Recommended Posts

 

Is it just me or the results of Perplexity and ChatGPT seem less smart. Not being able to answer to some basic questions or do analysis where they need to get data from multiple sources before compiling the results. The context window seems to be smaller. 

I do not recall that being the case. I notice a distinct degradation. I wonder if they’re “throttling” their compute to save cost.

  • Like 1
  • Love 1
Link to comment
https://www.neowin.net/forum/topic/1452803-is-ai-getting-dumber/
Share on other sites

Study: AI Is Making Us All Dumber

Well, it’s official: An increased reliance on artificial intelligence has been linked to diminished critical thinking abilities, a new study has found.

https://tech.co/news/study-ai-making-us-dumber#:~:text=According to a new study,workout that helps people learn.&text=Well%2C it's official%3A An increased,a new study has found.

On 20/03/2025 at 05:02, thexfile said:

Study: AI Is Making Us All Dumber

Well, it’s official: An increased reliance on artificial intelligence has been linked to diminished critical thinking abilities, a new study has found.

https://tech.co/news/study-ai-making-us-dumber#:~:text=According to a new study,workout that helps people learn.&text=Well%2C it's official%3A An increased,a new study has found.

I definitely believe that. People around me who would work on solving problems for hours a day now just get it done in a few minutes. No brain cells burned. 

On 19/03/2025 at 23:54, FMH said:

 

Is it just me or the results of Perplexity and ChatGPT seem less smart. Not being able to answer to some basic questions or do analysis where they need to get data from multiple sources before compiling the results. The context window seems to be smaller. 

I do not recall that being the case. I notice a distinct degradation. I wonder if they’re “throttling” their compute to save cost.

More that their results are being polluted by a rapidly increasing amount of stupid on the internet...

On 20/03/2025 at 00:40, FMH said:

I definitely believe that. People around me who would work on solving problems for hours a day now just get it done in a few minutes. No brain cells burned. 

I work for a large international software company and though we develop a few systems ourselves, the USE of external AI systems such as ChatGPT has been banned throughout the entire company...

So, a few things I have an opinion on here:

Is AI getting dumber: The LLMs now have massively more data to build upon and there are certainly more hallucinations than previously.  As a consumer of GenAI, then I feel it's valid to feedback to it when it's wrong, as it learns from this.  This was demonstrated and proven.  

But people are (generally) not feeding back, so it is significantly hampered in it's ability to learn from what is a good and bad result.

Is AI making people dumber: Some yes, some no.  I use GenAI frequently, and it has not made me "dumber", if anything it has improved the rate at which I digest information.  But again, I use it "correctly" and understand it is a work in progress and both feedback and analyse it's output.

I liken it to when GPS devices started appearing on people's dashboards, the early TomTom devices - they begat poorer drives with respect to lane discipline.  I am from an era of looking at a map, planning my route, knowing where I am going before I get there, but when people just started relying on TomToms, they would change lane at the last second because the device hadn't prepped them to turn left at the roundabout until it hit the roundabout.

We use GenAI significantly within our workspace, trained on our own data, and restricted to our own uses.  We have a robust policy about the use of CoPilot and others (e.g. ChatGPT).  I'm currently looking into Agentic AI  to go beyond basic automation of jobs, but as ever - I always check what it is wanting to do with data, how it uses it, where it's taking it and what it outputs.

  • Like 1

basically getting dumber... initial models were trained on approved and collected sources of data...... then they add more and more to it and then start harvesting crap like reddit for info then it all just goes to heck because its trained not smart..... it doesn't know what is real or not just keeps telling itself what i see is real so start merging that into what i already know

On 20/03/2025 at 13:22, neufuse said:

basically getting dumber... initial models were trained on approved and collected sources of data...... then they add more and more to it and then start harvesting crap like reddit for info then it all just goes to heck because its trained not smart..... it doesn't know what is real or not just keeps telling itself what i see is real so start merging that into what i already know

EXACTLY.  Unless you feed back that the output is tosh, it basically says to itself "Yeah, that looks right" and ups it in its preferred responses.

On 20/03/2025 at 13:26, Dick Montage said:

EXACTLY.  Unless you feed back that the output is tosh, it basically says to itself "Yeah, that looks right" and ups it in its preferred responses.

I remember a while ago I asked ChatGPT to write a simple bubble sort function in C# just as a simple test.  It didn't work.

