Tech It From Me

Open-Source AI vs. Big Players: Can Mistral, Ollama and LLaMA Compete?

Play Episode Pause Episode

Mute/Unmute Episode Rewind 10 Seconds 1x Fast Forward 30 seconds

00:00 / 9:41

Amazon Apple Podcasts Spotify YouTube

RSS Feed

Share

Link

Embed

Why Open-Source AI Matters

Open-source AI isn’t just a cheaper alternative to big tech models—it’s a philosophical and strategic shift in how artificial intelligence is developed, deployed, and shared. At its core, open-source AI democratizes access to cutting-edge technology, putting advanced tools in the hands of individuals, startups, educators, and researchers who might otherwise be locked out due to cost or licensing restrictions.

When a company like Meta releases LLaMA or a group like Mistral AI shares its model weights, they’re giving the community not just the software but the blueprint. This transparency allows for:

• Full inspection – Developers can review training methods, datasets, and parameters to understand precisely how the model works, a key step in building trust.
• Customization and innovation – Organizations can fine-tune these models for specific industries—healthcare, legal, finance—without sending sensitive data to external APIs.
• Education and research – Universities and independent researchers gain access to real-world, production-grade models for teaching and experimentation, driving academic advancement.

This mirrors the impact of past open-source revolutions. Linux became the backbone of the internet because anyone could contribute to its code and adapt it for their needs. Apache made web hosting accessible without expensive licensing fees. Today, Hugging Face plays a similar role for AI, hosting open models and fostering a collaborative ecosystem.

From a global perspective, open-source AI could help bridge the AI divide between wealthy nations and developing countries. Instead of relying entirely on expensive cloud-based APIs from Silicon Valley, local teams can run their own AI models, adapt them for cultural and linguistic nuances, and build region-specific solutions.

Of course, this openness comes with challenges—security risks, maintenance burdens, and the need for technical expertise—but it also represents one of the most promising pathways to a more equitable AI future.

For a deeper look at how we got to this point, check out The Real History of AI: From Turing to Transformers.

The Big Players vs. the Open Challengers

The AI landscape in 2025 feels like a heavyweight boxing match. In one corner, you have the big players—OpenAI with GPT-5, Google DeepMind with Gemini Ultra, and Anthropic with Claude 3.5—armed with billion-dollar budgets, proprietary datasets, and fleets of high-performance GPUs. In the other corner, a growing roster of open-source challengers—Mistral, Meta’s LLaMA, Falcon, StableLM—who are rewriting the playbook on how AI is built and distributed.

What Gives the Big Players Their Edge?

The giants of AI benefit from:

• Massive Compute Power – Access to thousands of cutting-edge GPUs like the Nvidia H100, often running 24/7 in specialized AI data centers.
• Proprietary Data – Exclusive, high-quality datasets—sometimes acquired through partnerships or licensing—that aren’t available to the public.
• Polished Ecosystems – Seamless integration with tools like Microsoft Office, Google Workspace, and cloud APIs makes adoption easy for enterprise customers.
• Dedicated Research Teams – Hundreds (or thousands) of full-time AI researchers pushing model capabilities to new heights.

These advantages mean that, for now, proprietary models tend to lead in raw performance, reasoning ability, and multi-modal integration (text, image, and audio).

What the Open Challengers Bring to the Table

While the big players dominate in infrastructure, open-source AI brings unique strengths:

• Agility and Innovation – Without corporate red tape, open-source teams can experiment and release updates quickly.
• Community-Driven Development – Thousands of independent contributors worldwide collaborate on model improvements, fine-tuning, and bug fixes.
• Accessibility – Anyone can download, run, and adapt these models, from a single GPU workstation to an on-premise server.
• Lower Cost Barriers – Instead of paying per-token usage fees, organizations can self-host and control expenses.

In some cases, open-source projects have even leapfrogged big tech in niche areas. For example, smaller models optimized for local inference can outperform massive proprietary models when latency, privacy, or cost is a top priority.

The Real Competition Isn’t Just About Speed or Accuracy

While benchmark leaderboards matter, the battle between big tech and open-source is increasingly about trust, control, and sustainability.

• Enterprises are asking: Do we want to send sensitive data to an external API?
• Developers are asking: Do we want to be locked into one provider’s pricing and policies?
• Governments are asking: Who should control the infrastructure of AI?

This growing tension is why both sides are watching each other closely. Big tech companies are beginning to incorporate open-source principles into certain releases, while open challengers are finding ways to leverage commercial partnerships for funding and compute resources.

For a closer look at the cost and infrastructure side of this arms race, check out The Real Cost of AI: Who’s Paying for the Compute Arms Race?.

Where Open-Source AI Wins

While proprietary models often steal the spotlight with jaw-dropping demos and billion-parameter bragging rights, open-source AI quietly wins in areas that matter most to businesses, developers, and researchers looking for control and flexibility.

1. Customization Without Limits

With open-source AI, you’re not stuck with a “one-size-fits-all” solution. You can fine-tune the model on your data, tailor it for a specific industry, or even optimize it for a single task. For example:

• A healthcare provider could adapt an open model for medical terminology and privacy compliance.
• A financial services firm could build a specialized chatbot that understands regulatory constraints without sending data to an external API.

Proprietary AI may offer customization via APIs, but it usually comes with hefty costs, rate limits, and data usage restrictions.

2. Cost Control and Predictability

One of the biggest frustrations with proprietary AI is the unpredictable—and often escalating—per-token billing model. Open-source AI changes that. By self-hosting models on local servers or private cloud instances, you can:

• Avoid per-query fees that pile up at scale.
• Repurpose existing infrastructure, like idle GPU servers.
• Budget for fixed hardware and energy costs instead of variable API charges.

