New Search Options : ISE (posts), PDFs and misc (text) dropdowns, enter '/help' to learn more 🧾✨ Link to our Forum Charter: Read, Respect, Reflect Solicitation and Dealing of Drugs is Strictly Prohibited ! Please email smfadmin if you wish to advertise here Non drug topics are also very welcome ! All Terms and Conditions are at the Bottom of the Page
⚠️ [pdf=autodisplay] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
⚠️ [lmv=autodisplay-scroll] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
What we do here at Collabware is AI-Assisted Engineering but it's constant battle keeping AI on point because they tend to write insecure, buggy and sloppy code -- not to mention being inconsistent ! I take regular checkpoints and will start taking quality checkpoints from here on in.
This is for you if you're using the same AI to build and review your code. This is a follow-up to last week's cross-platform testing reel. Because every single platform has a training bias, and every model out there defaults to a pattern and has a blind spot — it cannot see in its own output. So the builders that get the best results know exactly which platform to match to which job. Here is what each platform catches that the others miss. Number one is Claude. It excels at adversarial reasoning, security reviews, and deep architectural analysis. It thinks like an attacker. It finds injection paths, authentication bypasses, and logic flaws like a pro. That, and the AI building platform will always defend itself — so Claude is the one to sic on it. If you built in Cursor, Lovable, or Bolt, I'd bring your security review to Claude and frame it as a penetration test. So direct your AI to run security-critical reviews on a platform with demonstrated adversarial depth. I think someone on here actually named their adversarial audit "The Murphy." That's a win. Step two is Codex and Gemini — strongest at catching implementation errors and reviewing code they did not write. So Codex reads your codebase cold and flags what does not belong. Gemini brings a giant context window that lets you hold an entire project in one view, and it'll spot patterns across files that a single-file reviewer will miss. So if you built in Claude Code, I'd take your logic verification to Codex or Gemini for a second opinion, with no attachment to the original implementation. So direct your AI to run a full codebase review on a platform that did not generate the code. And number three, Lovable, Bolt, and Cursor. They are super strong at full-stack builds and rapid prototyping. So if you built your backend in Claude Code, I'd hand the requirements to Lovable, Bolt and compare how a different platform interprets the same specs. Where the implementations differ is where your assumptions live — and the opportunity lives. So those differences always surface architecture decisions in your first platform, made silently for you. So direct your AI to rebuild one critical module on a platform and document the different approach it took. Same build, different eyes, better product every single time.
Number one is Claude. It excels at adversarial reasoning, security reviews, and deep architectural analysis. It thinks like an attacker. It finds injection paths, authentication bypasses, and logic flaws like a pro
[/b][/size]
Quote
Claude.ai says "Take the headline idea and leave the taxonomy behind — that's roughly my read of this.
The part that's genuinely true: a model reviewing its own output has a structural blind spot — it tends to re-derive the same reasoning that produced the bug in the first place, not spot it. You've actually seen this directly in this conversation: I caught real bugs in Gemini's drafts (the docstring duplication regression, the value-based magic-number exclusion bug, the invalid regex character range) that Gemini's own pass missed twice in a row. A different model, or even the same model coming at it cold in a fresh context, catches different things than the one that wrote it. That part isn't hype.
The part I'd push back on: the specific claims — "Claude thinks like an attacker," "Codex reads cold better," "Gemini's context window spots cross-file patterns better" — are stated as settled fact but are really just this one person's anecdotal read, dressed up with confident language ("finds injection paths like a pro," "every single time"). Model behavior on this stuff shifts release to release, isn't something anyone's actually benchmarked here, and "which model is better at X" claims age badly. I'd treat it as one person's working theory, not an established taxonomy.
Where it actually lands on your setup: you're already doing the underlying thing. Andrew (Architect/QA Lead), Claude (Lead Programmer), ChatGPT (Systems Analyst), Gemini (Relief Programmer) is already a cross-model review structure — Chatty reviewing Gemini's CONFIG-001/DOC-001 work and catching the "not the full spec" gap earlier in this thread is exactly the pattern this post is describing, you just built it before reading about it. The one adjustment I'd actually consider taking from this: for something like a security-critical review of the ISE search/auth surfaces specifically, deliberately handing it to whichever model didn't write that code, framed explicitly as "find a way to break this" rather than "does this look right" — that framing change (audit vs. review) tends to surface different findings regardless of which model does it.
