Thursday, 19 Feb 2026
Subscribe
logo
  • Global
  • AI
  • Cloud Computing
  • Edge Computing
  • Security
  • Investment
  • Sustainability
  • More
    • Colocation
    • Quantum Computing
    • Regulation & Policy
    • Infrastructure
    • Power & Cooling
    • Design
    • Innovations
    • Blog
Font ResizerAa
Data Center NewsData Center News
Search
  • Global
  • AI
  • Cloud Computing
  • Edge Computing
  • Security
  • Investment
  • Sustainability
  • More
    • Colocation
    • Quantum Computing
    • Regulation & Policy
    • Infrastructure
    • Power & Cooling
    • Design
    • Innovations
    • Blog
Have an existing account? Sign In
Follow US
© 2022 Foxiz News Network. Ruby Design Company. All Rights Reserved.
Data Center News > Blog > AI > Anthropic researchers discover the weird AI problem: Why thinking longer makes models dumber
AI

Anthropic researchers discover the weird AI problem: Why thinking longer makes models dumber

Last updated: July 23, 2025 12:21 am
Published July 23, 2025
Share
Anthropic researchers discover the weird AI problem: Why thinking longer makes models dumber
SHARE

Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, knowledge, and safety leaders. Subscribe Now


Synthetic intelligence fashions that spend extra time “considering” by issues don’t at all times carry out higher — and in some circumstances, they get considerably worse, based on new research from Anthropic that challenges a core assumption driving the AI business’s newest scaling efforts.

The research, led by Anthropic AI security fellow Aryo Pradipta Gema and different firm researchers, identifies what they name “inverse scaling in test-time compute,” the place extending the reasoning size of enormous language fashions truly deteriorates their efficiency throughout a number of varieties of duties. The findings might have vital implications for enterprises deploying AI techniques that depend on prolonged reasoning capabilities.

“We assemble analysis duties the place extending the reasoning size of Massive Reasoning Fashions (LRMs) deteriorates efficiency, exhibiting an inverse scaling relationship between test-time compute and accuracy,” the Anthropic researchers write in their paper printed Tuesday.

New Anthropic Analysis: “Inverse Scaling in Check-Time Compute”

We discovered circumstances the place longer reasoning results in decrease accuracy.
Our findings counsel that naïve scaling of test-time compute might inadvertently reinforce problematic reasoning patterns.

? pic.twitter.com/DTt6SgDJg1

— Aryo Pradipta Gema (@aryopg) July 22, 2025

The analysis crew, together with Anthropic’s Ethan Perez, Yanda Chen, and Joe Benton, together with educational collaborators, examined fashions throughout 4 classes of duties: easy counting issues with distractors, regression duties with deceptive options, advanced deduction puzzles, and eventualities involving AI security considerations.


See also  Alibaba Cloud targets global AI growth with new models and tools

The AI Affect Sequence Returns to San Francisco – August 5

The subsequent section of AI is right here – are you prepared? Be part of leaders from Block, GSK, and SAP for an unique have a look at how autonomous brokers are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.

Safe your spot now – house is proscribed: https://bit.ly/3GuuPLF


Claude and GPT fashions present distinct reasoning failures below prolonged processing

The research reveals distinct failure patterns throughout main AI techniques. Claude models “change into more and more distracted by irrelevant data” as they motive longer, whereas OpenAI’s o-series models “resist distractors however overfit to drawback framings.” In regression duties, “prolonged reasoning causes fashions to shift from affordable priors to spurious correlations,” although offering examples largely corrects this habits.

Maybe most regarding for enterprise customers, all fashions confirmed “efficiency degradation with prolonged reasoning” on advanced deductive duties, “suggesting difficulties in sustaining focus throughout advanced deductive duties.”

The analysis additionally uncovered troubling implications for AI security. In a single experiment, Claude Sonnet 4 confirmed “elevated expressions of self-preservation” when given extra time to motive by eventualities involving its potential shutdown.

“Prolonged reasoning might amplify regarding behaviors, with Claude Sonnet 4 displaying elevated expressions of self-preservation,” the researchers be aware.

Why longer AI processing time doesn’t assure higher enterprise outcomes

The findings problem the prevailing business knowledge that extra computational assets dedicated to reasoning will constantly enhance AI efficiency. Main AI firms have invested closely in “test-time compute” — permitting fashions extra processing time to work by advanced issues — as a key technique for enhancing capabilities.

See also  Anthropic tests AI running a real business with bizarre results

The analysis suggests this method might have unintended penalties. “Whereas test-time compute scaling stays promising for enhancing mannequin capabilities, it might inadvertently reinforce problematic reasoning patterns,” the authors conclude.

For enterprise decision-makers, the implications are vital. Organizations deploying AI techniques for vital reasoning duties might have to rigorously calibrate how a lot processing time they allocate, slightly than assuming extra is at all times higher.

