November 30, 2024

Engineering complexity: because it works !

Why do we build complex products, like smart phones, AI, or Hermix?

Because complexity works. It delivers benefits and value, to the economy and society. And this is particularly true of modern engineering. We build complex products, for complex organizations, even if sometimes more fragile, or expensive, or risky. Because they work. These are thoughts about product and engineering complexity, from our Hermix webinar of 27. Nov 2024, with Arian Turhani, VP Capgemini for European institutions, and Alin Izvoran, public procurement expert: AI for public procurement and sales. Complexity has 2 major aspects: structural, and dynamic. Structural complexity means composed of many varied interrelated parts. A modern car has thousands of interdependent components. A modern AI neural network has hundreds of billions, even a trillion nodes. Dynamic complexity relates to uncertainty and ambiguity. It is characterized by emergence, nonlinearity, adaptiveness, propagation, chaos. This is where we get Taleb’s black swans and Lorenz’ butterfly effects. These systems are difficult to predict and control, even if we have sufficient information about their components. It was Aristotle who first described the phenomenon of really complex environments, 2300 years ago, as he famously said that the whole is not only more, but in fact it is different than the sum of the parts. And some of the effects of complexity are positive. The whole is not only different, but also better. These complexity manifestations also occur in relation to modern engineering technology. This is also what we are also building with Hermix. We build more than a tool with significant economic and efficiency benefits. We build a complex ecosystem. We create a new paradigm, with policy implications; such as increase in transparency and competition. We help create a better market environment.

Full video on Hermix website.



November 10, 2024

AI/GPT is a genius idiot slave

 I use AI/GPT as my genius idiot slave. 

Very pedantic. Obedient. Hard working. Great memory. 

Unable to understand the context.


It works. 

You must know HOW to ask,

how to evaluate the answer, 

AND you need patience.


So, what works great for me? 

Search and summarize a very general question, a book, a movie, a problem.

Correct my spelling, find synonyms, antonyms, rephrase text.

Write standard software code.


What doesn't work: solve a niche software problem, invent a new solution to a new problem, find a niche book, especially not published in English.

 

My basic example of tonight: I had a very niche problem: 'check if the value of an Excel cell is N/A or ""'

GPT4 was not able to solve it. Excel has a weird way to evaluate expressions and errors, so the simple solution always returns N/A, instead of True or False.



October 30, 2024

"AI Prompt Engineering" is an elementary skill.

 "AI Prompt Engineering" is not a new job. 

It is an elementary skill. 

Basic "digital literacy".


For at least 20 years, "Office, email and internet" have been mandatory in all CVs.

You must now add "Prompt engineering". 


Prompt engineering is the ability to ask questions and obtain relevant, useful answers or text from GPT (and other LLM/AI tools).


I could never list " MS Office" in my CV skills. Not only that I memorized most Office shortcuts, but I've been writing VBScript macros and Java/C++ OLE automation code ever since grad school (thanks Ionut, I learned so much from you). Listing Office or Internet as a skill would be silly: I can actually write their code.


Similarly, I cannot list "prompt engineering" as a skill. I strive to become an AI expert, of course. ChatGpt says that an AI Specialist is a master in machine learning, data science, and artificial intelligence; managing NLP; Python, R, or Java; ML libraries such as TensorFlow, PyTorch, Scikit-Learn.


But yes:

Prompt engineering is now the basic digital literacy skill, to add to all CVs.

 

L.e. Great comment 

from Tavi Paunescu on FaceBook: At least Office was a clearly defined suite, current AI big names are black boxes. So "engineering" against a black box :)

My reply: Any sufficiently advanced technology is not distinguishable from magic. See also Strugatki's picnick. ChatGpt says it's Arthur C Clarke.


L.e.2 Great comment 

from Smaranda Spizzico: how can I learn?

My reply: great question. Thanks!
ChatGpt says:

Prompt: How can you learn prompt engineering?

Answer: [...] Start with General Prompts: Experiment with simple, straightforward prompts. Notice how the wording, structure, and even tone can affect the output. For example, try using different formats like questions, commands, or step-by-step instructions.
Understand Prompt Types: There are descriptive prompts, directive prompts, few-shot prompts (where you give the model examples), and chain-of-thought prompts (where you guide the model step-by-step). Experimenting with each type will help you understand which works best for various tasks.[...]


October 20, 2024

A good salesman is someone who sells

I like simple definitions. 

There are tons of articles and books on this topic. They talk about what makes a good salesperson. Methods. Processes. Spin selling, solution selling. Pipelines, funnels, inbound, outbound, outdoors, network. Brand. Reputation, awareness, recognition. Tools. CRMs, content marketing, analytics, automation, lead generation. Skills. Empathy, curiosity, NLP, guts, courage, feeling, structure, negotiation.


I even wrote a couple of papers myself. About online digital communication and collaboration, and about my approach to B2G/public sector and B2E / enterprise sales. We develop ourselves tools for lead generation, market intelligence and bid automation.


