Close Menu
TechTost
  • AI
  • Apps
  • Crypto
  • Fintech
  • Hardware
  • Media & Entertainment
  • Security
  • Startups
  • Transportation
  • Venture
  • Recommended Essentials
What's Hot

Cloud AI startup Runpod hits $120M in ARR — and it started with a Reddit post

Chinese electric vehicles are closing in on the US as Canada slashes tariffs

From OpenAI offices to Eli Lilly deal – how Chai Discovery became one of the most impressive names in AI drug development

Facebook X (Twitter) Instagram
  • About Us
  • Contact Us
  • Privacy Policy
  • Terms and Conditions
  • Disclaimer
Facebook X (Twitter) Instagram
TechTost
Subscribe Now
  • AI

    From OpenAI offices to Eli Lilly deal – how Chai Discovery became one of the most impressive names in AI drug development

    16 January 2026

    Anthropic taps former Microsoft India Director to lead Bengaluru expansion

    16 January 2026

    Taiwan to invest $250 billion in US semiconductor manufacturing

    15 January 2026

    Mira Murati’s startup Thinking Machines Lab is losing two of its co-founders to OpenAI

    15 January 2026

    Musk denies knowledge of underage Grok sex images as California AG begins investigation

    14 January 2026
  • Apps

    TikTok is quietly launching a micro-drama app called ‘PineDrama’

    16 January 2026

    Google’s Trends Explore page gets new Gemini features

    16 January 2026

    After Italy, WhatsApp exempts Brazil from rival chatbot ban

    15 January 2026

    App downloads decline again in 2025, but consumer spending jumps to nearly $156 billion

    15 January 2026

    Netflix’s first original video podcasts feature Pete Davidson and Michael Irvin

    14 January 2026
  • Crypto

    Hackers stole over $2.7 billion in crypto in 2025, data shows

    23 December 2025

    New report examines how David Sachs may benefit from Trump administration role

    1 December 2025

    Why Benchmark Made a Rare Crypto Bet on Trading App Fomo, with $17M Series A

    6 November 2025

    Solana co-founder Anatoly Yakovenko is a big fan of agentic coding

    30 October 2025

    MoviePass opens Mogul fantasy league game to the public

    29 October 2025
  • Fintech

    Fintech firm Betterment confirms data breach after hackers sent fake crypto scam alert to users

    12 January 2026

    Flutterwave buys Nigeria’s Mono in rare African fintech exit

    5 January 2026

    Even as global crop prices fall, India’s Arya.ag attracts investors – and remains profitable

    2 January 2026

    These 21-year-old school dropouts raise $2 million to launch Givefront, a fintech for nonprofits

    18 December 2025

    Google deepens consumer loyalty drive in India with UPI-linked card

    17 December 2025
  • Hardware

    US slaps 25% tariffs on Nvidia’s H200 AI chips headed to China

    15 January 2026

    The weirdest tech announced at CES 2026

    15 January 2026

    Google’s Gemini will power Apple’s AI features like Siri

    14 January 2026

    Pebble founder says his new company ‘isn’t a startup’

    14 January 2026

    The ring founder details the era of the camera company’s “smart assistants.”

    13 January 2026
  • Media & Entertainment

    YouTube relaxes monetization guidelines for some controversial topics

    16 January 2026

    Bandcamp takes a stand against AI music, banning it from the platform

    15 January 2026

    Paramount filed a lawsuit against Warner Bros. amid the controversial Netflix merger

    13 January 2026

    Netflix had a huge night at the 2026 Golden Globes with 7 wins

    12 January 2026

    Spotify lowers monetization limit for video podcasts

    8 January 2026
  • Security

    Iran’s internet shutdown is now one of the longest as protests continue

    16 January 2026

    AI security company depthfirst announces $40M Series A

    14 January 2026

    Man pleads guilty to hacking US Supreme Court filing system

    14 January 2026

    Internet crashes in Iran amid protests over financial crisis

    9 January 2026

    Critics scrutinize spyware maker NSO’s transparency claims amid push to enter US market

