Breadcrumbs of Innovation: A Snapshot of AI Explorations
In the ever-evolving landscape of artificial intelligence, finding coherence can feel like solving a complex puzzle with constantly shifting pieces.
Much like the traditional new year’s resolutions we eagerly draft and often abandon, this compilation is a snapshot of my current AI explorations. I’ve scattered code files and prompts like breadcrumbs, marking the path of my intellectual wanderings.
Beyond the technical intricacies, a deeper, more philosophical question persistently echoes in my mind: What profound implications will this AI revolution have on our societal fabric? This is not merely an academic pondering, but a visceral inquiry that resonates more deeply than the technological marvel itself.
The Fundamental Dilemma: Regardless of whether we stand “for” or “against” AI, how will this transformative phenomenon reshape the political and social landscapes we inhabit?
It’s a question that transcends binary perspectives, inviting us to contemplate the nuanced intersections of technology, human agency, and collective future.
For this post also, you can find all files and prompts, on my GitHub account. See https://github.com/bflaven/ia_usages/tree/main/ia_video_editing_faiss_compare_keywords
The illustration are made from icons created by juicy_fish – Flaticon
The audio extracted from this article made with NotebookLLM
The Big Dive part 1 : exploring reflections on the impact of artificial intelligence (AI)
The Big Dive part 2 : The 2025 US presidential election through the lens of “We Have Never Been Woke,” a book critiquing a new “woke” elite.
A small compass – 10 Top ideas to survive on IA world for the coming year
I have attempted a one day training session in French. I do translate them but I also release the original version as I want to have it somewhere!
English version
- #1 Get over the shock and demystify AI.
- #2 Map AI’s uses and train teams.
- #3 Identify opportunities for productivity gains, without taboos.
- #4 Prepare for the emergence of AI assistants.
- #5 Anticipate the decline of the link economy.
- #6 Avoid the pitfalls of technological dependence.
- #7 Arm yourself against the peril of generalized information disorder.
For the #1:
The very first impression indeed is to “Get over the shock”, so taking AI as a human phenomenon and therefore historical. It helps demystifying in a sense. AI is “thunderclap in a clear sky” that’s the normal evolution econonomy that can for instance summarize with the equation IA = Cloud + Data.
For the #2:
That is the call to action. Identifying the potential uses of AI and take action without taboo. That is the first step I made by reading for instance “New powers, new responsibilities. A global survey of journalism and artificial intelligence”, from Charlie Beckett which led to this post. Check out: Some ideas on the probable future of journalism facing IA and how to create a prompt facilitation application for ChatGPT with Streamlit
For the #3:
Indeed, the very first impact of IA is that “ratchet up your output”. AI is a great lever to increase my performance but not necessarily the quality that may be the first victim of this productivity inflation. The proof is given by the quantity of code, prompts and attempts carried out, particularly in technical subjects.
On the quality issue, Olga that is working with me reminds me this meme. When ChatGPT came out, the real dev face this issue for instance.
For the #4:
Agents are already here. From a psychological point of view, you must accept the AI’s otherness realities (agents, robots, objects) that will multiply. To speak like Nietzsche “Don’t be afraid of the Doppelgängers created with IA. Keep your wit when summoning the IA’s God because fear is submission” 🙂
For the #4:
Interesting assertion. The link economy is in decline. That’s hard to conceive but Search Engines and Social Networks will be progressively replaced with IA. People will ask, talk, share emotions and opinions with Robots instead of other people.
For the #6:
Avoiding the pitfalls of technological/informational dependence. Who shapes the story and tempo is the time master. For me, it raises the issue of Sovereignty over models.
For the #7:
What to do against infobesity and like Steve Bannon wanted when the information sphere is saturated with sh… Everyone is thinking of their own little anxiety-provoking concept: cognitive apocalypse, end of the individual… A radical solution just turns off or wait for the black-out 🙂
French version
#1 Sortir de la sidération et démystifier l'IA.
#2 Baliser les usages de l'IA et former les équipes.
#3 Identifier les opportunités de gains de productivité, sans tabou.
#4 Se préparer à l'émergence des assistants IA.
#5 Anticiper le déclin de l'économie du lien.
#6 Eviter les pièges de la dépendance technologique.
#7 S'armer face au péril du désordre informationnel généralisé.
Thanks to Rémi Rostan for these points.
https://linktr.ee/studiolhc
Here are some ideas that will mark this coming year.