But these are methods.

I like simple definitions.


A good salesperson is someone who sells.

October 12, 2024

Regulations such as the GDPR and the AI Act hinder innovation, entrepreneurship, and private initiative in Europe


Regulations such as the GDPR and the AI Act hinder innovation, entrepreneurship, and private initiative in Europe.


The GDPR was drafted by bureaucrats and politicians who never managed a business, worked in the real economy, or engaged in sales.

The intentions were certainly good. Large corporations do need regulation. There are still significant areas requiring oversight, such as their monopolistic behavior toward suppliers, partners, and customers.

However, the GDPR has hurt small companies far more than large corporations. GDPR compliance is a fixed overhead. The big players immediately hired legal teams - they could afford it. Now, when you sign up for any platform, including Google, Microsoft, Amazon, or Facebook, you are inundated with hundreds of pages of consent forms and legal jargon, which small businesses or private individuals inevitably accept without reading.

A few years ago, I actually tried reading Google's documents related to data sharing and privacy. I managed to get through maybe two pages. There were countless annexes, totaling perhaps 500 or even 1,000 pages. For SMEs and startups, drafting or even reading such provisions is an impossible task.


A recent paper (summary below) confirms that GDPR hurt the economy, especially SMEs. Numerous analyses, including the famous Mario Draghi report from September 2024, as well as several official reports from the European Commission and DG Grow, also confirm that Europe is falling behind in innovation, competitiveness, and even GDP.


Yet, Europe continues to pile on more regulations.

The AI Act placed a new and immense burden on companies conducting AI and data research. The costs of understanding or complying are substantial. The entire business and research community has expressed concern (see examples below). Some large corporations moved their R&D activities from Europe to the US, while others absorbed the overhead. Who suffers? Startups and SMEs.


The EU continues to talk about innovation and growth. But in practice, the immediate reaction to new technology and innovation is to regulate and control. Then we wonder why we face poor economic performance, low innovation, low entrepreneurship, and a lack of unicorns.


Later edit

Great discussion on the related LinkedIn topic:

I strongly support the importance of privacy for consumers and private individuals.

My questions are:

1) is there any proof that the current GDPR regulation is effective? 

Imho, in practice, all companies deployed a cookie banner, and updated the small-text EULA that nobody reads anyway.

2) if the advantages of Gdpr are bigger than the disaster created to small businesses. 

I haven't seen any study on the societal impact.

The impact on the economy is proven: it's a disaster.

Notes

Summary and extracts from the analysis of the impact of the GDPR regulation

"Mounting evidence —including a critical paragraph in the recent Draghi (2024a) report— suggests it is doing more harm than good to Europe's tech ecosystem.

The "Brussels Effect," coined by Anu Bradford in a 2012 paper (and later a 2020 book of the same title) says that when the EU makes rules, because its market is so large, companies around the world will decide it's easier to just follow those rules everywhere rather than trying to have different rules for different places

Obviously, this argument is very appealing to EU bureaucrats and EU parliamentarians 

* Web traffic and online tracking fell by 10-15% after GDPR began

* The market has become more concentrated.

* Innovation has slowed. 

Web traffic and online tracking fell by 10-15% after GDPR began. Users often opt out when asked for consent. EU firms store 26% less data on average than US firms two years after the GDPR and reduce computation relative to US firms by 15%. 

The market has become more concentrated. Large firms with their own data gained market share. Small firms struggle more to comply and reach customers than large ones.

Innovation has slowed. New app entries in the Google Play Store halved after GDPR. Venture capital deals in the EU fell by 26.1% compared to the US. In particular AI innovation in Europe has been hindered- GDPR increase the cost of storing and processing the data required to train AI models.

Big Firms Win, Small Firms Lose

It appears GDPR has put smaller firms at a relative disadvantage and that it has increased market concentration, particularly benefiting very large firms, notably Google.  The reason is that GDPR increases the fixed costs of data manipulationGoogle can afford that.

...small Spanish B2B company which used to have a contact database of 300,000 firms. The day GDPR started, that dropped to 15,000 because they needed explicit consent

Less Innovation, Fewer Entrants"

The Sept. 2024 Meta letter, signed by 40 executives from Ericsson, Capgemini, Deutsche Bahn, Deutsche Bank, Nokia, Renault etc.

"the reality is Europe has become less competitive and less innovative compared to other regions and it now risks falling further behind in the AI era due to inconsistent regulatory decision making."

Deloitte analysis on the impact of the AI Act

"More than half of the [500] companies surveyed believe that their innovation opportunities in the field of AI will be restricted by regulation; less than a fifth think that the AI Act will have a positive impact on innovation opportunities."

Engineering complexity: because it works !

Why do we build complex products, like smart phones, AI, or Hermix ? Because complexity works. It delivers benefits and value, to the econo...