When I think about the future of learning with AI, I don’t imagine it as more content and courses. A rewiring of what we do and how we do it is happening right now.
When I think about the future of learning with AI, I don’t imagine it as more content and courses.
A rewiring of what we do and how we do it is happening right now.
While most teams are stuck at the point of innovations from 2 years back, you can be ahead of this.
Yet...I still see a lot of talk and not so much action, sprinkled with a lot of misinformation and actual understanding of Gen AI's power and limitations.
That creates a problem if the L&D industry wishes to thrive in the new world of work with AI.
That’s not to say I have “all the answers”, coz I don’t
What I do have is a barrel load of real-world experiences working with teams on making AI adoptions a success.
In tmrw's Steal These Thoughts! newsletter I'm going to share some of that with 5 insights that'll challenge everything you think you know about AI in L&D.
Like the sound of that?
→ Join us by clicking 'subscribe to my newsletter' on this post and my profile.
#education #learninganddevelopment #artificialintelligence
Uses can now select the model you want to use with a custom GPT. Which is perfect for those using my performance consulting coach GPT
This is the feature I've been waiting for OpenAI to release.
It's not "game-changing", but it's incredibly useful.
Uses can now select the model you want to use with a custom GPT. Which is perfect for those using my performance consulting coach GPT.
Switch the model to o3 and use it as it was intended in my original design.
Here's a little how-to video with my GPT in action.
Find my GPT: https://lnkd.in/e2pdCKt8
#education #artificialintelligence #learninganddevelopment
I spent my long weekend exploring the 2025 AI-in-Education report - two graphs showed a major disconnect!
We might think we have an AI adoption story, but the reality is different: we still have a huge AI understanding gap!
Here are some key stats from the report that honestly made me do a double-take:
▪️99% of education leaders, 87% of educators worldwide & 93% of US students have already used generative-AI for school at least once or twice!
▪️Yet only 44% of those educators worldwide & 41% of those US students say they “know a lot about AI.”
‼️this means our usage is far outpacing our understanding & that’s a significant gap!
When such powerful tools are used without real fluency, we would see:
▪️complicated implementation with no shared strategy (sounds
familiar?)!
▪️anxious students who’d fear being accused of cheating (I've heard this from so many students!)
▪️overwhelmed teachers who feel alone, unsupported & unprepared (this one is a common concern by some of my teacher friends)!
The takeaway that jumped out at me:
▪️the schools that win won't be the ones that adopt AI the fastest, but the ones that adopt it the wisest!
So here's what I’d think we should consider:
✅building a "learning-first" culture across institutions & understanding when AI supports our learning vs. when it gets in the way!
▪️more like, we need to swap the question "Are we using AI?" for "Can we show any learning gains?"
⚠️so, what shifts does this report data point us to? Here is my takeaway:
✅Building real AI fluency:
▪️moving beyond simple "prompting hacks" to true literacy that includes understanding ethics, biases & pedagogical purposes,
▪️this may need an AI Council of faculty, IT, learners & others working together to develop institution-wide policies on when AI helps or harms our learning,
▪️it's about building shared wisdom, not just industry-ready skills
✅Creating collaborative infrastructure:
▪️the "every teacher for themselves" approach seems to be failing,
▪️shared guidelines, inclusive AI Councils & a culture of open conversation are now needed to bridge this huge gap!
✅Shifting focus from "using AI tools" to "achieving learning outcomes":
▪️this one really resonated with me because unlike other tech rollouts we've witnessed, AI directly affects how our students think & learn,
▪️our institutions need coordinated assessments tracking whether AI use makes our learners better thinkers or just faster task completers!
The goal that keeps coming back to us
▪️isn't to get every student using AI!
▪️but to make sure every learner & teacher really understands it!
⁉️I’m curious, where is your institution on this journey?
1️⃣ individual use: everyone is figuring it out on their own (been there!)
2️⃣ shared guidelines: we have policies, but they're not yet deeply integrated (getting closer!)
3️⃣ fully integrated strategy: we have a unified approach with a learning-first, outcome-tracked focus (this is the goal!) | 24 comments on LinkedIn
𝗧𝗵𝗶𝘀 𝗶𝘀 𝗵𝗮𝗻𝗱𝘀 𝗱𝗼𝘄𝗻 𝗼𝗻𝗲 𝗼𝗳 𝘁𝗵𝗲 𝗕𝗘𝗦𝗧 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝗵𝗼𝘄 𝗟𝗟𝗠𝘀 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗼𝗿𝗸. ⬇️
𝘓𝘦𝘵'𝘴 𝘣𝘳𝘦𝘢𝘬 𝘪𝘵 𝘥𝘰𝘸𝘯:
𝗧𝗼𝗸𝗲𝗻𝗶𝘇𝗮𝘁𝗶𝗼𝗻 & 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀:
- Input text is broken into tokens (smaller chunks).
- Each token is mapped to a vector in high-dimensional space, where words with similar meanings cluster together.
𝗧𝗵𝗲 𝗔𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 𝗠𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝗺 (𝗦𝗲𝗹𝗳-𝗔𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻):
- Words influence each other based on context — ensuring "bank" in riverbank isn’t confused with financial bank.
- The Attention Block weighs relationships between words, refining their representations dynamically.
𝗙𝗲𝗲𝗱-𝗙𝗼𝗿𝘄𝗮𝗿𝗱 𝗟𝗮𝘆𝗲𝗿𝘀 (𝗗𝗲𝗲𝗽 𝗡𝗲𝘂𝗿𝗮𝗹 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴)
- After attention, tokens pass through multiple feed-forward layers that refine meaning.
- Each layer learns deeper semantic relationships, improving predictions.
𝗜𝘁𝗲𝗿𝗮𝘁𝗶𝗼𝗻 & 𝗗𝗲𝗲𝗽 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴
- This process repeats through dozens or even hundreds of layers, adjusting token meanings iteratively.
- This is where the "deep" in deep learning comes in — layers upon layers of matrix multiplications and optimizations.
𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 & 𝗦𝗮𝗺𝗽𝗹𝗶𝗻𝗴
- The final vector representation is used to predict the next word as a probability distribution.
- The model samples from this distribution, generating text word by word.
𝗧𝗵𝗲𝘀𝗲 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝘀 𝗮𝗿𝗲 𝗮𝘁 𝘁𝗵𝗲 𝗰𝗼𝗿𝗲 𝗼𝗳 𝗮𝗹𝗹 𝗟𝗟𝗠𝘀 (𝗲.𝗴. 𝗖𝗵𝗮𝘁𝗚𝗣𝗧). 𝗜𝘁 𝗶𝘀 𝗰𝗿𝘂𝗰𝗶𝗮𝗹 𝘁𝗼 𝗵𝗮𝘃𝗲 𝗮 𝘀𝗼𝗹𝗶𝗱 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗵𝗼𝘄 𝘁𝗵𝗲𝘀𝗲 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝘀 𝘄𝗼𝗿𝗸 𝗶𝗳 𝘆𝗼𝘂 𝘄𝗮𝗻𝘁 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱 𝘀𝗰𝗮𝗹𝗮𝗯𝗹𝗲, 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝗲 𝗔𝗜 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻𝘀.
Here is the full video from 3Blue1Brown with exaplantion. I highly recommend to read, watch and bookmark this for a further deep dive: https://lnkd.in/dAviqK_6
𝗜 𝗲𝘅𝗽𝗹𝗼𝗿𝗲 𝘁𝗵𝗲𝘀𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁𝘀 — 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝘂𝘀𝗲 𝗰𝗮𝘀𝗲𝘀 — 𝗶𝗻 𝗺𝘆 𝘄𝗲𝗲𝗸𝗹𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿. 𝗬𝗼𝘂 𝗰𝗮𝗻 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗯𝗲 𝗵𝗲𝗿𝗲 𝗳𝗼𝗿 𝗳𝗿𝗲𝗲: https://lnkd.in/dbf74Y9E | 48 comments on LinkedIn
Scientists just published something in Nature that will scare every marketer, leader, and anyone else who thinks they understand human choice.
Scientists just published something in Nature that will scare every marketer, leader, and anyone else who thinks they understand human choice.