This mirrors the cost-saving mindset seen in other tech areas, such as when some small IT teams choose Cloudflare over traditional firewalls.

3. Data Privacy and Control

In regulated industries—finance, healthcare, legal—sending sensitive data to an external AI provider can be a compliance nightmare. Open-source AI allows you to keep both the model and data inside your organization’s secure environment. This is a massive win for:

• Meeting GDPR and HIPAA requirements.
• Protecting proprietary research and intellectual property.
• Avoiding vendor lock-in that can lead to costly migrations later.

4. Transparency Builds Trust

With open-source models, you can audit the training process, inspect datasets, and verify ethical safeguards. This level of transparency isn’t possible with black-box proprietary systems. And in an era where deepfakes and misinformation are on the rise (see our piece on AI watermarking), that transparency matters more than ever.

5. Community Innovation Outpaces Corporate Timelines

Big tech companies follow roadmaps tied to product launches, investor calls, and quarterly earnings. Open-source AI, however, evolves at the speed of the community—sometimes releasing major updates or breakthrough features within weeks, not months.

Projects like Hugging Face Transformers have shown how a passionate global developer base can maintain, improve, and scale AI tools far faster than traditional corporate cycles.

The bottom line: Open-source AI may not yet beat proprietary models on every benchmark, but it consistently wins on flexibility, cost efficiency, and ethical transparency—and in the long game, those advantages could prove more transformative than raw compute power.

Where They Still Lag

For all its advantages, open-source AI still faces some steep uphill battles before it can fully match the capabilities—and market dominance—of proprietary giants like GPT-5, Claude 3.5, and Gemini Ultra.

1. Performance Gap on Cutting-Edge Benchmarks

While open-source models are making rapid progress, they often trail proprietary models in areas like:

• Complex reasoning across multiple steps or domains.
• Multi-modal integration—seamlessly combining text, images, and audio in a single query.
• Long-context handling, where big players’ models can process hundreds of thousands of tokens without losing coherence.

Many open models are optimized for smaller hardware footprints, which is great for accessibility but can limit raw performance.

2. User Experience & Integration

Proprietary AI platforms invest heavily in user-friendly interfaces and turnkey integrations with popular tools like Microsoft Office, Google Workspace, and enterprise CRM systems.
By contrast, open-source AI often requires:

• Command-line setup or manual configuration.
• Familiarity with Python, Docker, or specialized ML frameworks.
• Additional development work to connect the model to business workflows.

This can be a barrier for non-technical teams who need AI solutions that “just work” out of the box.

3. Resource Demands for Self-Hosting

Running an open-source LLM locally isn’t free—you still need:

• High-end GPUs (or expensive cloud compute) for fast inference.
• Reliable storage and networking infrastructure.
• Ongoing maintenance to keep the model secure and up to date.

This is where many organizations realize that while the model itself may be free, the total cost of ownership can still be significant. For more on the economics behind this, see The Real Cost of AI: Who’s Paying for the Compute Arms Race?.

4. Limited Enterprise Support

If a proprietary AI model goes down, customers can usually call a support line or file a priority ticket backed by an SLA. Open-source AI relies on:

• Community forums
• GitHub issues
• Volunteer maintainers

While these communities can be incredibly responsive, there’s no guarantee of 24/7 coverage or rapid resolution—something enterprise clients often require.

5. Security Risks and Abuse Potential

Open weights make it easier for bad actors to fine-tune models for malicious purposes, from generating deepfake content to creating phishing scripts that evade detection.
This risk is a growing concern for governments, businesses, and security researchers alike—and it’s why many are exploring model watermarking and AI-specific threat detection (see our coverage here).

The takeaway: Open-source AI is closing the gap quickly, but the last mile—performance parity, seamless usability, enterprise-grade support, and security hardening—remains a formidable challenge.

The Role of Compute & Infrastructure

Whether you’re running GPT-5 in the cloud or deploying LLaMA 3 on a local server, AI performance is ultimately limited by the hardware and infrastructure behind it. The difference is in who pays for—and who controls—that infrastructure.

Why Compute Power Is the True Bottleneck

Large language models (LLMs) are hungry. Even “lightweight” open-source models can require:

• High-performance GPUs such as Nvidia A100s or H100s for fast inference.
• Hundreds of gigabytes of RAM and storage for hosting model weights and datasets.
• Robust cooling and power systems to keep data centers running 24/7.

The most prominent players in AI have access to massive, specialized AI clusters purpose-built for model training and deployment. Open-source developers, on the other hand, often rely on a mix of community-donated compute, cloud credits, or smaller-scale local setups.

Cost Isn’t Just About Hardware

Even if you own the GPUs, you still face ongoing costs for:

• Electricity consumption – Running a high-end AI server is like keeping a powerful gaming PC on full blast all day.
• Network bandwidth – AI models often move vast amounts of data between compute nodes.
• Maintenance & updates – From replacing failed components to keeping dependencies up-to-date.

For a deeper look at the economics, check out The Real Cost of AI: Who’s Paying for the Compute Arms Race?.

The Cloud vs. On-Prem Trade-Off

Many organizations choose between:

• Cloud hosting – Faster setup, scalable resources, and lower upfront costs, but ongoing usage fees and potential data privacy concerns.
• On-premises hosting – Higher initial investment in hardware but full control over data, customization, and long-term costs.

Open-source AI offers flexibility here—you decide where to run it. Proprietary AI services typically lock you into their cloud platform.