 

So I did a quick search on Stack Overflow and found the EXACT same code, with the same bug.  So not only did it return buggy code, but it stole it from Stack Overflow without any kind of citation.  Double fail! :p

 

I’ve had it write code which I then noticed was using a deprecated function. I pointed this out and it agreed and provided more up to date code.

So… why give me the first code? Idiot AI

On 19/03/2025 at 19:54, FMH said:

 

Is it just me or the results of Perplexity and ChatGPT seem less smart. Not being able to answer to some basic questions or do analysis where they need to get data from multiple sources before compiling the results. The context window seems to be smaller. 

I do not recall that being the case. I notice a distinct degradation. I wonder if they’re “throttling” their compute to save cost.

Absolutely AI is getting dumber.  They're using social media sites like facebook, instagram, and reddit.  Social media is proof that not everyone's opinion is important.

  • Love 2
On 21/03/2025 at 11:45, Tomo said:

AI has never been smart, there is no intelligence whatsoever. It draws upon data that is already available on the Internet, whether that information is accurate or not.

Facile interpretation at best.

On 24/03/2025 at 10:30, Tomo said:

A more complex answer isn't required and would take awhile.

But provide more value.  All intelligence is based around pre-existing information and extrapolation - arguably we would point towards Wolfram Alpha as a perfect example of this.  Was it "smart"?  By who's definition?  I'd argue yes, because it wasn't just simple 1:1 answers, but rather taking the steps of abstraction away from the input question towards the relevancy between information sources and presenting an answer back - extrapolation.  Saying AI has never been "smart" and that it's just parroting it back shows a lack of understanding of what AI has been for decades.  Now with generative AI, yes - it still presents back answers based upon pre-existing data, but it's able to (when given opportunity) learn from that and apply that learning to other related cases.  That differentiates it greatly from parroting back information.

If a user consumes these responses without understanding how to tailor them, without giving the engine the ability to learn, and then writes it off as "not smart" - that speaks more towards a user than a model.  It may take a while (it's two words) to teach users.

  • Like 2

AI also trains on uninformed and incorrect data, think reddit and popular opinion with only a few corrective and downvoted replies, and then you have your answer.

A good AI is one that can distinguish from it but it is like people who are unable to inform themselves properly and then make the wrong uninformed decision.

Don't know what type of AI  needs you guys are using. I'm using based on services, specifically insurance in mexico with law fundamentals and it's been turned out very well.
Also one of my employees uses it for marketing and also has giving very good results.

In no one way should it be used instead of critical thinking. However if people use AI instead of their brains, they can start saying good bye to their jobs. 
Which overall that's a good thing. As a consumer and cliente of certain services, what I don't want is an air head answering my inquiries.

I have 4 employees and they all use it as guide. I'm responsable in having them in constant training so they don't rot their brain due to lack of use.

  • Like 1
  • Love 1
On 25/03/2025 at 00:04, freedonX said:

Don't know what type of AI  needs you guys are using. I'm using based on services, specifically insurance in mexico with law fundamentals and it's been turned out very well.
Also one of my employees uses it for marketing and also has giving very good results.

In no one way should it be used instead of critical thinking. However if people use AI instead of their brains, they can start saying good bye to their jobs. 
Which overall that's a good thing. As a consumer and cliente of certain services, what I don't want is an air head answering my inquiries.

I have 4 employees and they all use it as guide. I'm responsable in having them in constant training so they don't rot their brain due to lack of use.

You are using a well trained model, which has been provided valid data to work from - and I assume then further consumes the results of whatever it is being used for (which is another form of training that validates it's earlier results).  As such, I absolutely would expect much better results.  I'm doing similar using financial systems to forecast and then validate off the later bills - it's been flawless in this regard.

  • 8 months later...

I’ve noticed similar behavior. The models aren’t getting “dumber”; updates just adjust the balance of speed, safety, and cost, which can affect thoroughness in multi-step reasoning. Context handling varies too, leading to shorter or less detailed answers. It’s more fine-tuning than a capability drop.

  • Like 3
  • 4 weeks later...
On 22/12/2025 at 03:58, Sigiand said:

I think the feeling that ‘AI is getting dumber’ is often linked to expectations. When technology is new, it amazes us. Then we start using it every day and notice mistakes more often.

I recently searched for some info. on my new stove. The AI overview gave instructions on a microwave. I agree AI is dumb.

  • 3 weeks later...
On 04/01/2026 at 15:43, alexcc said:

You’re not alone in noticing this, but it’s likely not that the models are getting “dumber.” What’s changing is how they’re deployed.