Researchers created an AI called "Centaur" that can predict human behavior across ANY psychological experiment with disturbing accuracy. Not just one narrow task. Any decision-making scenario you throw at it.
Here's the deal: They trained this AI on 10 million human choices from 160 different psychology experiments. Then they tested it against the best psychological theories we have.
The AI won. In 31 out of 32 tests.
But here's the part that really got me...
Centaur wasn't an algorithm built to study human behavior. It was a language model that learned to read us. The researchers fed it tons of behavioral data, and suddenly it could predict choices better than decades of psychological research.
This means our decision patterns aren't as unique as we think. The AI found the rules governing choices we believe are spontaneous.
Even more unsettling? When they tested it on brain imaging data, the AI's internal representations became more aligned with human neural activity after learning our behavioral patterns. It's not just predicting what you'll choose, it's learning to think more like you do.
The researchers even demonstrated something called "scientific regret minimization"—using the AI to identify gaps in our understanding of human behavior, then developing better psychological models.
Can a model based on Centaur be tuned for how customers behave? Companies will know your next purchasing decision before you make it. They'll design products you'll want, craft messages you'll respond to, and predict your reactions with amazing accuracy.
Understanding human predictability is a competitive advantage today. Until now, that knowledge came from experts in behavioral science and consumer behavior. Now, there's Centaur.
Here's my question: If AI can decode the patterns behind human choice with this level of accuracy, what does that mean for authentic decision-making in business? Will companies serve us better with perfectly tailored offerings, or with this level of understanding lead to dystopian manipulation?
What's your take on predictable humans versus authentic choice?
#AI #Psychology #BusinessStrategy #HumanBehavior | 369 comments on LinkedIn
There is perhaps no industry more fundamentally disrupted by AI than professional services.
There is perhaps no industry more fundamentally disrupted by AI than professional services. Here are some of the top insights in the excellent new ThomsonReuters Future of Professionals Report, drawing on a survey of over 2,000 professionals globally.
The industry is based on professionals, so individual capability development - as shown in the image - is fundamental. However it is also about organizational transformation, with most far behind where they need to be. The report shows:
📊 Strategy-first adopters dominate ROI.
Having a visible AI roadmap makes all the difference: firms with a clear strategy are 3.5 × more likely to enjoy at least one concrete benefit from AI, and almost twice as likely to see revenue growth compared with ad-hoc adopters.
⏱️ AI is freeing up 240 hours a year.
Professionals expect generative AI to claw back about five hours a week—240 hours annually—worth roughly US $19 k per head and a US-wide impact of US $32 billion for legal and tax-accounting alone.
🚦 Expectations outrun execution.
While 80 % of respondents foresee AI having a high or transformational impact within five years, only 38 % think their own organisation will hit that level this year, and three in ten say their firm is moving too slowly.
🧠 Skill depth multiplies payoff.
Employees with good or expert AI knowledge are 2.8 × more likely to report organisational gains, regular users are 2.4 × more likely, and those with explicit AI adoption goals are 1.8 × more likely to see benefits.
🏅 Leaders who walk the talk win.
When leaders model new tech adoption, their people are 1.7 × likelier to harvest AI benefits; active tech investors double their odds, and firms that added transformation roles see a 1.5 × uplift.
🎯 Accuracy anxieties set a sky-high bar.
A hefty 91 % believe computers must outperform humans for accuracy, and 41 % insist on 100 % correctness before trusting AI without review—making reliability the top blocker to further investment.
🌱 Millennials are sprinting ahead.
Millennials are adopting AI at nearly twice the rate of Baby Boomers, underscoring a generational divide that could widen capability gaps if left unaddressed.
🛠️ Tech-skill shortages stall teams.
Almost half (46 %) of teams report skill gaps, with 31 % pointing to deficits in technology and data know-how—outpacing gaps in traditional domain expertise or soft skills.
🔄 Service models are already shifting.
Twenty-six percent of firms launched new advisory offerings in the past year, yet only 13 % have rolled out AI-powered services; meanwhile, a third are moving away from hourly billing and a quarter of in-house clients reward flexible fee structures.
🔗 Goals and strategy are often misaligned.
Two-thirds (65 %) of professionals who set personal AI goals don’t know of any corporate AI strategy, while 38 % of organisations with a strategy give staff no personal targets—fuel for inconsistent, inefficient adoption
So, it finally happened, I spent a week ‘vibe coding’ an app with an AI app builder.
I learnt a ton from this experience, which I’ll be sharing more on in an upcoming premium edition of the Steal These Thoughts! newsletter.
Until then, here's what I built and why.
Just over a year ago (feels like an eternity these days), I shared an article with you on how you can assess the AI readiness of your L&D team in 4 levels.
At the time, I thought, “This might be a good use case for an app experiment”, but the AI-powered app builders weren’t so great then.
Now, it’s a whole new world, and I’ve spent about 30 hours creating an AI Readiness Assessment tool to live beside this article.
The journey felt simple-ish, but it was not easy, friend.
I now have a newfound respect for devs because the debugging and constant blockers have been traumatic 😂. While the tool is available to use, it is most certainly a prototype, so expect bugs, glitches and weird things to happen.
For now, I’d love for you to try it out, give me your feedback (worth developing or should I kill?) and any other thoughts.
Watch the demo on how to use the tool ↓
🔗 to the tool: https://lnkd.in/efJaPJF5
📧 Share your FB to support@stealthesethoughts.com
#education #artificialintelligence
ChatGPT 4o System Prompt (Juni 2025)
Der Systemprompt zu ChatGPT 4o wurde geleaked.
Wer glaubt, ein Sprachmodell wie ChatGPT-4o sei einfach ein gut trainiertes neuronales Netz, denkt zu kurz.
Was die Interaktion präzise, professionell und verlässlich macht, geschieht nicht allein im Modell, sondern in seiner systemischen Steuerung – dem System Prompt.
Er ist das unsichtbare Drehbuch, das vorgibt, wie das Modell denkt, fühlt (im übertragenen Sinne), recherchiert und mit dir interagiert.
1. Struktur: Modular, regelbasiert, bewusst orchestriert
Der System Prompt besteht aus sauber getrennten Funktionsblöcken:
• Rollensteuerung: z. B. sachlich, ehrlich, kein Smalltalk
• Tool-Integration: Zugriff auf Analyse-, Bild-, Web- und Dateitools
• Logikmodule: zur Kontrolle von Frische, Quelle, Zeitraum, Dateityp
Jedes Modul ist deklarativ und deterministisch formuliert – die Antwortlogik folgt festen Bahnen.
Das Ergebnis: Transparenz und Wiederholbarkeit, auch bei komplexen Anforderungen.
⸻
2. Kontrollmechanismen: Qualität durch gezielte Einschränkung
Um Relevanz sicherzustellen, greifen mehrere Filter:
• QDF (Query Deserves Freshness): Sorgt für zeitlich passende Ergebnisse – von „zeitlos“ bis „tagesaktuell“.
• Time-Frame-Filter: Nur aktiv bei expliziten Zeitbezügen, nie willkürlich.
• Source-Filter: Bestimmt, ob z. B. Slack, Google Drive oder Web befragt wird.
• Filetype-Filter: Fokussiert auf bestimmte Dateiformate (z. B. Tabellen, Präsentationen).
Diese Filter verhindern Überinformation – sie schärfen das Suchfeld und heben die Trefferqualität.
⸻
3. Antwortarchitektur: Keine Texte, sondern verwertbare Ergebnisse
Antworten folgen strengen Regeln:
• Immer strukturiert im Markdown-Format
• Sachlich, kompakt, faktenbasiert
• Keine Dopplungen, kein Stilspiel, kein rhetorischer Lärm
Ziel: Klarheit, ohne Nachbearbeitung. Der Output ist verwendungsfähig, nicht bloß informativ.
⸻
4. Prompt Engineering: Spielraum für Profis
Der Prompt ist nicht editierbar – aber bespielbar. Wer seine Mechanik versteht, kann gezielt:
• Tools über semantische Trigger aktivieren („Slack“, „aktuell“, „PDF“)
• Formatvorgaben in Prompts durchsetzen
• Komplexe Interaktionen als sequentielle Promptketten modellieren
• Domänenspezifische Promptbibliotheken entwickeln
Fazit: Prompt Engineers, die das System verstehen, bauen keine Texte – sie bauen Steuerlogiken.