A Growing Need for Specialized AI Infrastructure

We’re also seeing a rise in AI-optimized hardware beyond just GPUs:

• AI-specific accelerators like Google’s TPUs.
• Edge AI devices that run smaller models close to the data source for faster, offline inference.
• AI networking solutions optimized for high-bandwidth, low-latency communication between servers.

As both open-source and proprietary models scale, compute capacity may become the new competitive currency—and organizations that invest early could gain a significant advantage.

The Security Question

As AI models—both proprietary and open-source—become more capable, security concerns are moving to the forefront. For open-source AI, the same transparency that fuels innovation can also open the door to misuse, manipulation, and exploitation.

The Double-Edged Sword of Open Weights

When an AI model’s weights and architecture are freely available, anyone can:

• Fine-tune it for beneficial, specialized use cases (like medical diagnosis or legal research).
• Or… fine-tune it for malicious purposes, such as generating convincing phishing emails, creating deepfake videos, or bypassing content moderation filters.

While proprietary models also face misuse risks, their closed nature makes it harder for bad actors to directly alter the model’s behaviour at the code level.

Model Poisoning: The Hidden Threat

One of the emerging risks is model poisoning, where malicious actors subtly modify an AI model during training or fine-tuning to introduce hidden biases or vulnerabilities. This can:

• Cause the model to behave incorrectly under specific conditions.
• Leak sensitive information that was never intended to be stored.
• Weaken defences against harmful outputs.

In large, collaborative open-source projects, guarding against poisoning requires strong community oversight, code reviews, and automated testing pipelines.

Data Privacy and Compliance Risks

For enterprises, another concern is how AI handles sensitive data. Without strict controls, an open-source model could unintentionally memorize and regurgitate private information from its training data.
That’s why secure deployment practices—like running AI behind the firewall—are critical for regulated industries. We discussed similar privacy concerns in How Cybercriminals Really Get Your Info.

The Push for AI Watermarking and Traceability

To combat misuse, researchers are exploring watermarking—embedding hidden signals in AI-generated content to help identify its source. This could:

• Assist in detecting misinformation campaigns.
• Help businesses verify the authenticity of content.
• Deter malicious actors by reducing plausible deniability.

For a deeper dive, see Watermarking AI: Will It Change the Way We Write Forever?.

Why Security Could Decide the Winner

In the battle between open-source and proprietary AI, security might become the ultimate differentiator. Enterprises will gravitate toward the option that offers:

• Clear guardrails against misuse.
• Strong governance frameworks.
• Compliance with emerging AI regulations worldwide.

If open-source communities can solve these challenges at scale, they may not just compete with big tech—they could redefine trust in AI.

The Road Ahead

The open-source vs. proprietary AI debate isn’t a zero-sum game—it’s a race toward two very different visions of the future. On one side, big tech will continue building massive, multi-modal AI systems with unmatched scale, integration, and enterprise polish. On the other hand, open-source AI will focus on transparency, adaptability, and putting powerful tools in the hands of the many rather than the few.

A Likely Hybrid Future

Much like the software ecosystem evolved with Linux and Windows coexisting, we can expect a hybrid AI landscape:

• Proprietary AI is leading in general-purpose consumer applications, cloud-hosted enterprise solutions, and high-stakes AI research.
• Open-source AI is thriving in specialized, privacy-sensitive, and cost-conscious use cases where customization and control matter most.

The Catalysts That Will Shape the Outcome

Several factors will determine how this balance plays out:

• Compute accessibility – Will affordable, high-performance AI hardware become mainstream?
• Security breakthroughs – Can open-source AI establish trust without limiting innovation?
• Regulation – How will governments approach AI governance, and will open-source models be regulated differently?
• Community momentum – Will the global developer ecosystem keep pace with proprietary advancements?

What This Means for Businesses and Developers

For organizations, the takeaway is clear: Don’t lock yourself into a single AI strategy. Explore both ecosystems, run pilots, and evaluate where each approach fits your needs.

• If speed to market and out-of-the-box performance are critical, proprietary AI may be the better starting point.
• If cost control, privacy, and domain-specific customization matter most, open-source AI could be the long-term winner.

For developers, the next few years will be an unprecedented opportunity to shape the direction of AI itself—either by contributing to open-source projects, innovating on top of them, or integrating them with proprietary platforms in creative ways.

Final Thoughts

Open-source AI has already proven it can stand toe-to-toe with big tech in certain arenas. The question isn’t whether it can compete—it’s how much of the future it will own.

For more insights on the evolution of technology leadership and AI’s role in business, check out From Y2K to AI: How IT Departments Have Changed and Where They’re Headed, or explore The Real Cost of AI for a deeper dive into the economics driving the AI arms race.

(Feature image generated with the help of DALL-E)

Sam Altman says the leap from GPT-4 to GPT-5 is so big that going back feels “miserable.” But is GPT-5 really the iPhone moment for AI—or is it just the latest upgrade in an already fast-moving race?

I’ve already spent some time using the new version, and I can confirm: many of the improvements are noticeable right away. The model feels faster, more confident in its answers, and much better at staying on track in more extended conversations. It’s still not perfect, but it’s a more enjoyable and productive experience.

Let’s unpack what’s new, what’s useful, and whether this release changes the game.

What’s New in GPT-5?

OpenAI officially rolled out GPT-5 to all users on Friday, August 8, 2025, marking what the company calls its most advanced ChatGPT release yet. For Pro subscribers, the upgrade is even bigger—you now get access to GPT-5, GPT-5-Pro, and GPT-5-Thinking, each tuned for different levels of depth and reasoning.

So, what exactly has changed? Quite a bit:

1. Smarter, Context-Aware Reasoning
GPT-5 takes “chain-of-thought” logic to the next level, breaking down complex problems into smaller, logical steps—without you having to spell it out. This means more accurate answers, fewer logical gaps, and better handling of multi-part queries.