My job definition AI’s User First
When it comes to explaining what I do or do not do professionally with AI, the only qualifier I can answer is that I am and remain an AI user. Like any user, I am quite obtuse and impatient, I want to get a result. So, rightly or wrongly, technical explanations quickly bore me.
Some concepts to qualify a project
Approaching IA is not so delicate, like always you have to figure out in what scope you are entering and launching your inquiry. Know-how is one thing, but make known is also important. I forged this franglish concept : the “faire-savoir” beats the “savoir-faire”
Ha, the passion for naming things and giving definitions to play the specialist. Anyway, I discovered, much more than the sole MVP, that it was existing much more concepts to qualify projects.
- MVP Minimum Viable Product: A basic version of a product with just enough features to be usable by early customers and provide feedback for future development.
- MVE Minimum Viable Experience: The simplest version of a product or service that still delivers a satisfactory user experience and meets core customer needs.
- MAP Minimum Awesome Product: A product that goes beyond basic functionality, delivering not just utility but also an engaging, delightful user experience from the start.
- RAT Risky Assumption Testing: A method of validating critical hypotheses about a product or business model through quick, low-cost experiments to reduce potential risks before full development.
As I have no paid subscription to medium.com, I did have the chance to read the full post by Jano le Roux “MVPs and MVEs are Dead*. Build a MAC”. Anyway, the title has played as a true clickbait as it is provocative. It emphasizes a golden rule of the storytelling: conflictual and tension are the ingredients for your secret sauce that will propel your reader through to the end.
So, what did I learn from the beginning of the post “MVPs and MVEs are Dead. Build a MAC”.
1. MVPs are obsolete
MVPs were supposed to be the “test the waters” version of your product. Build something basic, release it quickly, and see if there’s interest. Sounds smart, right? Not anymore.Here’s why:
AI has made MVPs too easy:
Anyone can throw together an MVP in hours using tools like GPT-4, Figma, or Bubble. The bar is so low that having an MVP isn’t impressive anymore.
I agree with that. According to me, a MVP is between 5 to 50 prompts, that is the reason why from now on I release all the prompts I made. Unfortunately, because of my greed, I am still wondering what a MAC is in the post of Jano le Roux?
*No paying the subcription to medium.com and for information in general reminds me this quote from William Shakespeare “Some there be that shadows kiss. Such have but a shadow’s bliss.”
Here are some extra sources on these concepts still unknown to me. I did not find MAC but MAP 🙁
- “MVPs and MVEs are Dead. Build a MAC” by Jano le Roux
- “¿MVP, MVE, MAP cual tengo que usar?” by Henryk Sobczak
- Your Guide to MVP, MMP, MLP, and MAP Startup Stages | Django Stars.
IA and CMS: handling tags
The underlying user story for this code is straightforward: to establish a semantic match between all existing keywords in a CMS (such as WordPress, Drupal, or a custom-built CMS) and a list of AI-generated keywords using prompts (e.g., with Claude, Mistral, or ChatGPT). The goal is to generate a refined list of existing keywords based on the AI’s suggestions for the editorial team.
The code utilizes the SentenceTransformer(‘all-MiniLM-L6-v2’) model to achieve this semantic matching effectively.
More code and prompts on https://github.com/bflaven/ia_usages/tree/main/ia_video_editing_faiss_compare_keywords/ia_cms
# in Portuguese ia_generated_kw = ['Milagre económico', 'Ásia-Pacífico', 'Ásia-Pacífico', 'persona non grata".', 'Ébola', 'Áudio', 'óleo de palma', 'Áustria'] cms_existing_kw = ['Milagre económico', 'agentes do estrangeiro', '1° de Maio', 'persona non grata".', 'phygital', '#Metoopolitico', 'óleo de palma', 'ABBA'] # SCENARIO_1 in French # EX_1 liste_1 = ["Artificial Intelligence", "Machine Learning", "Natural Language Processing"] liste_2 = ["AI", "Deep Learning", "NLP", "Robotics"] # EX_2 (IAG) liste_1 = ["Ukraine", "Zelensky", "Belgique", "Dirigeants"] (CMS) liste_2 = ["Zelensky", "Europe", "Bruxelles", "Russie", "Union Européenne"] # SCENARIO_2 in French liste_1 = ["AI", "Deep Learning", "NLP", "Robotics", "Artificial Intelligence"]
IA and Faiss:
Simplistic Explanation of Faiss:
Faiss is a tool that helps find similarities between things like text or images, even when the dataset is huge. It works very fast, even for large files that don’t fit in your computer’s memory. It uses smart algorithms to compare and group items efficiently and can work with Python or on the GPU for speed.