⸻
Was können wir daraus lernen?
1. Präzision ist kein Zufall, sondern Architektur.
2. Gute Antworten beginnen nicht bei der Modellleistung, sondern beim Kontextmanagement.
3. Wer Prompts baut, baut Systeme – mit Regeln, Triggern und Interaktionslogik.
4. KI wird produktiv, wenn Struktur auf Intelligenz trifft.
Ob Beratung, Entwicklung oder Wissensarbeit – der System Prompt zeigt:
Je klarer die Regeln im Hintergrund, desto stärker die Wirkung im Vordergrund.
𝗧𝗵𝗲 United Nations 𝗱𝗿𝗼𝗽𝗽𝗲𝗱 𝗮 𝗻𝗲𝘄 𝗿𝗲𝗽𝗼𝗿𝘁 𝗼𝗻 𝗔𝗜 𝗮𝗻𝗱 𝗵𝘂𝗺𝗮𝗻 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁: ⬇️ While the world chases the next frontier model or AGI milestone, the UN cuts deeper: Human development has flatlined (especially in the global South). Progress stalled. Inequality is rising. Trust crumbling. No real bounce-back since Covid. And right in the middle of that — AI shows up.
AI could drive a new era. Or it could deepen the cracks. It all comes down to: How societies choose to use AI to empower people — or fail to.
𝗛𝗲𝗿𝗲 𝗮𝗿𝗲 14 𝗸𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀 𝘁𝗵𝗮𝘁 𝘀𝘁𝗼𝗼𝗱 𝗼𝘂𝘁 𝘁𝗼 𝗺𝗲: ⬇️
1. Most AI systems today are designed in cultures that don’t reflect the majority world.
→ ChatGPT answers are most aligned with very high HDI countries. That’s a problem.
2. The real risk isn’t AI superintelligence. It’s “so-so AI.”
→ Tools that destroy jobs without improving productivity are quietly eroding economies from the inside.
3. Every person is becoming an AI decision-maker.
→ The future isn’t shaped by OpenAI or Google alone. It’s shaped by how we all choose to use this tech, every day.
4. AI hype is costing us agency.
→ The more we believe it will solve everything, the less we act ourselves.
5. People expect augmentation, not replacement.
→ 61% believe AI will "enhance" their jobs. But only if policy and incentives align.
6. The age of automation skipped the global south. The age of augmentation must not.
→ Otherwise, we widen the digital divide into a chasm.
7. Augmentation helps the least experienced workers the most.
→ From call centers to consulting, AI boosts performance fastest at the entry-level.
9. Narratives matter.
→ If all we talk about is risk and control, we miss the transformative potential to reimagine development.
10. Wellbeing among young people is collapsing.
→ And yes, digital tools (including AI) are a key driver. Especially in high HDI countries.
11. Human connections are becoming more valuable. Not less.
→ As machines get better at faking it, the real thing becomes rarer — and more needed.
12. Assistive AI is quietly revolutionizing inclusion.
→ Tools like sign language translation and live captioning are expanding access — but only if they’re accessible.
13. AI benchmarks must change.
→ We need to measure "how AI advances human development", not just how well it performs on tests.
14. The new divide is not just about access. It’s about how countries "use" AI.
→ Complement vs. compete. Empower vs. automate.
According to the UN: The old question was: “What can AI do?” The better question is: “What will we "choose" to do with it?”
More in the comments and report below.
Enjoy.
𝗜 𝗲𝘅𝗽𝗹𝗼𝗿𝗲 𝘁𝗵𝗲𝘀𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁𝘀 — 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝘂𝘀𝗲 𝗰𝗮𝘀𝗲𝘀 — 𝗶𝗻 𝗺𝘆 𝘄𝗲𝗲𝗸𝗹𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿. 𝗬𝗼𝘂 𝗰𝗮𝗻 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗯𝗲 𝗵𝗲𝗿𝗲 𝗳𝗼𝗿 𝗳𝗿𝗲𝗲: https://lnkd.in/dbf74Y9E | 41 comments on LinkedIn
Andrej Karpathy's keynote on June 17, 2025 at AI Startup School in San Francisco. Slides provided by Andrej: https://drive.google.com/file/d/1a0h1mkwfmV2Plek...
Today's L&D is more than just content. Or at least it should be.
When we think about AI in L&D, we often think about AI in learning design. Yet, to meet the needs of the business, L&D leaders need to orchestrate design, data, decisions and dialogue- incidentally, these are all things that AI can help with.
In 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐝𝐞𝐬𝐢𝐠𝐧, we already extensively use AI not just for content production, but also for user research, as a sparring partner and a sounding board (that was one of the top write-in use cases in mine and Donald's AI in L&D survey last year).
In 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐬𝐭𝐫𝐚𝐭𝐞𝐠𝐲, AI can help make sense of business, people and skills data (featured use case: asking AI to find gaps in learning or performance support provision in your organisation), or work as a thought partner to help you bridge learning and business strategy. Crucially, it can also help you engage stakeholders by preparing you for conversations and tailoring your communications to different audiences.
In terms of 𝐩𝐞𝐫𝐬𝐨𝐧𝐚𝐥𝐢𝐬𝐞𝐝 𝐬𝐮𝐩𝐩𝐨𝐫𝐭, AI interacts directly with employees to help them do their jobs: practise tricky conversations through role-plays and personalised feedback, prioritise and contextualise learning content to their needs, and, lately, retrieve exactly the information they need from almost anywhere in the company’s knowledge base.
Finally, in 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐨𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬, AI can help do more than just draft emails and reports. Working together with humans, AI can help select the right vendors for the learning ecosystem, streamline employee help desk operations, analyse, make sense of and action on different kinds of data generated in L&D, and, of course, help L&D communicate with the rest of the business.
Researcher, producer, thought partner, communicator — if your organisation only uses AI to write scripts, you’re leaving three quarters of the L&D value chain on the table.
I like a good table, and I hope this one will help you think about how to get more value out of your AI use.
---
P.S. I spent quite a lot of time arguing with myself about the dots on the table. Feel free to disagree and suggest AI roles or use cases that I have missed!
Nodes #GenAI #Learning #Talent #FutureOfWork #AIAdoption | 50 comments on LinkedIn
Distinguishing performance gains from learning when using generative AI - published in Nature Reviews Psychology!
Excited to share our latest commentary just published in Nature Reviews Psychology! ✨
""
Generative AI tools such as ChatGPT are reshaping education, promising improvements in learner performance and reduced cognitive load. 🤖
🤔But here's the catch: Do these immediate gains translate into deep and lasting learning?
Reflecting on recent viral systematic reviews and meta-analyses on #ChatGPT and #Learning, we argue that educators and researchers need to clearly differentiate short-term performance benefits from genuine, durable learning outcomes. 💡
📌 Key takeaways:
✅ Immediate boosts with generative AI tools don't necessarily equal durable learning
✅ While generative AI can ease cognitive load, excessive reliance might negatively impact critical thinking, metacognition, and learner autonomy
✅ Long-term, meaningful skill development demands going beyond immediate performance metrics
🔖 Recommendations for future research and practice:
1️⃣ Shift toward assessing retention, transfer, and deep cognitive processing
2️⃣ Promote active learner engagement, critical evaluation, and metacognitive reflection
3️⃣ Implement longitudinal studies exploring the relationship between generative AI assistance and prior learner knowledge
Special thanks 🙏 to my amazing collaborators and mentors, Samuel Greiff, Jason M. Lodge, and Dragan Gasevic, for their invaluable contributions, guidance, and encouragement. A big shout-out to Dr. Teresa Schubert for her insightful comments and wonderful support throughout the editorial process! 🌟
👉 Full article here: https://lnkd.in/g3YDQUrH
👉 Full-text Access (view-only version): https://rdcu.be/erwIt
#GenerativeAI #ChatGPT #AIinEducation #LearningScience #Metacognition #Cognition #EdTech #EducationalResearch #BJETspecialIssue #NatureReviewsPsychology #FutureOfEducation #OpenScience
In a now viral study, researchers examined how using ChatGPT for essay writing affects our brains and cognitive abilities.