    9 January 2026
  • Startups

    Cloud AI startup Runpod hits $120M in ARR — and it started with a Reddit post

    16 January 2026

    Parloa triples valuation in 8 months to $3 billion with $350 million raise

    16 January 2026

    AI video startup Higgsfield, founded by ex-Snap exec, valued at $1.3 billion

    15 January 2026

    India’s Emversity Doubles Valuation as It Scales Workers AI Can’t Replace

    15 January 2026

    Digg is launching its new rival Reddit to the public

    14 January 2026
  • Transportation

    Chinese electric vehicles are closing in on the US as Canada slashes tariffs

    16 January 2026

    Tesla will only offer subscriptions for full self-driving (Supervision) in the future.

    15 January 2026

    The FTC’s data-sharing order against GM was finally settled

    15 January 2026

    The American cargo technology company has publicly exposed its shipping systems and customer data on the web

    14 January 2026

    New York’s governor paves the way for robotaxis everywhere, with one notable exception

    13 January 2026
  • Venture

    Tiger Global loses India tax case linked to Walmart-Flipkart deal in blow to offshore playbook

    15 January 2026

    The super-organization is raising $25 million to support biodiversity startups

    13 January 2026

    These Gen Zers just raised $11.75 million to put Africa’s defense back in the hands of Africans

    12 January 2026

    The venture firm that ate up Silicon Valley just raised another $15 billion

    9 January 2026

    Why This VC Thinks 2026 Will Be ‘The Year of the Consumer’

    8 January 2026
  • Recommended Essentials
TechTost
You are at:Home»AI»Why most AI benchmarks tell us so little
AI

Why most AI benchmarks tell us so little

techtost.comBy techtost.com8 March 202405 Mins Read
Share Facebook Twitter Pinterest LinkedIn Tumblr Email
Why Most Ai Benchmarks Tell Us So Little
Share
Facebook Twitter LinkedIn Pinterest Email

On Tuesday, startup Anthropic released a family of AI models that it claims achieve best-in-class performance. Just days later, rival Inflection AI unveiled a model that it claims comes close to matching some of the most capable models out there, including OpenAI’s GPT-4, in quality.

Anthropic and Inflection are by no means the first AI companies to claim that their models have matched or beaten the competition by some objective measure. Google supported the same with its Gemini models at launch, and OpenAI said the same for GPT-4 and its predecessors, GPT-3, GPT-2, and GPT-1. The list goes on.

But what metrics are they talking about? When a seller says a model achieves top performance or quality, what exactly does that mean? Perhaps more to the point: Will a model that technically “performs” better than some other model in reality touch improved in a tangible way?

On that last question, not likely.

The reason – or rather the problem – lies in the benchmarks that AI companies use to quantify a model’s strengths and weaknesses.

Internal measures

Today’s most commonly used benchmarks for AI models — specifically chatbot-powered models such as OpenAI’s ChatGPT and Anthropic’s Claude — do a poor job of capturing how the average human interacts with the models being tested. For example, a benchmark cited by Anthropic in its recent announcement, GPQA (“A Graduate-Level Google-Proof Q&A Benchmark”), contains hundreds of PhD-level biology, physics, and chemistry questions — yet most people use chatbot for tasks like answering emails, writing cover letters and talking about their feelings.

Jesse Dodge, a scientist at the Allen Institute for AI, the nonprofit AI research organization, says the industry has reached a “crisis of evaluation.”

“Benchmarks are typically static and narrowly focused on evaluating a single capability, such as a model’s realism in a single domain or its ability to solve multiple-choice mathematical reasoning questions,” Dodge told TechCrunch in an interview. “Many benchmarks used for evaluation are more than three years old, from when AI systems were mainly used for research and did not have many real users. In addition, humans use genetic AI in many ways — they are very creative.”

Wrong measurements

It’s not that the most used benchmarks are completely useless. No doubt someone is asking Ph.D level math questions. in ChatGPT. However, as genetic AI models are increasingly positioned as mass-market, do-it-all systems, the old benchmarks are becoming less applicable.