2. Dramatically Fewer Hallucinations
While no AI is perfect, GPT-5 shows a noticeable reduction in factual errors. In my testing, I’ve seen cleaner, more consistent outputs that stay true to source material—especially when summarizing technical or niche topics.

3. ‘Vibe Coding’ for Developers
This is a game-changer for coders. Instead of simply producing functional code, GPT-5 can now interpret how you want your code to feel—whether that’s clean and minimalist, performance-optimized, or playful and experimental. It bridges the gap between intent and execution.

4. Unified Model Routing
Say goodbye to manually switching between browsing mode, code interpreter, or GPT-4 for different tasks. GPT-5 automatically chooses the right capabilities in the background, making the experience seamless and letting you focus on your work.

5. Deeper App Integrations
You can now connect GPT-5 directly to tools like Gmail, Google Calendar, and potentially more productivity platforms in the near future. This turns ChatGPT from a standalone assistant into something closer to a true personal AI workspace.

6. Custom Personalities with Memory
Want an AI writing partner with your exact tone? Or a coding assistant who remembers your preferred frameworks? GPT-5 now allows you to create named GPTs with persistent memory, so each assistant gets better the more you use it.

From my own use, two features stand out: Unified Model Routing, which makes every interaction feel frictionless, and Vibe Coding, which is almost eerie in how well it understands intent. For anyone who works with AI regularly, these aren’t just quality-of-life improvements—they’re workflow accelerators.

Is This Really the iPhone Moment?

Every major AI release seems to get compared to a tech milestone—Apple’s original iPhone launch, the debut of the Retina display, or even the first personal computer. With GPT-5, the buzz is no different. Some have already called it the “iPhone moment” for AI, suggesting it could be a turning point in how we interact with technology (source: The Verge).

But is that really the case?

If we look at the AI timeline, GPT-3 was arguably the real “wow” moment for the public—it introduced mainstream users to just how human-like a language model could feel (see my breakdown of AI’s history here). GPT-4 refined that experience, making it more reliable and capable. GPT-5? It’s more of a polished evolution than a radical leap. Think going from an iPhone 13 to an iPhone 15 Pro Max—sleeker, faster, with smarter features, but not a reinvention.

Here’s where I think the hype meets reality:

• Yes, GPT-5 is better. You’ll notice faster responses, better reasoning, and smarter integrations right away.
• No, it’s not magic. The leap isn’t as dramatic as some suggest—at least not yet.
• The AGI hype needs perspective. We’ve been here before with Web3, metaverse promises, and blockchain revolutions that fizzled (see my take on AI hype cycles).

That said, I can tell from my testing: it’s hard to go back to GPT-4 once you’ve used GPT-5 for a while. It feels like a more responsive, more intuitive assistant—one that’s inching closer to being a proper daily driver for professionals.

Where GPT-5 Actually Helps

While the headlines focus on bigger, brighter, and more human-like, the real question for most of us is: Where does GPT-5 actually make a difference in day-to-day work?

After using GPT-5 extensively across IT, writing, and research tasks, I’ve seen several clear advantages over GPT-4:

1. For Developers
With vibe coding and improved reasoning, GPT-5 doesn’t just spit out functional code—it writes in your preferred style, anticipates potential pitfalls, and suggests optimizations before you even ask. This means fewer rewrites and cleaner commits (see my breakdown on coding with AI here).

2. For Writers & Content Creators
From blog posts to video scripts, GPT-5 delivers clearer drafts that stay on-topic and require less editing. Its improved contextual memory also helps keep tone and style consistent across longer pieces—something GPT-4 often struggled with.

3. For Researchers & Analysts
Whether you’re summarizing dense reports or pulling insights from large datasets, GPT-5 processes and condenses information faster, with fewer hallucinations. This means you can spend more time thinking critically about the results instead of verifying every line (related: How AI can mislead if unchecked).

4. For Busy Professionals
The new Gmail and Google Calendar integrations mean you can draft emails, plan meetings, and generate follow-up notes—all without leaving ChatGPT. It’s not just answering questions; it’s helping you act on them.

5. For Long, Complex Requests
One of GPT-5’s most underrated upgrades is focus retention. Even in multi-step prompts, it stays on track instead of drifting into irrelevant tangents. This is especially valuable for IT workflows where one missed step can cause significant delays (see my AI adoption advice for IT leaders).

Bottom line: GPT-5 isn’t replacing knowledge workers anytime soon—but it shifts your workflow into a higher gear. You still need to steer; GPT-5 makes the road smoother and the trip faster.

You still need to drive the process. GPT-5 shifts it into a faster gear.

What This Means for IT Leaders & Businesses

For IT leaders, CIOs, and business owners, the GPT-5 release isn’t just another AI headline—it’s a signal that practical AI adoption is accelerating. But here’s the key takeaway: you don’t need to overhaul your entire stack tomorrow. Instead, now is the time to strategically experiment, govern, and integrate.

Here’s what I’d be thinking about right now:

1. Employee Readiness & Training
Your teams may have access to GPT-5 already, but do they know how to use it effectively—and securely? AI productivity gains come when staff are trained not just in prompt writing, but in understanding limitations, bias, and data privacy (see my AI job impact breakdown).

2. Governance & Compliance
Smarter AI doesn’t mean safer AI. Attackers will adapt to new models just as quickly as businesses adopt them (Gartner warns about AI security risks). Consider:

• Do you need a formal policy for AI-generated code and content?
• How will you audit AI-assisted decisions for compliance and accuracy?
• Who is responsible for AI usage oversight in your org?