A few factors at play:

Platforms often route users to different model variants depending on load, cost, or plan.

Safety and reliability filters can make responses feel more conservative or less exploratory.

Some tools optimize for speed over depth, which affects multi-step analysis.

Context handling can vary between sessions and interfaces.

The core models continue to improve, but real-world performance depends heavily on product decisions, not just model intelligence.

Actually, a large part of the problem is AI slop itself. As more and more of it is generated, much if it completely false, it gets fed back into the models, polluting the data and increasing the prevalence of "hallucinations".  It's a self-destructive feedback loop that will eventually make them unusable unless a way is found to exclude it.

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • OpenAI is rolling out a major upgrade to ChatGPT memory by Pradeep Viswanathan OpenAI is rolling out a major upgrade to ChatGPT's memory, making the system more capable, current, and scalable across long-term use. Memory allows ChatGPT to remember useful details about users, including their preferences, projects, and constraints. Instead of starting every conversation from scratch, ChatGPT can use this context to provide more relevant responses in future chats. OpenAI first launched saved memories in February 2024. That feature allowed users to explicitly ask ChatGPT to save information into its memory, such as travel plans or writing preferences. However, this system had limits because it depended heavily on users giving clear instructions to remember something. Additionally, saved memories could become stale over time. In April 2025, OpenAI expanded memory by allowing ChatGPT to reference past chat context outside the saved memories list. This was powered by a background process called “dreaming,” which automatically curates memories from chat history. This made ChatGPT better at learning from natural conversation without requiring users to manually save every detail. Today, OpenAI announced a more capable and compute-efficient memory architecture built on top of dreaming. This new system improves ChatGPT’s ability to carry forward useful context, follow user preferences, and remain accurate as time passes. According to OpenAI’s internal evaluations, the new system improves factual recall from 67.9% in 2025 to 82.8% in 2026. Preference adherence improves from 55.3% to 71.3%, while accuracy over time improves from 52.2% to 75.1%. The best part of this new system is a new memory summary page where users can review ChatGPT's memories. Users can even update details, correct information, or give instructions on what topics ChatGPT should bring up and when. This new, improved memory system is available to ChatGPT Plus and Pro users in the US starting today. It will roll out to more countries, as well as Free and Go users, in the coming weeks.
    • I work for a video production company in Australia. The camera operators shoot footage and then pass the SD card over to the editors. Much easier than handing over the entire camera. Plus, on a busy day you can hand off the SD card and then pop another in for the next shoot. Or, you might have used multiple SD cards because you need the extra space for a long shoot. I also use USB cables and wifi for transferring footage, but in many cases an SD card reader is the easiest method.
    • Microsoft Edge 149.0.4022.52 by Razvan Serea Microsoft Edge is a super fast and secure web browser from Microsoft. It works on almost any device, including PCs, iPhones and Androids. It keeps you safe online, protects your privacy, and lets you browse the web quickly. You can even use it on all your devices and keep your browsing history and favorites synced up. Built on the same technology as Chrome, Microsoft Edge has additional built-in features like Startup boost and Sleeping tabs, which boost your browsing experience with world class performance and speed that are optimized to work best with Windows. Microsoft Edge security and privacy features such as Microsoft Defender SmartScreen, Password Monitor, InPrivate search, and Kids Mode help keep you and your loved ones protected and secure online. Microsoft Edge has features to keep both you and your family protected. Enable content filters and access activity reports with your Microsoft Family Safety account and experience a kid-friendly web with Kids Mode. The new Microsoft Edge is now compatible with your favorite extensions, so it’s easy to personalize your browsing experience. Microsoft Edge 149.0.4022.52 changelog: Migration to improved V2 architecture for Workspaces. Workspaces, introduced in Edge in 2022, allows users to create durable sets of tabs that can be saved and shared with others. In order to improve reliability and performance of this feature, the following changes are being made: Migrating data for saved Workspaces from OneDrive/SharePoint to Edge Sync service Removing the collaboration/share functionality of this feature For organizations who have disabled Sync through policy, the existing v1 Workspace data will still be migrated to the new architecture. New v2 Workspaces created