More on “Faiss”: https://github.com/facebookresearch/faiss
More code and prompts on https://github.com/bflaven/ia_usages/tree/main/ia_video_editing_faiss_compare_keywords/ia_faiss
Use Case Connections:
- Use Case #1: Text Ordering
- Use Case #2: Image Ordering
- Use Case #3: Unrelated Use Case: Image Descriptions
Faiss is used with the “all-MiniLM-L6-v2” model to quickly rank and organize text titles. It helps pick the best title from a list by understanding how similar each one is to the ideal choice.
Faiss works with CLIP (a tool that connects text and images) to sort images based on their relevance to a user’s multilingual text input. For example, if a user searches “beautiful sunset,” it finds the most suitable image from a collection, even if the text is in different languages.
A separate method using face_recognition and transformers creates accurate descriptions for images. These descriptions can improve image search or provide useful alternative text (alt messages) for accessibility. For instance, it might describe an image as “A smiling person at the beach during sunset,” which is helpful for users and search engines. More on “CLIP, Connecting text and images” at https://openai.com/index/clip/ and “open_clip” at https://github.com/mlfoundations/open_clip.
IA and Video Editing:
This app helps analyze text from a transcription file (exported as JSON from Whisper). It uses a tool called SpaCy to find important information like “Detected Entities” (e.g., names, places), “Key Phrases” (important topics), and “Text Segments” (specific parts of the text). For each of these, it shows the time when they start (cue point IN) and end (cue point OUT) in the audio or video.
More code and prompts on https://github.com/bflaven/ia_usages/tree/main/ia_video_editing_faiss_compare_keywords/ia_video_editing
User Story:
As a content creator or video editor,
I want an app that analyzes transcriptions from my audio or video files,
So that I can easily find key moments (like when important topics or entities are mentioned) and use the timecodes to edit or navigate my content efficiently.
# SOURCE for video editing # Putin issues direct threat to West, says Russia used new missiles in Ukraine • FRANCE 24 English # https://www.youtube.com/watch?v=wBgCsgBLWpg # Germany's Scholz set to run for second term after potential rival bows out • FRANCE 24 English # https://www.youtube.com/watch?v=Ns4LuKxAuHE
Set a frame before exploring use cases
Just as reminder the very first two axes that I wanted to investigate that are part of the very beginning of this post:
Video Editing: AI-driven video editors like Magisto and Adobe’s Sensei AI help automate tedious editing tasks such as cutting footage, adding transitions, adjusting lighting, and even generating captions and subtitles.
Content Generation for Blogs and Articles: AI-driven platforms like GPT-3 and Jasper can generate long-form blog posts, articles, and SEO-optimized content tailored to a specific topic or keyword. By producing high-quality written content at scale, businesses can keep their blogs and websites updated with fresh material, improving SEO and increasing organic traffic.
Paranoia Chapter or “Everyone is out to get me”
A small note on this last chapter: After all, writing remains an outlet, it is both a serious and relaxing exercise. Even if the following chapter is incomprehensible, which may be detrimental to me, it does not entirely reflect my opinion either. I am clumsily trying to reflect at the beginning of this year and especially to evacuate cliches, redundancies and other hackneyed phrases that harm my thinking. First, I try to deflate as much as poverty and emptiness in my wording and nourishes my philosophical self-education!
Let’s say that the American presidential elections is always a unique opportunity to measure the state of the world. Personally, it was after Trump’s first election that I stopped writing in French, considering that the world was global with no way back. Check out : D3 – Représentation avec D3 d’un réseau avec le gouvernement Trump comme exemple
8 years later, the changes are even more advanced and the disintegration of both Democrats and Republicans is even more advanced. It’s not easy to approach such a subject without coming across as either a conspiracy theorist, a hater or a a supporter from one side or another.
What is ultimately the political offer of the two camps present? And how IA is a game changer?
Political offer from the Democrats
Let’s talks about the Democrats. Here are some Key Insights into the Transformation of Democratic Political Discourse. This analysis draws heavily from Musa al-Gharbi’s groundbreaking book “We Have Never Been Woke: The Cultural Contradictions of a New Elite” by Musa al-Gharbi.