3. Integration into Existing Workflows
With GPT-5’s new app integrations (Gmail, Google Calendar, and more to come), you can start embedding AI where your teams already work. Early pilot projects can uncover efficiency wins without a massive system change.

4. Cost-Benefit Analysis
Upgrading to GPT-5 Pro for your team could be a productivity multiplier—but you need to weigh it against licensing costs, usage limits, and security trade-offs. Many organizations are experimenting with a “center of excellence” model where a small AI-proficient group leads adoption and shares learnings internally.

5. Industry Benchmarking
If your competitors are using GPT-5 to speed up product development, improve customer support, or automate analysis, waiting too long could mean falling behind (see my Y2K-to-AI leadership lessons).

In short, GPT-5 is an opportunity, not a mandate. The businesses that will benefit most are those that experiment deliberately, govern wisely, and train their people first.

Are We Closer to AGI?

Whenever OpenAI drops a major update, the AGI (Artificial General Intelligence) conversation comes roaring back. GPT-5 is no exception—some headlines are already hinting that we’re on the brink of machines that “think” like humans (MIT Technology Review covers the debate).

But let’s pump the brakes.

OpenAI itself defines AGI as “highly autonomous systems that outperform humans at most economically valuable work” (source). GPT-5, while more capable than its predecessors, is still a specialized AI—brilliant at certain language and reasoning tasks, but not independently creative, curious, or self-motivated in the human sense.

Why GPT-5 Feels Closer to AGI

• Improved reasoning chains mean it can handle more complex, multi-step requests.
• Better memory and personalization make it feel more “aware” of you over time.
• Unified tool routing gives it the illusion of adaptability—choosing the best mode automatically.

Why We’re Still Not There

• GPT-5 has no true understanding of the world—only statistical associations.
• It lacks generalized goals outside of responding to prompts.
• It can still produce confident but factually wrong answers (hallucinations), even if less often than GPT-4 (I’ve covered the risks here).

In other words, GPT-5 is a powerful narrow AI with more polish and adaptability, but AGI remains a moving target—one that experts can’t agree will arrive in 5 years or 50 (see AI timeline debates from Stanford HAI).

The danger in treating GPT-5 like AGI is complacency: over-trusting its outputs, underestimating bias, and skipping human oversight. The real opportunity is using it as an augmented intelligence tool—boosting human decision-making without replacing it.

Final Thoughts

GPT-5 is, without question, the most capable version of ChatGPT yet. It’s faster, smarter, more focused, and—thanks to integrations—more useful in everyday workflows. For developers, writers, researchers, and IT leaders alike, the leap in practical productivity is real.

But it’s also important to keep perspective. GPT-5 isn’t magic, and it’s not AGI. Like every AI upgrade before it, the novelty will eventually fade, and what will matter most is how effectively you integrate it into your work. The real winners in this AI race won’t be the companies with the newest tools—they’ll be the ones that train their people, set strong governance, and adopt AI deliberately.

If you’re curious about where AI might head next, I’ve explored that in The Real History of AI: From Turing to Transformers and my breakdown of AI’s impact on the future of work.

In the meantime, GPT-5 is here, and it’s ready to work—if you are. Whether it becomes the iPhone moment for AI or just another step on the road to AGI will depend on how we, as users and leaders, choose to apply it.

Related Posts from Tech It From Me:

• Behind the Buzzwords: What Is a Large Language Model, Really?
• Will AI Really Replace All Mundane Jobs? My Take on Geoffrey Hinton’s Bold Claim
• The Real History of AI: From Turing to Transformers

(Feature image generated with the help of DALL-E.)

Rather than pushing for a single, centralized artificial general intelligence (AGI) — the “one AI to rule them all” approach — Zuckerberg outlined a bold new vision: a personalized super-AI for every individual.

Welcome to the age of personal superintelligence — and possibly, a new chapter in the AI arms race.

Tech It From Me

The Case for Personalized Superintelligence: Zuckerberg vs the AI Mainstream

Play Episode Pause Episode

Mute/Unmute Episode Rewind 10 Seconds 1x Fast Forward 30 seconds

00:00 / 16:56

Amazon Apple Podcasts Spotify YouTube

RSS Feed

Share

Link

Embed

Not One AI, But Billions of Them

For years, the prevailing narrative in AI has been the pursuit of a single, centralized artificial general intelligence — the “superbrain” that could answer any question, solve any problem, and serve the entire planet. Think of it as the Google Search of intelligence — one source, everyone taps in.

Zuckerberg’s vision flips that model on its head. Instead of one AI serving billions of people, he imagines billions of AIs serving one person each.

“Why share a general-purpose brain,” the logic goes, “when you could have one that knows you better than anyone else?”

Your personal AI could:

• Understand your schedule, routines, and energy patterns
• Mirror your tone, vocabulary, and sense of humour in every message
• Automatically sync with your social feeds, work tools, and AR glasses

Meta’s infrastructure already provides the connective tissue for this vision. Between WhatsApp, Messenger, Instagram, and Threads, they have a global user base of over 3.2 billion people — and all the cross-platform data streams to feed a highly personalized model.

It’s a leap from generic intelligence to deeply contextual intelligence — closer to a lifelong digital twin than a disposable chatbot session.

And if that sounds like something ripped from a sci-fi script, it’s because it is. From Her to Black Mirror, the idea of a constant AI companion has been a recurring trope. The difference now? The infrastructure to make it real already exists — from large language models to wearable displays and even early brain-computer interface research.