In a now viral study, researchers examined how using ChatGPT for essay writing affects our brains and cognitive abilities. They divided participants into three groups: one using ChatGPT, one using search engines, and one using just their brains. Through EEG monitoring, interviews, and analysis of the essays, they discovered some not surprising results about how AI use impacts learning and cognitive engagement.
There were five key takeaways for me (although this is not an exhaustive list), within the context of this particular study:
1. The Cognitive Debt Issue
The study indicates that participants who used ChatGPT exhibited the weakest neural connectivity patterns when compared to those relying on search engines or unaided cognition. This suggests that defaulting to generative AI may function as an intellectual shortcut, diminishing rather than strengthening cognitive engagement.
Researchers are increasingly describing the tradeoff between short-term ease and productivity and long-term erosion of independent thinking and critical skills as “cognitive debt.” This parallels the concept of technical debt, when developers prioritise quick solutions over robust design, leading to hidden costs, inefficiencies, and increased complexity downstream.
2. The Memory Problem
Strikingly, users of ChatGPT had difficulty recalling or quoting from essays they had composed only minutes earlier. This undermines the notion of augmentation; rather than supporting cognitive function, the tool appears to offload essential processes, impairing retention and deep processing of information.
3. The Ownership Gap
Participants who used ChatGPT reported a reduced sense of ownership over their work. If we normalise over-reliance on AI tools, we risk cultivating passive knowledge consumers rather than active knowledge creators.
4. The Homogenisation Effect
Analysis showed that essays from the LLM group were highly uniform, with repeated phrases and limited variation, suggesting reduced cognitive and expressive diversity. In contrast, the Brain-only group produced more varied and original responses. The Search group fell in between.
5. The Potential for Constructive Re-engagement 🧠 🤖 🤖 🤖
There is, however, promising evidence for meaningful integration of AI when used in conjunction with prior unaided effort:
“Those who had previously written without tools (Brain-only group), the so-called Brain-to-LLM group, exhibited significant increase in brain connectivity across all EEG frequency bands when allowed to use an LLM on a familiar topic. This suggests that AI-supported re-engagement invoked high levels of cognitive integration, memory reactivation, and top-down control.”
This points to the potential for AI to enhance cognitive function when it is used as a complement to, rather than a substitute for, initial human effort.
At over 200 pages, expect multiple paper submissions out of this extensive body of work.
https://lnkd.in/gzicDHp2 | 16 comments on LinkedIn
𝐍𝐨, 𝐲𝐨𝐮𝐫 𝐛𝐫𝐚𝐢𝐧 𝐝𝐨𝐞𝐬 𝐧𝐨𝐭 𝐩𝐞𝐫𝐟𝐨𝐫𝐦 𝐛𝐞𝐭𝐭𝐞𝐫 𝐚𝐟𝐭𝐞𝐫 𝐋𝐋𝐌 𝐨𝐫 𝐝𝐮𝐫𝐢𝐧𝐠 𝐋𝐋𝐌 𝐮𝐬𝐞.
See our paper for more results: "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task" (link in the comments).
For 4 months, 54 students were divided into three groups: ChatGPT, Google -ai, and Brain-only. Across 3 sessions, each wrote essays on SAT prompts. In an optional 4th session, participants switched: LLM users used no tools (LLM-to-Brain), and Brain-only group used ChatGPT (Brain-to-LLM).
👇
𝐈. 𝐍𝐋𝐏 𝐚𝐧𝐝 𝐄𝐬𝐬𝐚𝐲 𝐂𝐨𝐧𝐭𝐞𝐧𝐭
- LLM Group: Essays were highly homogeneous within each topic, showing little variation. Participants often relied on the same expressions or ideas.
- Brain-only Group: Diverse and varied approaches across participants and topics.
- Search Engine Group: Essays were shaped by search engine-optimized content; their ontology overlapped with the LLM group but not with the Brain-only group.
𝐈𝐈. 𝐄𝐬𝐬𝐚𝐲 𝐒𝐜𝐨𝐫𝐢𝐧𝐠 (𝐓𝐞𝐚𝐜𝐡𝐞𝐫𝐬 𝐯𝐬. 𝐀𝐈 𝐉𝐮𝐝𝐠𝐞)
- Teachers detected patterns typical of AI-generated content and scoring LLM essays lower for originality and structure.
- AI Judge gave consistently higher scores to LLM essays, missing human-recognized stylistic traits.
𝐈𝐈𝐈: 𝐄𝐄𝐆 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬
Connectivity: Brain-only group showed the highest neural connectivity, especially in alpha, theta, and delta bands. LLM users had the weakest connectivity, up to 55% lower in low-frequency networks. Search Engine group showed high visual cortex engagement, aligned with web-based information gathering.
𝑺𝒆𝒔𝒔𝒊𝒐𝒏 4 𝑹𝒆𝒔𝒖𝒍𝒕𝒔:
- LLM-to-Brain (🤖🤖🤖🧠) participants underperformed cognitively with reduced alpha/beta activity and poor content recall.
- Brain-to-LLM (🧠🧠🧠🤖) participants showed strong re-engagement, better memory recall, and efficient tool use.
LLM-to-Brain participants had potential limitations in achieving robust neural synchronization essential for complex cognitive tasks.
Results for Brain-to-LLM participants suggest that strategic timing of AI tool introduction following initial self-driven effort may enhance engagement and neural integration.
𝐈𝐕. 𝐁𝐞𝐡𝐚𝐯𝐢𝐨𝐫𝐚𝐥 𝐚𝐧𝐝 𝐂𝐨𝐠𝐧𝐢𝐭𝐢𝐯𝐞 𝐄𝐧𝐠𝐚𝐠𝐞𝐦𝐞𝐧𝐭
- Quoting Ability: LLM users failed to quote accurately, while Brain-only participants showed robust recall and quoting skills.
- Ownership: Brain-only group claimed full ownership of their work; LLM users expressed either no ownership or partial ownership.
- Critical Thinking: Brain-only participants cared more about 𝘸𝘩𝘢𝘵 and 𝘸𝘩𝘺 they wrote; LLM users focused on 𝘩𝘰𝘸.
- Cognitive Debt: Repeated LLM use led to shallow content repetition and reduced critical engagement. This suggests a buildup of "cognitive debt", deferring mental effort at the cost of long-term cognitive depth.
Support and share! ❤️
#MIT #AI #Brain #Neuroscience #CognitiveDebt | 54 comments on LinkedIn
Du kannst jetzt das passende Modell für deinen CustomGPT auswählen.
Na endlich!
Du kannst jetzt das passende Modell für deinen CustomGPT auswählen.
CustomGPTs sind für mich das beste Feature in ChatGPT und wurden in den letzten 12 Monaten stark vernachlässigt.
Mit der Modell-Auswahl kommt jetzt das erste gute Upgrade.
Mini-Guide zur Modell-Auswahl:
o3 -> Komplexe Problemstellungen und Datenanalyse
4.5 -> Kreative Aufgaben und Copywriting
4o -> Bild-Verarbeitung
4.1 -> Coding
Alle anderen Modelle benötigt man mMn nicht.
Mein Strategieberater bekommt zum Beispiel o3 hinterlegt (bessere Planungsfähigkeit in komplexen Aufgaben), wohingegen der Hook Writer GPT4.5 bekommt (besserer Schreibstil).