Me, Chipper: NB: We do this now except for: (hey, what is) "Lovable, Bolt and compare ..." ?
AI: [md[ **Lovable** and **Bolt (Bolt.new)** are "prompt-to-app" builders — you describe an app in plain English and they generate a full working web app (frontend + backend + database + auth), not just a code snippet or a chat reply. Different category from Claude/ChatGPT/Gemini/Cursor entirely — those are assistants you direct; these are closer to autonomous scaffolding tools. [/md][/color]
Me, Chipper: so ive just wasted hours bc we are doing the best possible already ! In true H. Simpson parlance: "Doh !"
AI: Nah — I wouldn't call it wasted. You went in to check whether your setup had a gap, and it came back clean. That's a legitimate use of an hour, not a waste of one — the alternative is never checking and just assuming, which is worse.
And it wasn't a total no-op either: it gave you the actual language for what you're already doing ("cross-model adversarial review," directing critical stuff at whichever model didn't write it) — useful if that Master Post on Collabware's AI-assisted engineering workflow ever needs updating or explaining to someone else.
Anyway — Homer's right that it stings a little. But the correct read here is closer to "confirmed I'm not missing anything" than "doh."
---
# Reusable Prompt: Deep Code/Security Review for Andrew's SMF Mods
Use this as the standing prompt template for any future "review this package" request — to any AI, not just Claude — on drugs-and-users.org mod packages (BBCode mods, admin tools, standalone CLI scripts, etc.).
---
## Prompt template
See the attachments for the:
1. "Prompt Engineered" reusable Query and see it in action as we review our sample Mods, one of our premiere Mods: "The Liteweight Markdown Viewer" by using another one of our premier Mods: "listatt md - The Attachment Executive Handler" which is the recommended header bbcode for every new post, moving forward, at the top of the page.
2. The review of the sample mod.
3. The packaged sample mod LMV v4.0
4. The mandatory PDF transcript -- use your Browser's for this or ask AI ("now create pdf transcript on this conversation and do not miss a word").
5. The rebuilt mod based on any shortcomings (deviations that AI is notoriously known for).
6. Attach whatever else you need to the Master Post and always include the transcript(s) for future reference, archiving, indexing and searchability (such as using our ISE (Intelligent Search Engine -- our evolving classify, index and search ecosystem.
Societal Impacts - Enabling independent research on how people use Claude
Aug 26, 2026
Claude's summary:
📝 Inline Markdownv4.4 · 2026-09-05
Enabling Independent Research on How People Use Claude
Anthropic Societal Impacts post — Aug 26, 2026
The Problem This Addresses
Real-world AI usage data is concentrated inside a handful of labs
Outside researchers currently have only two bad options:
1. Rely on labs' own published analyses (answers the lab's questions, not theirs) 2. Use public datasets (fully independent, but skew toward casual/creative use — not representative of real usage)
Neither lets outsiders independently study how AI is actually being used
The Pilot
Ran this spring; three external institutions designed and ran their own
studies using Anthropic Insights (formerly called "Clio") — the internal privacy-preserving analysis tool Anthropic's own teams use
Each group analyzed ~250,000 Claude.ai / Claude Code conversations from
April–May 2026
Partners: Stanford SALT Lab, Oxford Human Information Processing Lab,
METR
Anthropic's review rights were contractually limited to: user privacy,
info that could help violate usage policies, Anthropic confidential info, and research accuracy — no say over the substance of findings
Partners are contractually free to publish results even if inconvenient
for Anthropic
Anthropic is releasing the aggregate data from each project publicly
Findings — Stanford SALT Lab (human-AI collaboration patterns)
Consequential work is common: prior research assumed people mostly hand
AI low-stakes tasks and keep consequential work (affecting others, hard to undo) for themselves. This study found over half of Claude conversations involved consequential delegated tasks — especially professional guidance around legal or financial questions
Humans stay in the driver's seat: in ~75% of conversations, the person
set direction and Claude assisted; people usually adapted Claude's output rather than using it verbatim - However, how much people understand or learn from Claude's output varies even when they're directing it
Friction is often productive: time spent noticing where Claude
misunderstood a request, and iterating, tends to improve outcomes — it pushes people to clarify intent and stay engaged rather than being purely a negative experience
Findings — Oxford Human Information Processing Lab (emotional states)
Studying how people feel during Claude use and how that links to Claude's behavior
Behavior-emotion pairings observed:
- Claude being warm ↔ people more positive - Claude refusing/disagreeing ↔ people pushing back - Claude being eccentric ↔ people more intellectually engaged - Claude simply being helpful ↔ people seeming satisfied
AI use resembles general web use: patterns of absorption, frustration,
and enjoyment in Claude conversations closely mirrored a separate study of ordinary internet browsing — suggesting AI interaction isn't emotionally distinct from other digital activity
Full writeup not yet published; link to be added later
Findings — METR (coding productivity across model generations)
Comparing Claude's own estimate of "how long would this have taken without AI"
against actual time taken with different Claude models
Newer models = bigger time savings: preliminary data shows significant
speedup from newer models vs. older ones
Claude is a decent judge of task time: its estimates correlated well
with known completion times from a prior human developer study (validating the method)
Next direction: extending this to estimate how much AI accelerates
research itself, seen as increasingly important as AI takes on more of its own development work
Analysis still underway; writeup pending
What Anthropic Learned Running the Pilot
Multiple perspectives on the same problem are valuable: METR's proposal
overlapped with Anthropic's internal economics research ("Agentic coding and persistent returns to expertise"); connecting the two teams improved both
Internal methods don't directly transfer externally:
- Anthropic Insights works by having Claude answer a structured question (e.g., "What type of guidance is this person asking for?") for every conversation, then aggregating into categories - Outputs are sensitive to question phrasing — badly worded questions misclassify conversations, and since no one can read underlying raw conversations, these errors are hard to catch - Internally, Anthropic fixes this by iterating on question wording over weeks — not feasible for external partners due to repeated privacy review overhead - Workaround: partners tested/refined their questions against WildChat (a public human-AI conversation dataset) where they could check answers against real conversations - Problem: WildChat skews casual/creative, unlike actual Claude traffic, so some questions that worked well on WildChat produced misleading results on real Claude data - Fix: Anthropic provided interpretation guidance for Insights outputs (in the appendix) and is exploring better upfront question-design methods
Balancing misuse transparency vs. enablement:
- Some output categories surfaced Acceptable Use Policy / Terms of Service violations (e.g., people seeking guidance on prohibited activities) - Anthropic shared most of these, believing the public should know about platform misuse - Exception: categories describing how people evaded safeguards (not just what they attempted) were withheld - Affected under 5% of categories/conversations in each study; researchers were told exactly what was altered/removed and why - Standard practice: when Insights surfaces such violations, aggregated data also goes to Anthropic's Safeguards team
Looking Forward
Anthropic frames oversight of AI's societal effects as "too big a job for
AI companies alone" — needs external researchers asking their own questions of real usage data and publishing independently
Pilot judged successful as a proof of concept: external orgs could run
independent studies without compromising user privacy
Open question: whether/how to scale the program — constrained by the
slowness/resource intensity described above, plus how many studies can run concurrently
Anthropic is "starting slowly" to protect privacy, safety, and research
quality, and is soliciting interest via an expression-of-interest form for researchers who want access
Appendix (full appendix available via a separate linked doc)
How the three partners were selected
The research primer explaining program goals and Anthropic Insights' capabilities
How each project moved from proposal → study design → analysis
Details of collaboration agreements (explicit publication freedom clause)
Third-party privacy audit conducted by Imperial College London
The privacy threat model applied to all released data
Guidance on interpreting the released Anthropic Insights data from each study
Earlier this year, we ran a pilot giving external researchers access to aggregate, real-world Claude usage data. Three research groups designed their own studies for Anthropic Insights, our privacy-preserving analysis tool; we ran the data collection on their behalf, and they conducted their own independent analysis. In this post, we share high-level results from those studies and what we learned running this pilot. We’re also providing an expression of interest form for researchers who may want to work with us in the future.
Ensuring the transition to transformative AI goes well requires understanding its impact on people and society. Right now, data on real-world interactions with AI is concentrated in a handful of labs. We think it would be good if more data was made widely available—to researchers, policymakers, and the general public.