Why This Is a Big Deal

Most of today’s best-known AI tools — ChatGPT, Google Gemini, Anthropic Claude — are brilliant generalists. They can write an essay, debug code, or explain quantum physics. But here’s the catch:

They don’t know you. Every conversation starts from scratch. You’re reintroducing yourself over and over, like meeting a stranger with amnesia.

Zuckerberg’s personalized superintelligence vision flips that model. Instead of an AI that knows everything in general, you’d have one that knows you in particular:

• Context-aware: It remembers your goals, tone, and preferences.
• Memory-rich: It recalls last week’s conversation about your upcoming trip or that tricky project at work.
• Proactive: It takes initiative — preparing a presentation before you ask, flagging conflicts in your schedule, or suggesting a weekend plan based on your past choices.

It’s not just a smarter assistant — it’s the beginnings of a second brain.

If you want a deeper understanding of how these systems work at the model level, I break that down in Behind the Buzzwords: What Is a Large Language Model, Really?.

The possibilities are huge:

• Summarizing your meetings while you’re still in them
• Writing reports in your voice and style
• Coaching your presentations in real time
• Helping your kids with homework based on their learning styles

But here’s the price tag for that level of personalization: your AI would need access to almost everything about you — your files, private messages, biometric data, location history, even your voice and facial ID.

That level of intimacy is both its superpower and its Achilles’ heel. With that much of “you” inside one system, the question becomes:

Who do you trust to hold that much of yourself?

Meta, Microsoft, Apple, Google… and OpenAI

Zuckerberg’s personal superintelligence pitch doesn’t exist in a vacuum. It’s part of a much bigger battle for AI dominance — one that will define not just who leads the market, but how we interact with machines for the next decade.

Here’s how the other tech titans are playing their cards:

• Microsoft is betting on workplace supremacy. Its Copilot is embedded across Excel, Word, Outlook, and Teams, making AI an invisible productivity layer in the daily workflow of millions. As I explored in Microsoft Joins the $4 Trillion Club, this AI-first pivot is already paying off in valuation and market share.
• Apple is leaning into privacy as a selling point. With Apple Intelligence, processing happens on-device whenever possible, minimizing cloud exposure and appealing to users who distrust big data collection.
• Google is trying to weave Gemini into everything — from Search to Gmail to Android — but its strategy is split between innovating for users and protecting its ad revenue machine.
• OpenAI is staying the course toward AGI — a single, universal intelligence that can do everything for everyone. I unpack the history and ambition behind this in The Real History of AI: From Turing to Transformers.

Meta’s gamble is that people won’t just want a chatbot that answers questions — they’ll want a lifelong AI companion. A digital twin that evolves with them, remembers their milestones, and adapts to their changing needs.

If Microsoft’s Copilot is a co-worker, Apple’s AI is a privacy-first concierge, and Google’s Gemini is a search-enhanced assistant, then Meta wants to be your second self — always learning, always present, and always in your pocket (or on your face via AR glasses).

Goodbye Open Source?

Here’s where Zuckerberg’s vision takes a sharp turn.

For years, Meta positioned itself as the champion of open-source AI, releasing LLaMA models that researchers, startups, and even competitors could adapt freely. That openness was a major differentiator from the walled gardens of OpenAI, Google, and Apple.

Now, Zuckerberg says that era is over.

The reason? “Safety.” Meta claims that keeping future models closed will reduce the risk of misuse — such as generating deepfakes, automated cyberattacks, or large-scale disinformation campaigns.

But critics aren’t buying the whole story. Some argue this pivot is less about safety and more about control — keeping Meta’s next-gen models exclusive to its ecosystem and locking in a competitive edge.

This debate isn’t new. The tension between innovation and control has been simmering for years, especially as AI becomes more powerful. I’ve explored similar dynamics in Watermarking AI: Will It Change the Way We Write Forever?, where we looked at how transparency tools can both protect users and stifle creativity.

The irony is hard to miss:

• Open source fuels innovation — countless AI breakthroughs came from publicly available research and models.
• Closed systems protect IP — and keep dangerous capabilities out of the wrong hands.

The question is whether Meta can convince the world that closing the door on open AI is a step toward safety, not monopoly.

The Future: Helpful or Creepy?

The idea of a personal AI that knows you inside out sounds like a dream to some — and a nightmare to others.

On one hand, it’s the ultimate productivity tool: a tireless assistant that anticipates your needs, manages your time, and even shields you from digital clutter. Imagine waking up to an AI that has already sorted your inbox, rescheduled your meeting to avoid traffic, and prepped a personalized workout based on your sleep data.

On the other hand, it’s the perfect surveillance device — one that sees, hears, and remembers everything you do. And once that data exists, the risk of misuse, hacking, or corporate overreach becomes very real.

We’ve already seen how technology can cross the line:

• Facial recognition systems are deployed without consent in public spaces
• Algorithmic bias is influencing job offers, policing, and even medical treatment
• Phishing campaigns that exploit personal data — something I break down in How Cybercriminals Really Get Your Info

This is where Zuckerberg’s vision straddles the line between empowering and manipulative.

A personal AI could feel like an ally — or like a Black Mirror episode waiting to happen. The difference will come down to:

• Governance: Who sets the rules for how these systems operate?
• Consent: Do users truly understand what they’re agreeing to?
• Control: Can you turn it off — and can you delete everything it knows about you?

Because once an AI becomes your “second self,” the stakes aren’t just technical. They’re personal.

Conclusion

Whether you’re fascinated or unnerved by Zuckerberg’s “AI for everyone” vision, one thing is clear: the AI race is no longer just about building smarter models — it’s about controlling the interface between humans and machines.

The real power isn’t in the code alone. It’s in who gets to mediate your decisions, shape your habits, and filter your reality.