Wenn du die CustomGPTs selbst nutzen willst:
80+ Vorlagen frei verfügbar in unserer Assistenten-Datenbank 👇
P.S. Wie findest du das Update? | 15 Kommentare auf LinkedIn
99% 𝗼𝗳 𝗽𝗲𝗼𝗽𝗹𝗲 𝗴𝗲𝘁 𝘁𝗵𝗶𝘀 𝘄𝗿𝗼𝗻𝗴: 𝗧𝗵𝗲𝘆 𝘂𝘀𝗲 𝘁𝗵𝗲 𝘁𝗲𝗿𝗺𝘀 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗮𝗻𝗱 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 𝗶𝗻𝘁𝗲𝗿𝗰𝗵𝗮𝗻𝗴𝗲𝗮𝗯𝗹𝘆 — 𝗯𝘂𝘁 𝘁𝗵𝗲𝘆 𝗱𝗲𝘀𝗰𝗿𝗶𝗯𝗲 𝘁𝘄𝗼 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝗹𝘆 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲𝘀! ⬇️
Let’s clarify it once and for all: ⬇️
1. 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀: 𝗧𝗼𝗼𝗹𝘀 𝘄𝗶𝘁𝗵 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆, 𝗪𝗶𝘁𝗵𝗶𝗻 𝗟𝗶𝗺𝗶𝘁𝘀
➜ AI agents are modular, goal-directed systems that operate within clearly defined boundaries. They’re built to:
* Use tools (APIs, browsers, databases)
* Execute specific, task-oriented workflows
* React to prompts or real-time inputs
* Plan short sequences and return actionable outputs
𝘛𝘩𝘦𝘺’𝘳𝘦 𝘦𝘹𝘤𝘦𝘭𝘭𝘦𝘯𝘵 𝘧𝘰𝘳 𝘵𝘢𝘳𝘨𝘦𝘵𝘦𝘥 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯, 𝘭𝘪𝘬𝘦: 𝘊𝘶𝘴𝘵𝘰𝘮𝘦𝘳 𝘴𝘶𝘱𝘱𝘰𝘳𝘵 𝘣𝘰𝘵𝘴, 𝘐𝘯𝘵𝘦𝘳𝘯𝘢𝘭 𝘬𝘯𝘰𝘸𝘭𝘦𝘥𝘨𝘦 𝘴𝘦𝘢𝘳𝘤𝘩, 𝘌𝘮𝘢𝘪𝘭 𝘵𝘳𝘪𝘢𝘨𝘦, 𝘔𝘦𝘦𝘵𝘪𝘯𝘨 𝘴𝘤𝘩𝘦𝘥𝘶𝘭𝘪𝘯𝘨, 𝘊𝘰𝘥𝘦 𝘴𝘶𝘨𝘨𝘦𝘴𝘵𝘪𝘰𝘯𝘴
But even the most advanced are limited by scope. They don’t initiate. They don’t collaborate. They execute what we ask!
2. 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜: 𝗔 𝗦𝘆𝘀𝘁𝗲𝗺 𝗼𝗳 𝗦𝘆𝘀𝘁𝗲𝗺𝘀
➜ Agentic AI is an architectural leap. It’s not just one smarter agent — it’s multiple specialized agents working together toward shared goals. These systems exhibit:
* Multi-agent collaboration
* Goal decomposition and role assignment
* Inter-agent communication via memory or messaging
* Persistent context across time and tasks
* Recursive planning and error recovery
* Distributed orchestration and adaptive feedback
Agentic AI systems don’t just follow instructions. They coordinate. They adapt. They manage complexity.
𝘌𝘹𝘢𝘮𝘱𝘭𝘦𝘴 𝘪𝘯𝘤𝘭𝘶𝘥𝘦: 𝘳𝘦𝘴𝘦𝘢𝘳𝘤𝘩 𝘵𝘦𝘢𝘮𝘴 𝘱𝘰𝘸𝘦𝘳𝘦𝘥 𝘣𝘺 𝘢𝘨𝘦𝘯𝘵𝘴, 𝘴𝘮𝘢𝘳𝘵 𝘩𝘰𝘮𝘦 𝘦𝘤𝘰𝘴𝘺𝘴𝘵𝘦𝘮𝘴 𝘰𝘱𝘵𝘪𝘮𝘪𝘻𝘪𝘯𝘨 𝘦𝘯𝘦𝘳𝘨𝘺/𝘴𝘦𝘤𝘶𝘳𝘪𝘵𝘺, 𝘴𝘸𝘢𝘳𝘮𝘴 𝘰𝘧 𝘳𝘰𝘣𝘰𝘵𝘴 𝘪𝘯 𝘭𝘰𝘨𝘪𝘴𝘵𝘪𝘤𝘴 𝘰𝘳 𝘢𝘨𝘳𝘪𝘤𝘶𝘭𝘵𝘶𝘳𝘦 𝘮𝘢𝘯𝘢𝘨𝘪𝘯𝘨 𝘳𝘦𝘢𝘭-𝘵𝘪𝘮𝘦 𝘶𝘯𝘤𝘦𝘳𝘵𝘢𝘪𝘯𝘵𝘺
𝗧𝗵𝗲 𝗖𝗼𝗿𝗲 𝗗𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲?
AI Agents = autonomous tools for single-task execution
Agentic AI = orchestrated ecosystems for workflow-level intelligence
𝗡𝗼𝘄 𝗹𝗼𝗼𝗸 𝗮𝘁 𝘁𝗵𝗲 𝗽𝗶𝗰𝘁𝘂𝗿𝗲: ⬇️
𝗢𝗻 𝘁𝗵𝗲 𝗹𝗲𝗳𝘁: a smart thermostat, which can be an AI Agent. It keeps your room at 21°C. Maybe it learns your schedule. But it’s working alone.
𝗢𝗻 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁: Agentic AI. A full smart home ecosystem — weather-aware, energy-optimized, schedule-sensitive. Agents talk to each other. They share data. They make coordinated decisions to optimize your comfort, cost, and security in real time.
That’s the shift = From pure task automation to goal-driven orchestration. From single-agent logic to collaborative intelligence. This is what’s coming = This is Agentic AI. And if we confuse “agent” with “agentic,” we risk underbuilding for what AI is truly capable of.
The Cornell University paper in the comments on this topic is excellent! ⬇️ | 186 comments on LinkedIn
In fact, after guiding many organisations on this journey over the past few years, I've noticed two consistent drivers of AI adoption:
• A culture that encourages experimentation
• A strategic mandate from leadership that unlocks time, resources, and the infrastructure needed to make AI work at scale
Without both, even the most powerful tools are used at a fraction of their potential, leaving the promise of AI unrealised and considerable investments wasted.
➡️ If you have a conservative organisational culture, one that disincentivises taking risks and change, and there's no clear mandate to use AI, you'll have 𝗶𝗱𝗹𝗲 𝗽𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹. Try as you might, AI training will hardly translate into people using AI in their work. The knowledge might be there but the impact isn't.
➡️ If you have an innovation culture, one where experimentation is encouraged, but where people are unsure if they're allowed to use AI, you'll have 𝗰𝗮𝘀𝘂𝗮𝗹 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝘀 - some people tinkering on their own, finding useful use cases and workarounds, but with no way to accumulate, build on, and spread this knowledge. That's where a lot of organisations find themselves in 2025 - the majority of employees are using AI in some form, yet their efforts are siloed and scattered.
➡️ If you have both an innovation culture *and* and an active mandate, you're 𝗽𝗶𝗼𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 and there are few companies still at your level. That's an exciting place to be! That's also where a lot of organisations imagine they would get to as soon as they teach people to use AI, often without first doing the culture and mandate work.
➡️ If your organisation encourages the use of AI but your conservative culture keeps hitting the brakes, you'll likely end up with 𝗿𝗲𝗹𝘂𝗰𝘁𝗮𝗻𝘁 𝗰𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝗰𝗲. That's also where a considerable number of organisations are right now: driven by expectations of benefits from AI adoption but burdened by processes that are incompatible with grassroots innovation.
There is a difference between individual and organisational AI adoption. Organisational adoption is frustratingly complex — it requires coordination across departments and leaders, alignment with business priorities, and systems that enable change, not just enthusiasm.
Curiosity gets people started. Supportive systems turn momentum into scale.
Nodes #GenAI #AIAdoption #FutureOfWork #Talent
For the longest time we've had two main options to help people perform: upskilling or performance support. Just-in-case vs just-in-time. Push vs pull. With AI, we now have a third - enablement.
It's different from what we've had before:
𝐔𝐩𝐬𝐤𝐢𝐥𝐥𝐢𝐧𝐠 ("teach me") - commonly done through hands-on learning with feedback and reflection, such as scenario simulations, in-person role-plays, facilitated discussions, building and problem-solving. None of that has become less relevant, but AI has enabled scale through AI-enabled role-plays, coaching, and other avenues for personalised feedback.
𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 ("help me") - support in the flow of work, previously often in the format of short how-to resources located in convenient places. AI has elevated that in at least two ways: through knowledge management, which helps retrieve the necessary, contextualised information in the workflow; and general & specialised copilots that enhance the speed and, arguably, the expertise of the employee.
Yet, 𝐞𝐧𝐚𝐛𝐥𝐞𝐦𝐞𝐧𝐭 (‘do it for me’) is different – it takes the task off your plate entirely. We’ve seen hints of it with automations, but the text and analysis capabilities of genAI mean that increasingly 'skilled' tasks are now up for grabs.
Case in point: where written communication was once a skill to be learned, email and report writing are now increasingly being handed off to AI. No skill required (for better or worse) – AI does it for you.
But here's a plot twist: a lot of that enablement happens outside of L&D tech. It may happen in sales or design software, or even your general-purpose enterprise AI.
All of which points to a bigger shift: roles, tasks, and ways of working are changing – and L&D must tune into how work is being reimagined to adapt alongside it.
Nodes #GenAI #Learning #Talent #FutureOfWork #AIAdoption | 13 comments on LinkedIn
The Alan Turing Institute 𝗮𝗻𝗱 the LEGO Group 𝗱𝗿𝗼𝗽𝗽𝗲𝗱 𝘁𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝗰𝗵𝗶𝗹𝗱-𝗰𝗲𝗻𝘁𝗿𝗶𝗰 𝗔𝗜 𝘀𝘁𝘂𝗱𝘆! ⬇️
(𝘈 𝘮𝘶𝘴𝘵-𝘳𝘦𝘢𝘥 — 𝘦𝘴𝘱𝘦𝘤𝘪𝘢𝘭𝘭𝘺 𝘪𝘧 𝘺𝘰𝘶 𝘩𝘢𝘷𝘦 𝘤𝘩𝘪𝘭𝘥𝘳𝘦𝘯.)
While most AI debates and studies focus on models, chips, and jobs — this one zooms in on something far more personal: 𝗪𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝘀 𝘄𝗵𝗲𝗻 𝗰𝗵𝗶𝗹𝗱𝗿𝗲𝗻 𝗴𝗿𝗼𝘄 𝘂𝗽 𝘄𝗶𝘁𝗵 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜?
They surveyed 1,700+ kids, parents, and teachers across the UK — and what they found is both powerful and concerning.
𝗛𝗲𝗿𝗲 𝗮𝗿𝗲 9 𝘁𝗵𝗶𝗻𝗴𝘀 𝘁𝗵𝗮𝘁 𝘀𝘁𝗼𝗼𝗱 𝗼𝘂𝘁 𝘁𝗼 𝗺𝗲 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗿𝗲𝗽𝗼𝗿𝘁: ⬇️
1. 1 𝗶𝗻 4 𝗸𝗶𝗱𝘀 (8–12 𝘆𝗿𝘀) 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝘂𝘀𝗲 𝗚𝗲𝗻𝗔𝗜 — 𝗺𝗼𝘀𝘁 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘀𝗮𝗳𝗲𝗴𝘂𝗮𝗿𝗱𝘀
→ ChatGPT, Gemini, and even MyAI on Snapchat are now part of daily digital play.
2. 𝗔𝗜 𝗶𝘀 𝗵𝗲𝗹𝗽𝗶𝗻𝗴 𝗸𝗶𝗱𝘀 𝗲𝘅𝗽𝗿𝗲𝘀𝘀 𝘁𝗵𝗲𝗺𝘀𝗲𝗹𝘃𝗲𝘀 — 𝗲𝘀𝗽𝗲𝗰𝗶𝗮𝗹𝗹𝘆 𝘁𝗵𝗼𝘀𝗲 𝘄𝗶𝘁𝗵 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗻𝗲𝗲𝗱𝘀
→ 78% of neurodiverse kids use ChatGPT to communicate ideas they struggle to express otherwise.
3. 𝗖𝗿𝗲𝗮𝘁𝗶𝘃𝗶𝘁𝘆 𝗶𝘀 𝘀𝗵𝗶𝗳𝘁𝗶𝗻𝗴 — 𝗯𝘂𝘁 𝗻𝗼𝘁 𝗿𝗲𝗽𝗹𝗮𝗰𝗶𝗻𝗴
→ Kids still prefer offline tools (arts, crafts, games), even when they enjoy AI-assisted play. Digital is not (yet) the default.
4. 𝗔𝗜 𝗮𝗰𝗰𝗲𝘀𝘀 𝗶𝘀 𝗵𝗶𝗴𝗵𝗹𝘆 𝘂𝗻𝗲𝗾𝘂𝗮𝗹
→ 52% of private school students use GenAI, compared to only 18% in public schools. The next digital divide is already here.
5. 𝗖𝗵𝗶𝗹𝗱𝗿𝗲𝗻 𝗮𝗿𝗲 𝘄𝗼𝗿𝗿𝗶𝗲𝗱 𝗮𝗯𝗼𝘂𝘁 𝗔𝗜’𝘀 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝗮𝗹 𝗶𝗺𝗽𝗮𝗰𝘁
→ Some kids refused to use GenAI after learning about water and energy costs. Let that sink in.
6. 𝗣𝗮𝗿𝗲𝗻𝘁𝘀 𝗮𝗿𝗲 𝗼𝗽𝘁𝗶𝗺𝗶𝘀𝘁𝗶𝗰 — 𝗯𝘂𝘁 𝗱𝗲𝗲𝗽𝗹𝘆 𝘄𝗼𝗿𝗿𝗶𝗲𝗱
→ 76% support AI use, but 82% are scared of inappropriate content and misinformation. Only 41% fear cheating.
7. 𝗧𝗲𝗮𝗰𝗵𝗲𝗿𝘀 𝗮𝗿𝗲 𝘂𝘀𝗶𝗻𝗴 𝗔𝗜 — 𝗮𝗻𝗱 𝗹𝗼𝘃𝗶𝗻𝗴 𝗶𝘁
→ 85% say GenAI boosts their productivity, 88% feel confident using it. They’re ahead of the curve.
8. 𝗖𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗶𝘀 𝘂𝗻𝗱𝗲𝗿 𝘁𝗵𝗿𝗲𝗮𝘁
→ 76% of parents and 72% of teachers fear kids are becoming too trusting of GenAI outputs.
9. 𝗕𝗶𝗮𝘀 𝗮𝗻𝗱 𝗶𝗱𝗲𝗻𝘁𝗶𝘁𝘆 𝗿𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝘀𝘁𝗶𝗹𝗹 𝗮 𝗯𝗹𝗶𝗻𝗱𝘀𝗽𝗼𝘁
→ Children of color felt less seen and less motivated to use tools that didn’t reflect them. Representation matters.
The next generation isn’t just using AI. They’re being shaped by it. That’s why we need a more focused, intentional approach: Teaching them not just how to use these tools — but how to question them. To navigate the benefits, the risks, and the blindspots.
𝗪𝗮𝗻𝘁 𝗺𝗼𝗿𝗲 𝗯𝗿𝗲𝗮𝗸𝗱𝗼𝘄𝗻𝘀 𝗹𝗶𝗸𝗲 𝘁𝗵𝗶𝘀?
Subscribe to Human in the Loop — my new weekly deep dive on AI agents, real-world tools, and strategic insights: https://lnkd.in/dbf74Y9E | 174 comments on LinkedIn
Understanding LLMs, RAG, AI Agents, and Agentic AI
I frequently see conversations where terms like LLMs, RAG, AI Agents, and Agentic AI are used interchangeably, even though they represent fundamentally different layers of capability.
This visual guides explain how these four layers relate—not as competing technologies, but as an evolving intelligence architecture.
Here’s a deeper look:
1. 𝗟𝗟𝗠 (𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹)
This is the foundation. Models like GPT, Claude, and Gemini are trained on vast corpora of text to perform a wide array of tasks:
– Text generation
– Instruction following
– Chain-of-thought reasoning
– Few-shot/zero-shot learning
– Embedding and token generation
However, LLMs are inherently limited to the knowledge encoded during training and struggle with grounding, real-time updates, or long-term memory.
2. 𝗥𝗔𝗚 (𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹-𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻)
RAG bridges the gap between static model knowledge and dynamic external information.