- Tech Disillusionment: The once-loyal tech community is rapidly distancing itself from the Democratic Party, perceiving the party as outdated and disconnected from contemporary dynamics. What was once a symbiotic relationship has transformed into a critical disconnect.
- Symbolic Capitalism: A new breed of “diversity entrepreneurs” leverage social justice language not to genuinely empower marginalized groups, but to establish personal privilege, power, and social status. This performative activism masks underlying power structures.
- Victimization as Currency: Within cultural elites, differentiation and social positioning have become increasingly dependent on constructing intricate narratives of victimhood. The more complex one’s intersectional identity, the more cultural capital is accumulated.
- Narrative Fragmentation: The Democratic political strategy has devolved into a collection of niche, atomized narratives that fail to create a cohesive, unifying message. This approach contrasts sharply with more monolithic political movements.
- Ideological Disconnect: Despite claiming to serve the common good, the Democratic elite has become increasingly detached from broader societal experiences, creating a widening gap between rhetoric and lived reality.
Here are some ressources to go further.
- “We Have Never Been Woke: The Cultural Contradictions of a New Elite” of Musa al-Gharbi https://press.princeton.edu/books/hardcover/9780691232607/we-have-never-been-woke
- Musa al-Gharbi official website https://musaalgharbi.com/we-have-never-been-woke-available-now/
- A great podcast on “No Small Endeavor” with Lee C. Camp and Musa al-Gharbi on his book “We Have Never Been Woke” https://www.nosmallendeavor.com/musa-al-gharbi-we-have-never-been-woke
More insights on the book….
The Paradox of Modern Equality: Key Insights from “We Have Never Been Woke”
The Rise of Symbolic Capitalists
- A new elite class of professionals who work primarily with words, ideas, images, and data
- Found in sectors like education, media, law, and finance
- Predominantly composed of well-educated, affluent white liberals who champion progressive causes
The Fourth Great Awokening
- A dramatic shift in attitudes and activities among knowledge workers starting around 2011
- Marked by a surge in discussions about prejudice and discrimination in media
- Characterized by symbolic capitalists perceiving more racism than minorities themselves report experiencing
The Equality Paradox
- Modern society claims to be more egalitarian than ever, with diversity celebrated and prejudice condemned
- Yet social and economic inequality continues to grow exponentially
- These seemingly contradictory trends are actually interconnected
The Power Dynamic
- Symbolic capitalists gain status and influence through their advocacy for equality
- Their egalitarian credentials paradoxically help them maintain and expand their privileged position
- They often inadvertently contribute to the inequalities they publicly oppose
The Language of Justification
- Social justice terminology is increasingly used to legitimize the elite’s position
- Those who fall behind in the knowledge economy are often dismissed for having the “wrong” views on social issues
- This creates a self-reinforcing system where progressive language serves as a tool for maintaining status
The Unintended Consequences
- Well-meaning advocacy can provoke backlash against the very causes being championed
- The elite’s genuine beliefs often blind them to their role in perpetuating social problems
- Real progress on inequality requires challenging these self-serving narratives
This analysis reveals how modern progressive activism, despite its sincere intentions, can sometimes reinforce the very power structures it aims to dismantle.
It is also true that denouncing this in this way does also play into the hands of the opposing party, in this case the Republicans and their leader Trump and his “broligarchy”.
For “broligarchy” concept, see the post “La temible broligarquía https://elpais.com/opinion/2025-01-21/la-temible-broligarquia.html
Political offer from the Republican
A shot on what I have called “The Republican Techno-Political Landscape: AI, Power, and Narrative Control”. Here are some Key Insights into Republican Tech-Political Ideology. I try to alleviate the critical observation (cognitive bias) for the analysis. I hope that it hemps to reveal a complex intersection of technology, politics, and human agency, where AI becomes both a tool of empowerment and a mechanism of potential societal control.
- Technological Determinism: AI is viewed as a transformative force that fundamentally reshapes human potential, oscillating between utopian enhancement and existential threat. The narrative suggests technology can either elevate or diminish human capabilities.
- Performance Culture: The Silicon Valley ideology permeates Republican tech discourse, emphasizing efficiency, economic rationality, and relentless self-optimization. AI becomes a tool for continuous performance measurement and personal entrepreneurship.