The next decade will be defined by choices being made right now — in corporate boardrooms, in policy discussions, and in the way each of us adopts (or rejects) these tools.

If your company still treats AI as “just another chatbot,” it’s already behind. If you’re still thinking of AI as something separate from your daily life, you may be surprised how quickly it becomes woven into it — by choice or by default.

Because whether your AI lives in your phone, your glasses, or someday in your neural implant, the most important question will remain the same:

Is it working for you — or are you working for it?

(Feature image generated with the help of DALL-E)

Every day, regular people—students, small business owners, retirees, even IT professionals—fall victim to cybercrime. And it’s not because they’re stupid, careless, or unlucky. It’s because today’s cyberattacks are automated, convincing, and wildly effective.

In this post, I’m breaking down how cybercriminals get your info, based on real-world experience from almost three decades in IT leadership roles. I’ll also give you 5 simple things you can do today to reduce your risk—no technical expertise required.

Tech It From Me

Tech It to the Bank: How Cybercriminals Really Get Your Info

Play Episode Pause Episode

Mute/Unmute Episode Rewind 10 Seconds 1x Fast Forward 30 seconds

00:00 / 16:37

Amazon Apple Podcasts Spotify YouTube

RSS Feed

Share

Link

Embed

The Most Common Ways Hackers Steal Your Data

Let’s ditch the Hollywood imagery. Nobody is hammering away at your firewall by hand. Cybercriminals use automation to compromise thousands of people at once—and they’re incredibly efficient at it.

Here’s how:

1. Phishing

Fake emails or texts that mimic trusted brands like Amazon, PayPal, or Microsoft. They get you to click a link, log in, and boom—your credentials are stolen.

2. Social Engineering

You get a phone call from someone claiming to be IT support. Or HR. Or even your CEO. It sounds urgent. They need your info now. You panic. You give it.

3. Credential Stuffing

If you’ve reused the same password across multiple sites, this one’s for you. Hackers take stolen credentials from past breaches (like LinkedIn or Dropbox) and try them on hundreds of other platforms. It works a lot.

4. Fake Public Wi-Fi

Ever connected to “Free Airport Wi-Fi”? Sometimes it’s legit. Sometimes it’s a hacker’s laptop. Once you’re on their network, they can intercept your data.

5. Malicious Apps and Extensions

Shady mobile apps or browser extensions ask for more access than they need, then abuse it. They log your keystrokes, scan your clipboard, or upload your data without you knowing.

Why You’re More Vulnerable Than You Think

You don’t need to be rich or famous to get hacked. You need to:

• Reuse passwords
• Skip MFA
• Click one bad link
• Be on the wrong Wi-Fi network

Cybercriminals aren’t targeting you—they’re targeting everyone.

They run bots that test leaked credentials from old data breaches across platforms like Gmail, Amazon, and PayPal. If you’ve ever used the same password twice, you could be exposed.

Go check haveibeenpwned.com to see if your email has been in a breach. Spoiler: it probably has.

5 Things You Can Do to Protect Yourself

1. Use a Password Manager

I use 1Password—and I’m not compensated or affiliated with them in any way.

One of the reasons I use it is that it includes built-in support for MFA codes, so I don’t need a separate app like Google Authenticator. It syncs across all my devices and makes secure logins effortless.

Other great options include:

• Bitwarden
• Dashlane
• iCloud Keychain (for Apple-only users)

2. Turn on MFA

Multi-factor authentication (MFA) adds a second layer of protection to your accounts. If someone steals your password, they still need your phone or device to get in.

Turn this on for:

• Email
• Banking
• Social media
• Cloud storage

3. Stop Reusing Passwords

Every account should have a unique, strong password. If one gets compromised, the rest stay safe. Password managers make this easy.

4. Be Skeptical of Public Wi-Fi

Don’t trust open networks. They’re convenient, but risky. Avoid logging into sensitive accounts when you’re on them. I have a separate article that talks about issues around wi-fi, and how you can resolve them.

5. Use a VPN—Like Cloudflare WARP

A VPN encrypts your internet traffic, making it unreadable to snoops on public networks.

I personally use Cloudflare WARP. It’s free, fast, and easy to install—not just on your phone, but on your Mac or Windows PC too.

WARP isn’t a full VPN—it doesn’t spoof your location—but it does encrypt your data using Cloudflare’s secure network.
Perfect for cafés, airports, or anywhere with sketchy Wi-Fi.

I’ll be doing a full episode soon on VPNs: how they work, which ones to avoid, and how to choose the right one.

Lessons from the IT Trenches

As someone who’s spent their career in IT leadership, I’ve seen this stuff go sideways.

The most significant security breaches I’ve witnessed didn’t happen because of bad software.
They happened because of one person:

• Clicking a fake DocuSign email
• Reusing an old password
• Ignoring a security warning

Technology doesn’t get hacked—people do.

So my job as an IT leader is to make secure behaviour easier than insecure behaviour. That’s why I advocate for tools like 1Password and Cloudflare WARP. They reduce friction, and they encourage good habits.

One good decision can stop an entire breach chain.

Final Thoughts

Let’s recap:

• You are a target, even if you don’t feel like one.
• Cybercriminals count on laziness, reuse, and distraction.
• Password managers, MFA, and VPNs aren’t optional anymore—they’re essential.

You don’t have to fix everything overnight.
Just start with one thing.

Add MFA to your email.
Start using a password manager.
Try a VPN next time you’re on public Wi-Fi.

Security doesn’t have to be scary. It just has to be intentional.

Like this post? Share it with a friend.

Listen to the full episode of Tech It From Me wherever you get your podcasts from.

Or check out all episodes at techitfromme.com.

