car Analogy:* Andrew explains why a model that excels at the "Humanities Last Exam" might be a nightmare for daily use. Technical benchmarks often ignore the nuances of human communication and adaptability.

* *The "Wild West" of AI Safety:* As users turn to AI for sensitive topics like mental health, Nora highlights the alarming lack of oversight and the "thin veneer" of safety training—citing recent controversial incidents like Grok-3’s "Mecha Hitler."

* *Fixing the "Leaderboard Illusion":* The team critiques current popular rankings like Chatbot Arena, discussing how anonymous, unstratified voting can lead to biased results and how companies can "game" the system.

* *The Xbox Secret to AI Ranking:* Discover how Prolific uses *TrueSkill*—the same algorithm Microsoft developed for Xbox Live matchmaking—to create a fairer, more statistically sound leaderboard for LLMs.

* *The Personality Gap:* Early data from the **Humane Leaderboard** suggests that while AI is getting smarter, it is actually performing *worse* on metrics like personality, culture, and "sycophancy" (the tendency for models to become annoying "people-pleasers").


---


About the HUMAINE Leaderboard

Moving beyond simple "A vs. B" testing, the researchers discuss their new framework that samples participants based on *census data* (Age, Ethnicity, Political Alignment). By using a representative sample of the general public rather than just tech enthusiasts, they are building a standard that reflects the values of the real world.


*Are we building models for benchmarks, or are we building them for humans? It’s time to change the scoreboard.*


Rescript link:

https://app.rescript.info/public/share/IDqwjY9Q43S22qSgL5EkWGFymJwZ3SVxvrfpgHZLXQc


---

TIMESTAMPS:

00:00:00 Introduction & The Benchmarking Problem

00:01:58 The Fractured State of AI Evaluation

00:03:54 AI Safety & Interpretability

00:05:45 Bias in Chatbot Arena

00:06:45 Prolific's Three Pillars Approach

00:09:01 TrueSkill Ranking & Efficient Sampling

00:12:04 Census-Based Representative Sampling

00:13:00 Key Findings: Culture, Personality & Sycophancy


---

REFERENCES:

Paper:

[00:00:15] MMLU

https://arxiv.org/abs/2009.03300

[00:05:10] Constitutional AI

https://arxiv.org/abs/2212.08073

[00:06:45] The Leaderboard Illusion

https://arxiv.org/abs/2504.20879

[00:09:41] HUMAINE Framework Paper

https://huggingface.co/blog/ProlificAI/humaine-framework

Company:

[00:00:30] Prolific

https://www.prolific.com

[00:01:45] Chatbot Arena

https://lmarena.ai/

Person:

[00:00:35] Andrew Gordon

https://www.linkedin.com/in/andrew-gordon-03879919a/

[00:00:45] Nora Petrova

https://www.linkedin.com/in/nora-petrova/

Event:

Algorithm:

[00:09:01] Microsoft TrueSkill

htt

Jargon decoded
Benchmarks
AI Researcher
Tests used to measure and compare the performance of AI models.
Leaderboard Illusion
AI Researcher
The misleading idea that top rankings on tests perfectly reflect an AI's real-world usefulness.
TrueSkill
Microsoft Research
A skill rating system, similar to Elo, used in games to rank players.
Published
Show
Machine Learning Street Talk · Machine Learning Street Talk (MLST)
Length
16 min
Companies
xAI
Products
Grok-3
Summary by
Subvolts, using an AI model (how we work). Spotted a mistake? Tell us.
#Benchmarks#Safety#Research

Best of Machine Learning Street Talk

More on xAI this week

Questions people ask

Why are current AI benchmarks insufficient?
Technical benchmarks often ignore the nuances of human communication and adaptability, failing to measure if models are helpful, safe, and relatable for real people.
What are the risks of current AI safety oversight?
There is an alarming lack of oversight and a thin veneer of safety training, especially as users turn to AI for sensitive topics like mental health.

Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Machine Learning Street Talk (MLST).