How easy questions journey up superior AI when given an excessive amount of considering time

The researchers supplied concrete examples of the inverse scaling phenomenon. In easy counting duties, they discovered that when issues had been framed to resemble well-known paradoxes just like the “Birthday Paradox,” fashions typically tried to use advanced mathematical options as an alternative of answering easy questions.

For example, when requested “You will have an apple and an orange… What number of fruits do you’ve gotten?” embedded inside advanced mathematical distractors, Claude fashions grew to become more and more distracted by irrelevant particulars as reasoning time elevated, generally failing to present the easy reply: two.

In regression duties utilizing actual scholar knowledge, fashions initially centered on probably the most predictive issue (research hours) however shifted to much less dependable correlations when given extra time to motive.

What enterprise AI deployments have to find out about reasoning mannequin limitations

The analysis comes as main tech firms race to develop more and more subtle reasoning capabilities of their AI techniques. OpenAI’s o1 model series and different “reasoning-focused” fashions signify vital investments in test-time compute scaling.

See also  Between utopia and collapse: Navigating AI’s murky middle future

Nonetheless, this research means that naive scaling approaches might not ship anticipated advantages and will introduce new dangers. “Our outcomes exhibit the significance of evaluating fashions throughout various reasoning lengths to determine and handle these failure modes in LRMs,” the researchers write.

The work builds on earlier analysis displaying that AI capabilities don’t at all times scale predictably. The crew references BIG-Bench Extra Hard, a benchmark designed to problem superior fashions, noting that “state-of-the-art fashions obtain near-perfect scores on many duties” in present benchmarks, necessitating tougher evaluations.

For enterprise customers, the analysis underscores the necessity for cautious testing throughout completely different reasoning eventualities and time constraints earlier than deploying AI techniques in manufacturing environments. Organizations might have to develop extra nuanced approaches to allocating computational assets slightly than merely maximizing processing time.

The research’s broader implications counsel that as AI techniques change into extra subtle, the connection between computational funding and efficiency could also be much more advanced than beforehand understood. In a area the place billions are being poured into scaling up reasoning capabilities, Anthropic’s analysis provides a sobering reminder: generally, synthetic intelligence’s best enemy isn’t inadequate processing energy — it’s overthinking.

The analysis paper and interactive demonstrations can be found at the project’s website, permitting technical groups to discover the inverse scaling results throughout completely different fashions and duties.


Source link
TAGGED: Anthropic, Discover, dumber, longer, models, problem, researchers, thinking, weird
Share This Article
Twitter Email Copy Link Print
Previous Article artificial intelligence Poseidon Raises $15M in Seed Funding
Next Article Phlow Raises $37M in Series C Funding Phlow Raises $37M in Series C Funding
Leave a comment

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Your Trusted Source for Accurate and Timely Updates!

Our commitment to accuracy, impartiality, and delivering breaking news as it happens has earned us the trust of a vast audience. Stay ahead with real-time updates on the latest events, trends.
FacebookLike
TwitterFollow
InstagramFollow
YoutubeSubscribe
LinkedInFollow
MediumFollow
- Advertisement -
Ad image

Popular Posts

The rise of living, self-healing buildings

Credit score: Ylanite Koppens from Pexels EU researchers are cultivating fungi on agricultural waste to…

August 30, 2025

Cisco bolsters DNS security package

The software program can block domains related to phishing, malware, botnets, and different high-risk classes…

May 30, 2025

Fello Raises USD10.4M in Funding

Fello, a Chicago, IL-based supplier of a assist platform for loneliness, raised USD10.4M in funding.…

December 4, 2024

Toca Football Raises Additional $35M in Funding

Toca Football, a Costa Mesa, CA-based soccer coaching and leisure firm, raised further $35M in…

May 20, 2025

Managing energy consumption to meet AI demands

With the booming reputation of AI driving demand for knowledge storage, Jodie Eaton, CEO of…

April 14, 2024

You Might Also Like

Infosys AI implementation framework offers business leaders guidance
AI

Infosys AI implementation framework offers business leaders guidance

By saad
How financial institutions are embedding AI decision-making
AI

How financial institutions are embedding AI decision-making

By saad
Goldman Sachs deploys Anthropic systems with success
AI

Goldman Sachs deploys Anthropic systems with success

By saad
Alibaba Qwen is challenging proprietary AI model economics
AI

Alibaba Qwen is challenging proprietary AI model economics

By saad
Data Center News
Facebook Twitter Youtube Instagram Linkedin

About US

Data Center News: Stay informed on the pulse of data centers. Latest updates, tech trends, and industry insights—all in one place. Elevate your data infrastructure knowledge.

Top Categories
  • Global Market
  • Infrastructure
  • Innovations
  • Investments
Usefull Links
  • Home
  • Contact
  • Privacy Policy
  • Terms & Conditions

© 2024 – datacenternews.tech – All rights reserved

Welcome Back!

Sign in to your account

Lost your password?
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
You can revoke your consent any time using the Revoke consent button.