(Feature image generated with the help of DALL-E)

Tech It From Me

Why Wi-Fi Still Sucks in 2025

Play Episode Pause Episode

Mute/Unmute Episode Rewind 10 Seconds 1x Fast Forward 30 seconds

00:00 / 21:05

We’ve come a long way since the early days of wireless networking. Today, we’ve got gigabit internet, mesh routers, Wi-Fi 6 (and now Wi-Fi 7), and entire smart homes running on wireless. So why does your video call still freeze when you move rooms? Why does your smart TV buffer even though your phone shows full bars?

It’s 2025—and Wi-Fi still sucks for most people.

In this article (based on Episode 8 of my podcast Tech It From Me), I’ll explain why Wi-Fi problems are still so common, even when you’ve “done everything right.” I’ll also walk you through what’s worked for me at home, and what you can do today to make your Wi-Fi faster, stronger, and more reliable.

Wi-Fi ≠ Internet

Let’s start with the biggest misconception:
Wi-Fi is not your internet connection.

Wi-Fi is just the wireless bridge between your device and your local network. Your internet comes into your home through your modem, which connects to your ISP (Internet Service Provider).

Here’s an analogy:
Think of Wi-Fi like the hallway in your house, and the internet like the road outside. If you can’t get through the hallway because it’s blocked with furniture, that’s not the road’s fault.

If Netflix is buffering or your smart devices are sluggish, don’t assume it’s your internet provider. The issue might be inside your house.

My Setup: What’s Actually Working

I’m on a 1 Gbps down / 30 Mbps up cable connection—pretty typical here in Canada. Like most people, I started out using the modem/router combo from my ISP. But it didn’t take long to realize it wasn’t cutting it.

I shut off the Wi-Fi on the ISP modem and replaced it with a TP-Link Deco X60 mesh system. I’m not sponsored by TP-Link—I just did the research and landed on something that worked for my space.

Here’s what makes the difference:

• Three Deco units, one per floor
• All wired with Ethernet backhaul (not relying on wireless mesh links)
• IP passthrough enabled on the modem, so my Deco handles routing directly—no double NAT or firewall conflicts

The result?
A strong, stable, and consistent network throughout the house—even with 70+ devices online.

Why Wi-Fi Still Sucks (Even in 2025)

Even with good hardware, Wi-Fi can still let you down. Here are the biggest culprits:

1. Signal Interference

The 2.4 GHz band is a mess. It’s crowded with microwaves, baby monitors, Bluetooth, and your neighbours’ Wi-Fi. Even fridges, TVs, and metal furniture can distort or block signals.

I’ve had smart plugs drop offline just for being near a metal shelf. It’s invisible noise—but it matters.

2. Poor Placement

Too many routers are hidden in closets, behind TVs, or stuffed in basement corners. Wi-Fi needs space to breathe.

For the best signal:

• Place routers high up
• Keep them in open space
• Center them in your home if possible

3. Device Overload

Smart homes come with a price: constant chatter. Phones, tablets, smart TVs, security cameras, thermostats, smart lights, vacuum robots—and in my case, even 3D printers and Christmas light controllers.

Each one is always doing something—pinging the cloud, updating firmware, or syncing. That’s a lot of traffic for your router to manage.

4. Unrealistic Expectations

Mesh systems are great—but they’re not magic. If your secondary node is hanging on by a thread to your main unit’s signal, you’ve just extended weak coverage further.

That’s why wired backhaul matters. My mesh units are all connected via Ethernet, turning them into true access points.

What You Can Actually Do (Starting Today)

No need to rip out drywall or rewire your entire house. Just follow these real-world tips:

1. Turn Off Your ISP’s Wi-Fi

Your modem’s built-in router is likely weak. Disable its Wi-Fi and let your own mesh or router system take over. Avoid signal overlap and network confusion.

2. Nail Your Placement

Get your access points off the floor, out of closets, and away from metal. Use a Wi-Fi analyzer app (like NetSpot, WiFiman, or WiFi Analyzer) to test signal strength and channel congestion. It’s free and makes a huge difference.

3. Hardwire What You Can

Ethernet still rules.
If a device doesn’t move—wire it. That includes mesh nodes, desktops, TVs, gaming consoles, and even streaming boxes. Each wired device takes pressure off your wireless network.

One warning:
Never run Ethernet through cold-air returns or HVAC ducts.
It’s a fire hazard and likely violates building code. Instead, use:

• Surface raceways
• Joist spaces
• Crawlspaces
• Wall cavities with low-voltage brackets

4. Use Your Own Router

If you’re relying on an all-in-one modem/router, it’s time to upgrade. Even a mid-range system from TP-Link, ASUS, or Netgear will give you better coverage and control.

5. Segment Your Smart Devices

Most IoT gadgets still use 2.4 GHz and don’t need full access.
If your router supports it, create a separate SSID just for them. For more advanced setups, use a VLAN—which is like creating virtual traffic lanes on your network.

Pro Tip: Apartment & Condo Users

If you live in a dense building, your Wi-Fi will always be fighting with others. So focus on managing interference, not eliminating it.

• Use 5 GHz or 6 GHz bands where possible
• Wire devices when you can
• Create a guest network for noisy smart plugs
• Schedule firmware updates and cloud syncs for off-peak hours

Sometimes, stability beats speed.

The Bottom Line

Wi-Fi doesn’t suck because the technology is bad.
It sucks because most people don’t realize it’s an ecosystem—made up of placement, hardware, signal quality, congestion, and interference.

The good news?
You don’t need to be a network engineer to fix it.
You just need to think like one.

Be intentional.
Plan your setup.
Use the right tools.
And don’t expect miracles from budget gear or lazy installs.