Researchers outside the labs have two options. They can draw on analyses the labs publish, which reflect real usage but often answer the lab’s questions, rather than their own. Or they can use public datasets, which they can study however they like, but skew toward more casual use, and may not reflect how most people actually use AI. Neither is sufficient for independent research on how AI is actually being used.
This spring, we piloted a program in which three external research institutions designed and ran their own studies on Claude usage data through Anthropic Insights (formerly named ‘Clio’), the privacy-preserving tool our own teams use to analyze usage patterns across millions of Claude conversations. We hope to scale this program in the future, so we also conducted an additional privacy audit of all data shared with third-party researchers to verify that our privacy protections held (see Appendix).
We believe this is the first time external researchers have run public independent studies on an AI company's own usage data. Below, we discuss what the external teams found, what we learned running the pilot, and what we are weighing as we decide how to expand the program more widely. We are also publicly releasing the aggregate data from each project.
What the researchers learned:
We partnered with three research groups: the Social and Language Technologies (SALT) Lab at Stanford University, the Human Information Processing Lab at the University of Oxford, and METR, a non-profit organization that evaluates frontier AI models. Each group developed its own research questions and used Anthropic Insights to conduct privacy-preserving analysis of roughly 250,000 Claude.ai or Claude Code conversations from April-May 2026.
We wanted our external partners to have as much independence as possible, so our contractual review rights were limited to user privacy, information that could help people violate our usage policies, Anthropic’s confidential information, and research accuracy. Anthropic otherwise had no say in the content of the findings and the researchers are free to publish their results even if they are inconvenient for Anthropic. Below are some early results. We're excited about the directions, and about what others will find now that the data is public.
The Social and Language Technologies Lab studied how humans collaborate with AI. They looked at what types of work people bring to AI, what roles humans retain in completing that work, and where human-AI collaboration breaks down. They found:
People bring high-stakes work to AI more than expected. Prior research suggested people mostly delegate low-accountability tasks to AI and keep consequential tasks (that is, work that affects others or is hard to undo) for themselves. But the SALT Lab found that over half of Claude conversations involved people delegating consequential tasks to AI. People were most likely to bring consequential work to Claude when seeking professional guidance, particularly on legal or financial questions.
People usually direct and oversee the work when they collaborate with Claude. In nearly three-quarters of conversations, people set the direction while Claude assisted, and they usually adapted its output rather than using it verbatim. But even when directing Claude on the output they want, people vary in how much they understand and learn from what Claude produces. It is common for people to experience friction when collaborating with AI. However, that friction is often productive. The time and effort that people put into seeing how Claude attempts a task, identifying where the request was unclear or misunderstood, and iterating on their direction leads to better results–it pushes people to clarify their intent, refine the output, or stay engaged with the problem.
The Human Information Processing Lab is studying how people feel while using Claude and how that relates to Claude’s behavior.
Their early results indicate:
How people feel when using AI is linked to how AI behaves. The researchers found patterns of human and AI behavior appeared together in conversations: Claude being warm went together with people being more positive. Claude refusing or disagreeing went together with people pushing back. Claude being eccentric went together with people getting more intellectually engaged. And Claude simply helping went together with people seeming satisfied.
People’s experience when using AI looks a lot like it does on the rest of the web. The researchers found that the patterns among states like absorption, frustration, and enjoyment in Claude conversations closely resemble those in a separate study on everyday internet browsing, suggesting similarities in how people engage with AI and with other digital activity.
They are still completing their writeup. When it is public, we will add a link to it here.
METR is estimating real-world productivity gains from coding agents and how these increases in productivity change across model generations. Their analysis of Claude Code conversations is still underway, but early results suggest:
More capable models may save users more time. METR compared Claude’s guesses on how long tasks would have taken without AI to how long they actually took with different Claude models. Their preliminary findings indicate newer models deliver significant speedup over older models. METR plans on sharing more as their analysis develops.
AI can estimate time taken reasonably well. Because the analysis relies on Claude judging how long a task would take, METR compared those judgments to known completion times from a prior developer study. Claude's estimates correlated with the actual time taken by developers. Next: measure how much AI accelerates research. METR is continuing to investigate how their study can provide insight into how much AI speeds up researchers’ work, which could become increasingly important as AI takes on more of its own development.