after migration won't sync across devices and will remain local to each device. This update occurs on a progressive rollout beginning in Edge Stable v145 and will continue rolling out in Edge v149. For more information, see Getting started with Microsoft Edge Workspaces. Feature Updates Passkey Sync for Enterprise Users. Microsoft Edge is introducing support for passkey synchronization for enterprise users, enabling secure, passwordless authentication across devices. Passkeys created in Edge can now be synced seamlessly, improving sign-in experience while maintaining strong security standards. Note: This is a controlled feature rollout. If you don't see this change, check back as we continue the rollout. Enterprise WebView2 runtime downgrade via DowngradeVersion policy. Administrators can temporarily roll back specific applications to a previous WebView2 Evergreen Runtime version (N-1 or N-2) using the new DowngradeVersion policy in msedgewebview2.admx. The Downgrade Version policy allows enterprises to mitigate critical regressions by specifying per-application exe-to-version mappings. The Edge Updater installs the target version side-by-side, and the WebView2 Loader redirects targeted apps accordingly. Downgrades auto-expire with each new WebView2 release: apps pinned to N-1 remain on the same version (now becoming N-2) and will auto-update in the next release, while apps pinned to N-2 will revert to the current Evergreen version. The policy applies only to enterprise-managed devices (domain-joined or MDM-enrolled). For more information, see Microsoft Edge WebView2 Policy Documentation | Microsoft Learn. Collections retirement. Collections has been removed in this update. Users can no longer access or use the feature. To keep saved content, users can export it, or move all pages to Favorites before updating to Microsoft Edge Stable 149. For more information, see Organize your ideas with Collections in Microsoft Edge - Microsoft Support. Modern, unified, and updated Look and Feel. Microsoft Edge has updated the Look and Feel to give customers a unified experience across all of Microsoft AI surfaces including Copilot and Bing. This changes multiple elements of the UX such as spacing, corners, fonts, default colors, etc. Clarify choices surrounding third-party cookie settings. Language under Settings > Privacy, search, and services > Cookies are clarified to better describe the choices users have in managing third-party cookies. Custom primary password retirement. Users are no longer able to create a new custom primary password in Edge Settings edge://settings/autofill/passwords/settings. Any users who are still using a custom primary password will be automatically migrated to device authentication. Additionally, the PrimaryPasswordSetting policy will no longer support the WithCustomPrimaryPassword option. For more information, see Keep your saved passwords private in Microsoft Edge | Microsoft Support. Unifying Copilot Chat policy controls. The Microsoft365CopilotChatIconEnabled policy is the standard for configuring Copilot Chat. Previously, this behavior was controlled by blocking the Copilot extension, either explicitly or by using the * wildcard via the ExtensionSettings or ExtensionInstallBlockList policies. Extension and sidebar policies no longer affect the appearance or functionality of Copilot Chat. Copilot address bar suggestions were also tied to extension policy settings. Starting in Microsoft Edge version 149, admins can use the CopilotAddressBarSuggestionsEnabled policy to manage this behavior. Intune MAM Protected Downloads. The protected downloads feature for Intune MAM is now available for BYOD (Bring Your Own Device) devices, which aren't managed by a tenant. Policy Updates / New policies CopilotAddressBarSuggestionsEnabled - Enable Copilot address bar suggestions CpuPerformanceTierOverride - Override for the CPU performance tier DataUrlInWebWorkerOpaqueOriginEnabled - Enable opaque origins for data URLs in Web Workers DefaultLocalFontsSetting - Default Local Fonts permission setting ForceForegroundPriorityForUrls - Force foreground priority for specific URLs LocalFontsAllowedForUrls - Allow Local Fonts permission on these sites LocalFontsBlockedForUrls - Block Local Fonts permission on these sites Deprecated policies WalletDonationEnabled - Wallet Donation Enabled (deprecated) EdgeWalletEtreeEnabled - Edge Wallet E-Tree Enabled (deprecated) Additional policy changes ForceForegroundPriorityForUrls - ForceForegroundPriorityForOrigins is renamed to ForceForegroundPriorityForUrls OnSecurityEventEnterpriseConnector - Add macOS platform support ProtectedContentIdentifiersAllowed - Remove macOS platform support Download: Microsoft Edge (64-bit) | 193.0 MB (Freeware) Download: Microsoft Edge (32-bit) | 170.0 MB Download: Microsoft Edge (ARM64) | 188.0 MB View: Microsoft Edge Website | Release History Get alerted to all of our Software updates on Twitter at @NeowinSoftware
  • Popular Contributors

    1. 1
      +primortal
      471
    2. 2
      PsYcHoKiLLa
      247
    3. 3
      Skyfrog
      79
    4. 4
      FloatingFatMan
      67
    5. 5
      Michael Scrip
      59
  • Tell a friend

    Love Neowin? Tell a friend!