David Widder, a postdoctoral researcher at Cornell who studies artificial intelligence and ethics, notes that many of the common tests of reference skills—from solving school-level math problems to determining whether a sentence contains an anachronism—will never be relevant to the majority of users.

“Earlier AI systems were often built to solve a specific problem in a context (e.g. medical AI expert systems), making a deep understanding of what constitutes good performance in that particular context more possible,” Widder said. at TechCrunch. “As systems are increasingly seen as ‘general purpose’, this is less possible, so we’re increasingly seeing a focus on testing models across a variety of benchmarks in different fields.”

Errors and other defects

In addition to misalignment with use cases, there are questions about whether some benchmarks are properly measuring what they are supposed to measure.

One analysis of HellaSwag, a test designed to assess common sense reasoning in models, found that over a third of the test questions contained typos and “stupid” writing. Somewhere else, MMLU (short for “Massive Multitask Language Understanding”), a benchmark highlighted by vendors such as Google, OpenAI and Anthropic as proof that their models can reason through logic problems, asks questions that can be solved through memorization verbatim.

Test questions from the HellaSwag benchmark.

“[Benchmarks like MMLU are] more about memorizing and associating two keywords together,” Widder said. “I can find [a relevant] article quickly enough and answer the question, but that doesn’t mean I understand the causal mechanism or that I could use my understanding of that causal mechanism to actually reason and solve new and complex problems in unpredictable contexts. Not even a model can.”

Fixing what’s broken

So benchmarks are broken. But can they be fixed?

Dodge believes so – with more human involvement.

“The right way forward, here, is a combination of evaluation benchmarks with human evaluation,” he said, “prompting a model with a real user question and then hiring a human to evaluate how good the response is.”

As for Widder, he’s less optimistic that benchmarks today — even with corrections for the most obvious mistakes, like typos — can be improved to the point where they would be informative to the vast majority of AI model users. Instead, he believes that tests of models should focus on the downstream effects of those models and whether the effects, good or bad, are seen as desirable by those affected.

“I would ask for what specific goals we want AI models to be able to be used for and assess whether they would be – or are – successful in such contexts,” he said. “And hopefully that process also includes evaluating whether we should be using AI in such contexts.”

All included benchmarks genAI Generative AI reference points Research
Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
Previous ArticleApple will ease the transition to Android by fall 2025
Next Article LLMs are ready to make logging business intelligence tools easier and faster to use
bhanuprakash.cg
techtost.com
  • Website

Related Posts

Cloud AI startup Runpod hits $120M in ARR — and it started with a Reddit post

16 January 2026

From OpenAI offices to Eli Lilly deal – how Chai Discovery became one of the most impressive names in AI drug development

16 January 2026

Anthropic taps former Microsoft India Director to lead Bengaluru expansion

16 January 2026
Add A Comment

Leave A Reply Cancel Reply

Don't Miss

Cloud AI startup Runpod hits $120M in ARR — and it started with a Reddit post

16 January 2026

Chinese electric vehicles are closing in on the US as Canada slashes tariffs

16 January 2026

From OpenAI offices to Eli Lilly deal – how Chai Discovery became one of the most impressive names in AI drug development

16 January 2026
Stay In Touch
  • Facebook
  • YouTube
  • TikTok
  • WhatsApp
  • Twitter
  • Instagram
Fintech

Fintech firm Betterment confirms data breach after hackers sent fake crypto scam alert to users

12 January 2026

Flutterwave buys Nigeria’s Mono in rare African fintech exit

5 January 2026

Even as global crop prices fall, India’s Arya.ag attracts investors – and remains profitable

2 January 2026
Startups

Cloud AI startup Runpod hits $120M in ARR — and it started with a Reddit post

Parloa triples valuation in 8 months to $3 billion with $350 million raise

AI video startup Higgsfield, founded by ex-Snap exec, valued at $1.3 billion

© 2026 TechTost. All Rights Reserved
  • About Us
  • Contact Us
  • Privacy Policy
  • Terms and Conditions
  • Disclaimer

Type above and press Enter to search. Press Esc to cancel.