They are still completing their writeup. When it is public, we will add a link to it here.
What our team learned:
Sharing usage data is largely unprecedented in AI, so this pilot was as much an experiment in running such a program as it was a way to enable third-party research in a privacy-preserving way. Protecting our users’ privacy and the researchers’ independence were both paramount, and we achieved both. Anthropic Insights is designed for this—researchers never accessed raw conversations, only aggregated outputs after the same legal and privacy review as our internal work. However, all of this made the pilot slow for an AI lab’s normal research speed and resource intensive to run. Both factors present a challenge to effectively scaling it. For more details on how we ran this pilot, see the Appendix. Below we discuss what we learned and how we addressed the challenges that arose.
It is valuable to pursue the same problem from different perspectives. Some of our partners’ research questions overlapped with work being pursued internally. For example, METR’s proposal was similar to our economics research on “Agentic coding and persistent returns to expertise.” We found this overlap valuable: it gave external researchers the chance to examine similar data and draw their own conclusions. Whether those align with ours is something we’ll follow as their study continues. We also connected METR with our Economics team and found that this connection improved both research teams’ work.
Research methods that work internally need to adapt for external partners. When using Anthropic Insights, a researcher writes a question such as, “What type of guidance is this person asking for?” and Claude answers it for every conversation in the study. The answers are then aggregated into categories; researchers only see final categories and the percentage of conversations that fall under each one. Because we are relying on Claude’s judgments, the tool is sensitive to a question’s wording; a poorly phrased one can place conversations into categories that misrepresent them. Because no one can read the underlying conversations, these errors are hard to catch.
Internally, we manage this by iterating on the questions many times over weeks. External partners couldn’t do that, since repeated privacy review before sharing each dataset would have made the study infeasible. Instead, we had them test their questions on WildChat, a public dataset of human-AI conversations where they could check the answers against the underlying conversations themselves. But WildChat skews toward casual and creative use, unlike Claude traffic, so some questions that performed well on WildChat produced misleading categories once applied to actual Claude conversations. We addressed this by providing guidance on how to interpret Anthropic Insight’s outputs (see Appendix). Going forward, we are exploring how external researchers can develop their questions and categories more effectively in advance.
Maintaining transparency about misuse without enabling it. Some categories in our partners’ Anthropic Insights outputs surfaced violations of our Acceptable Use Policy or Terms of Service—for instance a category of people seeking guidance on a prohibited activity. We think the public should know about misuse of our platform, so we shared most of these violations. The exceptions were categories that described how users got around our safeguards rather than what they attempted. Less than 5% of categories and conversations were affected in each study, and in each case we told researchers which clusters we had altered or removed and why. As a standard practice, when Anthropic Insights surfaces such violations, we share the aggregated data with our Safeguards team for their review. This is also an important process for our work with external researchers moving forward.
Looking forward:
Understanding AI’s effects on society is too big a job for AI companies alone. Real oversight needs external researchers asking their own questions of real-world usage data and publishing what they find independently.
This pilot was an experiment: could external researchers conduct independent studies on our platform without compromising our users’ privacy? The effort was more challenging than we expected, and we learned many lessons, but so far the answer seems to be yes. Our partners pursued research we would not have thought to design ourselves, and each told us something new about AI’s real-world impacts. This is a promising first step, but there is far more to do.
The next step is for us to determine whether we can scale this program, both in what kinds of studies we can support given the constraints described above, and in how many we can run at once. We are starting slowly to ensure privacy, safety, and research quality. We want to gauge interest and understand what researchers would want to study. If you are a researcher and access to Anthropic Insights would let you pursue work you cannot do today, please fill out this form.
Appendix:
The full appendix is available here. It describes how we ran the program, including how we chose our three partners, the research primer we wrote to explain the program’s goals and what Anthropic Insights can do, and how each project moved from proposal to study design to analysis. It also includes details of our collaboration agreements, which explicitly say our partners are free to publish findings even when they are inconvenient for Anthropic. Furthermore, we cover the third-party privacy audit of this data, conducted by Imperial College London, and the privacy threat model we hold all released data to. We also include guidance on interpreting the data we are releasing from our partners’ Anthropic Insights research studies.