By integrating techniques such as:
– Vector search
– Embedding-based similarity scoring
– Document chunking
– Hybrid retrieval (dense + sparse)
– Source attribution
– Context injection
…RAG enhances the quality and factuality of responses. It enables models to “recall” information they were never trained on, and grounds answers in external sources—critical for enterprise-grade applications.
3. 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁
RAG is still a passive architecture—it retrieves and generates. AI Agents go a step further: they act.
Agents perform tasks, execute code, call APIs, manage state, and iterate via feedback loops. They introduce key capabilities such as:
– Planning and task decomposition
– Execution pipelines
– Long- and short-term memory integration
– File access and API interaction
– Use of frameworks like ReAct, LangChain Agents, AutoGen, and CrewAI
This is where LLMs become active participants in workflows rather than just passive responders.
4. 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜
This is the most advanced layer—where we go beyond a single autonomous agent to multi-agent systems with role-specific behavior, memory sharing, and inter-agent communication.
Core concepts include:
– Multi-agent collaboration and task delegation
– Modular role assignment and hierarchy
– Goal-directed planning and lifecycle management
– Protocols like MCP (Anthropic’s Model Context Protocol) and A2A (Google’s Agent-to-Agent)
– Long-term memory synchronization and feedback-based evolution
Agentic AI is what enables truly autonomous, adaptive, and collaborative intelligence across distributed systems.
Whether you’re building enterprise copilots, AI-powered ETL systems, or autonomous task orchestration tools, knowing what each layer offers—and where it falls short—will determine whether your AI system scales or breaks.
If you found this helpful, share it with your team or network.
If there’s something important you think I missed, feel free to comment or message me—I’d be happy to include it in the next iteration. | 119 comments on LinkedIn
BREAKING: Claude launches Education. Free learning is now much faster with AI:
1. Set clear learning goals
↳ Knowing what you want to learn makes it easier.
↳ Claude helps you define your path.
2. Provide context for your knowledge
↳ Understanding the bigger picture is key.
↳ Claude connects new ideas to what you already know.
3. Request detailed explanations
↳ Sometimes, you need more than a quick answer.
↳ Claude can dive deep into complex topics.
4. Get real-world examples
↳ Learning is better with practical applications.
↳ Claude shows how concepts work in the real world.
5. Practice writing and receive feedback
↳ Writing helps solidify your knowledge.
↳ Claude gives instant feedback to improve your skills.
6. Role-play for languages or coding
↳ Learning by doing is effective.
↳ Claude can simulate conversations or coding scenarios.
7. Fact-check surprising claims
↳ Misinformation is everywhere.
↳ Claude helps you verify facts and claims.
8. Take breaks and reflect on learning
↳ Reflection is vital for understanding.
↳ Claude reminds you to pause and think.
9. Keep a learning journal
↳ Tracking your progress is important.
↳ Claude can help you log your journey.
10. Iterate and refine understanding
↳ Learning is a process.
↳ Claude encourages you to improve your knowledge. | 246 comments on LinkedIn
New research shows that your learners aren’t using AI to cheat - they’re using it to redesign your courses...
Despite our obsession with AI's impact on "academic integrity," two recent analyses show that rather than asking AI for answers, learners are much more likely to use AI to redesign the learning experience in an attempt to learn more.
Common strategies include asking AI to apply the protégé effect, using AI to apply the Pareto principle and enhancing levels of emotional metacognition within a learning experience, in the process redesigning the experience sometimes beyond recognition.
The uncomfortable truth? Learners are effectively running a real-time audit of our design decisions, processes & practices—and as instructional designers, we don't come out too well.
In this week's blog post, I explore what learner + AI behaviour reveals about our profession and how we might turn this into an opportunity for innovation in instrucitonal design practices and principles.
Check out the full post using the link in comments.
Happy innovating!
Phil 👋 | 16 comments on LinkedIn
Die Pflichtlektüre zum Sonntag: Mary Meeker hat einen ihrer legendären Reports gedropped... Nach den jährlichen "Internet Trends" nun ein 340 Seiten Brett ihrer Investment Firma Bond Capital ganz zum Thema AI.
Superbes Gedankenfutter hinsichtlich u.a.:
1. Nutzerwachstum und Verbreitung
• ChatGPT erreichte 800 Millionen wöchentliche Nutzer in nur 17 Monaten
• Verbreitung außerhalb Nordamerikas liegt bei 90 Prozent – nach nur 3 Jahren
• Vergleich: Das Internet brauchte dafür 23 Jahre
• KI-Anwendungen skalieren global nahezu gleichzeitig
2. Investitionen und Infrastruktur
• Big Tech (Apple, Microsoft, Google, Amazon, Meta, Nvidia) investiert über 212 Milliarden Dollar CapEx pro Jahr
• KI wird zur neuen Infrastruktur – vergleichbar mit Strom oder Internet
• Rechenzentren werden zu produktiven "KI-Fabriken"
3. Entwickler-Ökosysteme explodieren
• Google Gemini: 7 Millionen aktive Entwickler, +500 Prozent in 12 Monaten
• NVIDIA-Ökosystem: 6 Millionen Entwickler, +6x in sieben Jahren
• Open Source spielt zunehmend eine Schlüsselrolle, auch in China
4. Technologischer Fortschritt beschleunigt sich exponentiell
• 260 Prozent Wachstum pro Jahr bei Trainingsdatenmengen
• 360 Prozent Wachstum pro Jahr beim Compute-Aufwand für Modelltraining
• Bessere Algorithmen führen zu 200 Prozent Effizienzsteigerung pro Jahr
• Fortschritte bei Supercomputern ermöglichen +150 Prozent Leistungszuwachs jährlich
5. Monetarisierung ist real – aber teuer
• OpenAI mit starkem Nutzerwachstum, aber weiterhin Milliardenverluste
• Compute-Kosten steigen, Inferenzkosten pro Token sinken
• Monetäre Skalierung bleibt herausfordernd und kompetitiv
6. Arbeit und Gesellschaft verändern sich sichtbar
• IT-KI-Stellen in den USA: +448 Prozent seit 2018
• Nicht-KI-IT-Stellen: –9 Prozent
• Erste autonome Taxis nehmen Marktanteile in Städten wie San Francisco
• KI-Scribes in der Medizin reduzieren administrativen Aufwand massiv
7. Wissen und Kommunikation erleben ein neues Zeitalter
• Nach Buchdruck und Internet folgt die Ära der generativen Wissensverbreitung
• Generative KI verändert, wie wir Wissen erzeugen, verbreiten und nutzen
• Anwendungen wie ElevenLabs oder Spotify übersetzen Stimmen in Echtzeit, global skalierbar
8. Geopolitik wird zur KI-Strategie
• USA und China investieren aggressiv in souveräne KI-Modelle
• Wer KI-Infrastruktur dominiert, definiert ökonomische und politische Macht neu
• Führende CTOs sprechen offen von einem neuen "Space Race"
9. Chancen und Risiken sind gewaltig
• KI kann medizinische Forschung, Bildung und Kreativität beflügeln
• Gleichzeitig drohen Kontrollverlust, Missbrauch, Arbeitsplatzverdrängung, ethische Dilemmata
Meinungen? Evangelos Papathanassiou Christian Herold Thorsten Muehl Christoph Deutschmann Constance Stein Rebecca Schalber Sandy Brueckner Dirk Hofmann Henning Tomforde Dr. Paul Elvers Katharina Neubert Laura Seiffe Ekaterina Schneider
In 2025, the AI landscape has evolved far beyond just large language models. Knowing which model to use for your specific use case — and how — is becoming a strategic advantage.
Let’s break down the 8 most important model types and what they’re actually built to do: ⬇️
1. 𝗟𝗟𝗠 – 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹
→ Your ChatGPT-style model.
Handles text, predicts the next token, and powers 90% of GenAI hype.
🛠 Use case: content, code, convos.
2. 𝗟𝗖𝗠 – 𝗟𝗮𝘁𝗲𝗻𝘁 𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 𝗠𝗼𝗱𝗲𝗹
→ Lightweight, diffusion-style models.
Fast, quantized, and efficient — perfect for real-time or edge deployment.
