I love data science. But, my story is a bit different. 

I began as an Excel-Guru / Mechanical Engineer. Not a traditional background for a data scientist by any means. 

I used Excel until I reached its limits. 

I found R, and the rest is history:

- I began learning R, which led to applying data science to business 

- By applying data science, my team went from $3M/year to $15,000,000 per year

- I began consulting, which led to trainings all over the world

- I found my passion for teaching, and made a company out of it (Business Science!)

Since finding data science, I haven't worked a day in my life. It's been amazing. 

Let me know if you'd like to learn R, Python, or just want to chat data science.

Matt Dancho

Vincent Granville

Chief AI Scientist, GenAItechLab.com

Full-stack entrepreneur with multimillion-dollar exit. Co-Founder at BondingAI.io. Author (Wiley, Elsevier). Creator of open-source Python libraries / Web APIs such as xLLM and GenAI-Evaluation. Builder of high performance in-memory agentic multi-LLMs for professional users and enterprise, with real-time finetuning, self-tuning, no weight, no training, no latency, no hallucinations, no GPU. Made from scratch, leading to replicable results, leveraging explainable AI, adopted by Fortune 100. With a focus on delivering concise, exhaustive, relevant, and in-depth search results, references, and links. 

Vincent is a leading expert, mathematician, patent owner, former post-doc at Cambridge University, former VC-funded executive, with 20+ years of corporate experience including CNET, NBC, Visa, Wells Fargo, Microsoft, eBay.  Vincent also founded and co-founded a few startups, including one with a successful exit (Data Science Central acquired by Tech Target). He is also a top influencer for brands such as Nvidia, SingleStore, and The Linux Foundation.

From 2005 to this day, Vincent successfully arbitraged the real estate market, buying, selling and renting out homes with significant growth and return over time: Building wealth starting from negative net worth when he first arrived in the US in 1995.

Chief AI Scientist

Privately funded GenAI and ML Optimization research lab, creating and disseminating open source intellectual property. Developing explainable and generative AI, synthetic data, computer vision, NLP and generic ML technologies. Check out our certifications, Python libraries, eBooks, articles, courses and podcasts, including our best seller "Synthetic Data and Generative  AI" (Elsevier). Visit our company website at MLtechniques.com.

I’m happy to share that I’m starting a new position as CAIO at BondingAI!

Chief AI Officer. Building architecture, IP, and new AI agents for xLLM, including: • Predictive analytics on retrieved tables from PDFs, databases, the Internet or Excel repositories • Anomaly detection for fraud detection / cybersecurity • Medical data: automatically recovering insights (electrocardiogram data, predicting heart attacks) and data compression • Tabular data synthesis: best and fastest on the market, no deep neural network involved • No-Blackbox deep neural networks and other alternatives (96% correct prediction for next token)

See also new book on these topics, entitled "No-Blackbox, Secure, Efficient AI and LLM Solutions", at https://lnkd.in/g5cmxHE3

See also my alternative to DNNs, where there is no gradient descent, and the minimum of the loss function is obtained in closed form with no iterative algorithm. There is a whole chapter on the topic in my new book, see table of contents at https://lnkd.in/g5cmxHE3

Quantitative Researcher | Author of O’Reilly’s Deep Learning for Finance

Most people default to MSE when training regression models.

That’s fine, until it isn’t.

Mean Squared Error (MSE) aggressively penalizes large errors.
- Great when your data is clean.
- Terrible when it isn’t.

In real financial and economic data, outliers are not bugs, they are features.

That’s where Huber loss quietly does a better job.

What’s the difference?
• MSE squares every error
 → Large mistakes dominate the optimization
 → Models become overly sensitive to rare shocks

• Huber loss behaves differently
 → Quadratic for small errors (like MSE)
 → Linear for large errors (like MAE)

In plain terms:
Huber protects your model from being hijacked by a few extreme observations.

Source: LinkedIn

"If solved, it would destroy modern cryptography."
No, it would not. Modern cryptography does not rely on prime numbers. You get a lot better with irrational numbers. Details in my book 0 and 1 – From Elemental Math to Quantum AI at https://lnkd.in/gQD86nAP

An MIT lecturer just used ChatGPT to "understand" a problem mathematicians haven’t solved in 160 years.

And he failed big time.

Let me explain what actually happened.

He published a piece on Forbes showing his ChatGPT conversations on the Riemann Hypothesis.

It's one of the deepest unsolved problems in mathematics. 
If solved, it could destroy modern cryptography.

And there's a 1 million $ prize for anyone who is capable of doing so. 

The lecturer claimed he didn’t understand it, as he is not a mathematician. But he believed ChatGPT could help.

He tried to prove this by asking very naive and wrong questions to ChatGPT, just like a layman would do.

And in the end, he did not understand the problem.
He just understood ChatGPT’s simplifications of his own confusion.

This is the new failure mode we’re not talking about enough.

1) Generative AI doesn’t increase understanding, it just increases confidence.

ChatGPT is very good at: 
– sounding coherent
– regurgitating internet materials
– producing explanations that feel complete, but are usually wrong

What it CANNOT do is guarantee correctness, depth, novelty, or epistemic understanding.

But the user doesn’t see that. They see fluency.

That’s textbook Dunning–Kruger, now automated.

2) Explaining is different from knowing.

The article proudly shows how GPT can: 
– simplify definitions
– paraphrase known explanations
– restate standard intuitions and basic concepts

There is no mathematical insight or reasoning. Just random words.

The Riemann Hypothesis isn’t unsolved because people don’t understand the math. 

It’s unsolved because existing demonstration attempts failed.

No amount of conversational fluency changes that.

3) AI doesn’t know when it’s wrong, and neither do non-expert users.

ChatGPT doesn’t know:
– which simplifications are harmless
– which nuances are essential
– which mistakes invalidate the reasoning

And users who lack domain expertise cannot tell the difference.

So errors don’t look like errors. They look like insights.
That’s the danger.

The article concludes that ChatGPT helps us understand why the Hypothesis is unsolved.

But the interaction itself shows the opposite: 
– shallow questions
– circular explanations
– analogies that replace rigor
– philosophical confusion dressed as insights

This is just an illusion of understanding.

It's not true that AI lowered the barrier to knowledge. What it did was lowering the barrier to "feeling informed".

And that’s far more dangerous.

Because when everyone believes they understand everything, expertise collapses, standards erode, and confidence detaches from competence even more.

And that’s exactly why automation bias + Dunning-Kruger effect is such a dangerous mix.

Fluency is not understanding. 
And AI is very good at masking the difference.

----
Follow me Chiara Gallese, Ph.D. for more on AI & Tech risks

taplio

The Battle

Matt Dancho

Vincent Granville

Followers

Posts

Engagements

Likes

Performance ScoreVincent Granville's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Matt Dancho

Vincent Granville

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Vincent Granville's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreVincent Granville's LinkedIn Performance