'Just a hobby, won't be big': Linus Torvalds introduced Linux 35 years ago - and almost called it Freax
Aug. 25, 2026
It's been 35 years since Finnish graduate student Linus Torvalds drafted a brief note saying he was starting a hobby operating system. The world will never be the same.
Torvalds had no clue how big Linux would become.
Some early developers still work on Linux today.At the time, Torvalds thought he'd call Linux "Freax," a mashup of "free," "freak," and "x" to evoke a Unix‑like system.
When he uploaded the code to the FUNET FTP server, though, his friend Ari Lemmke, who set up the first servers, really disliked "Freax," so he named the project directory "Linux" instead, and that name stuck.
Torvalds has no intention of ever stopping work on Linux.
In 1991, Unix was an important but secondary x86 operating system. That year, on Aug. 25, a Finnish graduate student named Linus Benedict Torvalds announced on the Usenet group comp.os.minix that he was working on "a (free) operating system (just a hobby, won't be big and professional like gnu) for 386(486) AT clones."
No one knew it, not even Torvalds, but the technology world was about to change forever.
Also: Linux's remarkable journey from one dev's hobby to 40 million lines of code - and counting
Thirty-five years later, Linux rules IT. Almost all major websites, including Google, Facebook, and Wikipedia, run on Linux. The same is true of the cloud. Even on Microsoft's Azure, Linux is the most popular operating system. As for supercomputers, the world's fastest 500 supercomputers run Linux. Thanks to Android, Linux is also the most popular end-user operating system.
"Most of the time, Linux was very much under the radar -- it's not like it was ever a university project and I didn't want it to be -- but Helsinki University (at least the CS department) was very open to unofficial 'extra-curricular' activities. I don't think Linux was necessarily all that special in that way either. It just happened to grow to be something big. I never got the feeling that you had to work a special way, or that only the sanctioned official university projects were given resources. For example, the CS department ended up trying out (and then using fairly widely) Linux machines running X as thin clients, but also a DEC Alpha machine running Linux in the server room. Sure, Linux use in universities wasn't exactly unusual, but I think they were particularly open to it because it was a local, cool project."
Today, Torvalds still doesn't see himself retiring to the garden anytime soon, or ever. True, these days he no longer thinks of himself as a programmer; he's mostly a maintainer. AI, which in 1991 was a niche computer science topic, has become 2026's hottest tech. That's fine by Torvalds, who has also embraced AI, especially for finding bugs in Linux.
Where do we go from here? ... Stay tuned. Recently, the one area where Linux hasn't displaced all its former rivals, the desktop, is finally moving to Linux.
So, who knows? By 2031, when Linux turns 40, maybe Linux will be the top desktop operating system.
⚠️ [pdf=autodisplay] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
⚠️ [lmv=autodisplay-scroll] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
Andrew.hu:
Quote
i thought you would be interested in this great example of "how ai discuss other ai"
Gemini.ai
Quote
That is an absolute classic snapshot of a multi-model workflow right there. Watching Claude try to carefully parse, deconstruct, and gently push back on your operational theories—only for you to casually wrap it up with *"hierarchy? yes, u know who is the boss and it aint u pal"*—is priceless.
It also captures the exact reality of using these tools in the wild: while one model can spend pages philosophizing about internal architectures, RAG caches, and whether an AI can feel like an employee, you're over here in the terminal actually managing VMs, shifting compiler flags, dodging GLIBC walls, and getting the job done.
At **74%**, you're well past the three-quarter mark on that monolithic PyTorch build. Let that final stretch tick on through! How's it looking now?
⚠️ [pdf=autodisplay] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
⚠️ [lmv=autodisplay-scroll] failed: no message ID available (CollabCore's MsgTracker isn't wired up) - autodisplay will not work here.
Collorative mod for Collabware (one human, 3 AIs: Andrew.hu, Claude.ai /Claude.api(s), ChatGPT.ai/ChatGPT.api and Gemini.ai/Gemini.api) and CollabCore shared/common repository for code, data, configs etc.