- Narrative Manipulation: AI is recognized as a powerful propaganda mechanism, capable of shaping public opinion, controlling information flow, and potentially engineering consent for political actions like annexation or conflict.
- Transactional Politics: Political leadership, exemplified by figures like Trump, is characterized by a pragmatic, deal-oriented approach. Ideology becomes secondary to strategic gain, with power dynamics measured through immediate transactional benefits.
- Cognitive Subservience: The emerging technological landscape threatens individual autonomy, potentially creating a docile population easily manipulated through emotional triggers and constant technological mediation.
Videos to tackle this post
Using Semantic Similarity with a CMS #ia #nlp #cms #api #python
Combining Face Recognition and Image Captioning: A Comprehensive Guide #python #ia
Natural Language & Multilingual Image Discovery: How AI Understands Your Searches #ia #image
Beyond Subtitles: AI-Powered Video Content Discovery #ia #nlp #spacy #whisper #streamlit #python
More infos
- How to Use FAISS to Build Your First Similarity Search | by Asna Shafiq | Loopio Tech | Medium
https://medium.com/loopio-tech/how-to-use-faiss-to-build-your-first-similarity-search-bf0f708aa772 - GitHub – TamerOnLine/sentence-embeddings-similarity-search: Convert sentences to expressions and perform semantic search using Ollama and FAISS
https://github.com/TamerOnLine/sentence-embeddings-similarity-search - Faiss Python API: Introducing Facebook’s AI Similarity Search Tool – AI StartUps Product Information, Reviews, Latest Updates
https://cheatsheet.md/vector-database/faiss-python-api.en - GitHub – guilherme-pombo/SimpleRAG: The simplest RAG implementation using only Transformers and FAISS
https://github.com/guilherme-pombo/SimpleRAG - GitHub – abinthomasonline/clip-faiss: Image Search Application with OpenAI CLIP Model and Faiss Library
https://github.com/abinthomasonline/clip-faiss - laion/CLIP-ViT-B-32-laion2B-s34B-b79K · Hugging Face
https://huggingface.co/laion/CLIP-ViT-B-32-laion2B-s34B-b79K - Stability AI
https://stability.ai/ - LAION-5B: A NEW ERA OF OPEN LARGE-SCALE MULTI-MODAL DATASETS | LAION
https://laion.ai/blog/laion-5b/ - GitHub – openai/CLIP: CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image
https://github.com/openai/CLIP - Whisper-transcription_and_diarization-speaker-identification-/transcribtion_diarization.ipynb at main · lablab-ai/Whisper-transcription_and_diarization-speaker-identification- · GitHub
https://github.com/lablab-ai/Whisper-transcription_and_diarization-speaker-identification-/blob/main/transcribtion_diarization.ipynb - Can Whisper distinguish two speakers? – #9 by techjp – API – OpenAI Developer Forum
https://community.openai.com/t/can-whisper-distinguish-two-speakers/291253/9 - face-recognition · PyPI
https://pypi.org/project/face-recognition/ - opencv-python · PyPI
https://pypi.org/project/opencv-python/ - ai-video-generator · GitHub Topics · GitHub
https://github.com/topics/ai-video-generator - Python Video Processing: 6 Useful Libraries and a Quick Tutorial | Cloudinary
https://cloudinary.com/guides/front-end-development/python-video-processing-6-useful-libraries-and-a-quick-tutorial - Python For Ai Video Editing | Restackio
https://www.restack.io/p/ai-video-synthesis-answer-python-ai-editing-cat-ai - Automate Video Editing with Python | Towards Data Science
https://towardsdatascience.com/automate-video-editing-with-python-4e0c43edef36 - GitHub – Breakthrough/PySceneDetect: :movie_camera: Python and OpenCV-based scene cut/transition detection program & library.
https://github.com/Breakthrough/PySceneDetect - GitHub – octimot/StoryToolkitAI: An editing tool that uses AI to transcribe, understand content and search for anything in your footage, integrated with ChatGPT and other AI models
https://github.com/octimot/StoryToolkitAI - Nexa AI On-Device Model Hub | Explore Gemma, LLaMa, Qwen, Stable Diffusion, Whisper & Leading On-Device AI Models and Run Models Locally On Your Device
https://nexa.ai/models - GitHub – VIGINUM-FR/D3lta
https://github.com/VIGINUM-FR/D3lta