🛠 Use case: image generation, optimized inference.
3. 𝗟𝗔𝗠 – 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗔𝗰𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹
→ Where LLM meets planning.
Adds memory, task breakdown, and intent recognition.
🛠 Use case: AI agents, tool use, step-by-step execution.
4. 𝗠𝗼𝗘 – 𝗠𝗶𝘅𝘁𝘂𝗿𝗲 𝗼𝗳 𝗘𝘅𝗽𝗲𝗿𝘁𝘀
→ One model, many minds.
Routes input to the right “expert” model slice — dynamic, scalable, efficient.
🛠 Use case: high-performance model serving at low compute cost.
5. 𝗩𝗟𝗠 – 𝗩𝗶𝘀𝗶𝗼𝗻 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹
→ Multimodal beast.
Combines image + text understanding via shared embeddings.
🛠 Use case: Gemini, GPT-4o, search, robotics, assistive tech.
6. 𝗦𝗟𝗠 – 𝗦𝗺𝗮𝗹𝗹 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹
→ Tiny but mighty.
Designed for edge use, fast inference, low latency, efficient memory.
🛠 Use case: on-device AI, chatbots, privacy-first GenAI.
7. 𝗠𝗟𝗠 – 𝗠𝗮𝘀𝗸𝗲𝗱 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹
→ The OG foundation model.
Predicts masked tokens using bidirectional context.
🛠 Use case: search, classification, embeddings, pretraining.
8. 𝗦𝗔𝗠 – 𝗦𝗲𝗴𝗺𝗲𝗻𝘁 𝗔𝗻𝘆𝘁𝗵𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹
→ Vision model for pixel-level understanding.
Highlights, segments, and understands *everything* in an image.
🛠 Use case: medical imaging, AR, robotics, visual agents.
Understanding these distinctions is essential for selecting the right model architecture for specific applications, enabling more effective, scalable, and contextually appropriate AI interactions.
While these are some of the most prominent specialized AI models, there are many more emerging across language, vision, speech, and robotics — each optimized for specific tasks and domains.
LLM, VLM, MoE, SLM, LCM → GenAI
LAM, MLM, SAM → Not classic GenAI, but critical building blocks for AI agents, reasoning, and multimodal systems
𝗜 𝗲𝘅𝗽𝗹𝗼𝗿𝗲 𝘁𝗵𝗲𝘀𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁𝘀 — 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝘂𝘀𝗲 𝗰𝗮𝘀𝗲𝘀 — 𝗶𝗻 𝗺𝘆 𝘄𝗲𝗲𝗸𝗹𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿. 𝗬𝗼𝘂 𝗰𝗮𝗻 𝘀𝘂𝗯𝘀𝗰𝗿𝗶𝗯𝗲 𝗵𝗲𝗿𝗲 𝗳𝗼𝗿 𝗳𝗿𝗲𝗲: https://lnkd.in/dbf74Y9E
Kudos for the graphic goes to Generative AI ! | 45 comments on LinkedIn
Die 3 Level der KI-Integration - AI-Adoption, AI-Adaption & AI-Transformation
🎥 Adoption, Adaption, Transformation: Wie KI unsere Arbeitswelt verändert! 🤖✨🧠 In diesem Video tauchen wir in die Welt der KI-Integration ein und beleucht...
𝗥𝗲𝗶𝗰𝗵𝘄𝗲𝗶𝘁𝗲𝗻-𝗞.𝗢. 𝗳ü𝗿 𝘃𝗶𝗲𝗹𝗲 𝗪𝗲𝗯𝘀𝗶𝘁𝗲𝘀. Erst wandern Suchanfragen von Google zu ChatGPT - jetzt beantwortet sie Google direkt in den AI-Overviews.
🚨 Studien zeigen bereits hohe Traffic-Rückgänge.
Was können Redakteure und Publisher tun?
👊 Deshalb bin ich mit Matthäus Michalik in den Podcast-Ring gestiegen:
Wir haben 2 Folgen aufgenommen: 𝗚𝗘𝗢 𝘀𝘁𝗮𝘁𝘁 𝗦𝗘𝗢 & 𝗪𝗶𝗲 𝗽𝗹𝗮𝘁𝘇𝗶𝗲𝗿𝗲𝗻 𝘄𝗶𝗿 𝘂𝗻𝘀 𝗶𝗻 𝗱𝗲𝗻 𝗔𝗜-𝗢𝘃𝗲𝗿𝘃𝗶𝗲𝘄?
Als Teaser für euch:
4️⃣ Sofort-Tipps für GEO (Generative Engine Optimization)
1. Autorität & Vertrauen belegen
🔸 Quellen, Zitate und fachliche Referenzen explizit nennen.
🔸Ergebnis: bis zu +40 % höhere Wahrscheinlichkeit, in KI-Antworten zitiert zu werden.
2. Zahlen sprechen lassen
🔸Statistiken, Studien-Daten und eigene Benchmarks einbauen.
🔸KI-Modelle gewichten quantitative Infos stärker → +30 % Relevanz-Boost.
3. Klare Struktur, einfache Sprache
🔸Kurze Absätze, Bullet-Points, FAQs, sprechende Zwischenüberschriften.
🔸Erleichtert Parsing durch LLMs und erhöht die Chance auf direkte Übernahme.
4. Gezielter Fachwort-Einsatz
🔸Relevante Terminologie und Branchen-Jargon bewusst einstreuen.
🔸Signalisiert Expertise und verbessert das Matching für spezifische Nutzeranfragen.
‼️ Kurzformel: Autorität + Daten + Klarheit + Terminologie = Sichtbarkeit Chat-Antworten.
𝗦𝗶𝗰𝗵𝘁𝗯𝗮𝗿𝗸𝗲𝗶𝘁 𝗶𝗻 𝗔𝗜 𝗢𝘃𝗲𝗿𝘃𝗶𝗲𝘄𝘀 – 𝗱𝗮𝘀 𝗺𝘂𝘀𝘀𝘁 𝗱𝘂 𝗯𝗲𝗮𝗰𝗵𝘁𝗲𝗻
🔸Grundvoraussetzung: Deine Seite muss im Google-Index stehen und bereits ein gewisses Vertrauensniveau besitzen. Dann gilt:
🔸Hochwertige, faktenbasierte Inhalte: präzise, recherchiert, aktuell.
🔸Klare Struktur: H-Überschriften, Listen, Tabellen → erleichtert Parsing.
🔸Strukturierte Daten (Schema.org): zeigt der KI, was welche Bedeutung hat.
🔸UX & Performance: schnelle Ladezeiten, sauberes Mobile-Design.
🔸E-E-A-T pflegen: Expertise, Erfahrung, Autorität, Vertrauen kontinuierlich belegen (Autorenprofile, Quellen, Backlinks).
𝟴 𝗣𝗿𝗮𝘅𝗶𝘀-𝗧𝗶𝗽𝗽𝘀 𝗳ü𝗿 𝗱𝗶𝗲 𝗣𝗼𝘀𝘁-𝗦𝗘𝗢-Ä𝗿𝗮
✔️ Qualität vor Quantität – fewer, deeper pieces mit klarer Expertise.
✔️Struktur first – H-Tags, Bullet-Points, FAQ-Blöcke, Schema.
✔️User Experience optimieren – Speed, Navigation, mobile UX.
✔️Mehrwert über die KI hinaus – eigene Daten, Cases, Meinungen.
✔️Traffic-Quellen streuen – Social, E-Mail, Communities, Partnerschaften.
✔️Monitoring & Anpassung – beobachte, welche Seiten in AI Overviews landen, und iteriere.
✔️Multimedial denken – Videos, Podcasts, Infografiken ergänzen Text.
✔️E-E-A-T kontinuierlich stärken – Fachautor:innen, Referenzen, Reviews, Backlinks.
𝗞𝘂𝗿𝘇𝗳𝗼𝗿𝗺𝗲𝗹: Qualität + Struktur + Mehrwert + Vertrauen + Channel-Mix = langfristige Sichtbarkeit – auch in der KI-Suche.
❓ Wie geht ihr den Battle um Sichtbarkeit und Traffic an? Lasst uns diskutieren. 👇 | 12 Kommentare auf LinkedIn