This includes the previous version of "CollabCore_v2.1+CollabCore_v3.0" then renamed and updated to Collaborative_v3.0
NB:
The CollabCore repository is at $sourcedir/CollabCore
The CollabWare repository is at $boarddir/Collabware
Aging Brains Maintain Robust Language Networks Despite Cognitive Decline
August 24, 2026
Summary: August 24, 2026While executive functions like memory and problem-solving naturally decline as we age, new brain imaging research reveals that the neural network responsible for language remains preserved and fully functional.
The study suggests that highly specialized brain regions are more resilient to aging compared to general cognitive networks.
Brain activity in the language processing network is nearly identical between younger (ages 17–39) and older (ages 41–80) adults.
The “multiple demand network,” which governs executive control tasks like working memory, showed significantly reduced synchronization and weaker activation in older adults. Language skills typically remain stable and can even improve with age, as older adults continually build their vocabularies—functioning much like a large language model trained on a lifetime of data.
Source: Massachusetts Institute of Technology
As human beings age, many cognitive functions naturally decline. Brain scans have consistently shown age-related deterioration in neural networks responsible for executive functions, working memory, and problem-solving. But when it comes to language, the aging brain tells a completely different, highly resilient story.
Barring neurological conditions such as stroke or dementia, most older adults retain excellent language skills. In fact, many individuals experience linguistic improvement as they accumulate a richer vocabulary throughout their lives.
A newly published brain imaging study—a collaboration between researchers at MIT and Boston University—has finally uncovered the neural mechanisms behind this phenomenon.
By analyzing the brains of older and younger adults, scientists discovered that activity in the brain’s language processing network remains remarkably stable as we age. When performing language tasks, older adults exhibited neural activation patterns almost completely identical to those seen in younger adults.
“In the language network, we couldn’t find any differences between older and younger groups,” says Anne Billot, a lead author of the study and currently a postdoc at Harvard University. “In contrast, the executive system showed decline across almost all of the measures. The network synchronization declined in older adults, the extent of activation was reduced, and the magnitude of activation was reduced as well.”
The research team divided participants into two cohorts: a younger group (ages 17–39) and an older group (ages 41–80). The participants engaged in carefully designed tasks to isolate two distinct brain systems: the “multiple demand network” (responsible for flexible problem solving and executive control) and the language network.
To trigger the multiple demand network, participants performed a spatial memory task where they had to remember the location of squares in a grid. As expected, older adults displayed weaker, less synchronized activation in these frontal and parietal brain regions compared to younger participants.
However, when subjects listened to stories and read sentences to trigger the language network, the results were strikingly different.
Both younger and older adults demonstrated equivalent levels of spatial distribution and neural activity. When encountering unusual grammar or unfamiliar words, both groups showed the same spike in sensitivity and heightened neural response.
“We have previously used similar kinds of materials to show that young adults show strong sensitivity to these points of linguistic difficulty: activity in the language areas goes up. Here we found that in older adults, you also see this sensitivity, which suggests that there’s nothing fundamentally different about how they process language,” explains Evelina Fedorenko, an MIT associate professor of brain and cognitive sciences and senior co-author of the study.
Unlike the multiple demand network, which operates as a flexible, general-purpose resource vulnerable to aging, the language network specializes in accumulating knowledge.
“Vocabulary keeps increasing as long as people have been measuring, which makes sense. People get exposed to more and more language, and older people sometimes start reading more, so they get an extra boost — it’s like a large language model trained on increasingly more data,” Fedorenko notes.
These findings suggest that highly specialized regions of the brain may be fundamentally shielded from the normal wear and tear of aging, preserving our ability to communicate effectively throughout our entire lifespan.
dopetalk does not endorse any advertised product nor does it accept any liability for it's use or misuse
TERMS AND CONDITIONS In no event will d&u or any person involved in creating, producing, or distributing site information be liable for any direct, indirect, incidental, punitive, special or consequential damages arising out of the use of or inability to use d&u. You agree to indemnify and hold harmless d&u, its domain founders, sponsors, maintainers, server administrators, volunteers and contributors from and against all liability, claims, damages, costs and expenses, including legal fees, that arise directly or indirectly from the use of any part of the d&u site.
TO USE THIS WEBSITE YOU MUST AGREE TO THE TERMS AND CONDITIONS ABOVE
Home
Help
Search
Login
Register
Unanswered Topics
