Rapport IA 2024

This report is the result of weeks of research and daily monitoring of generative AI. The report outlines major trends, practical applications, and avenues for leveraging AI in your business. The report is published by Side School, online bootcamps for applying AI in one's career. Learn at the end how these courses can assist you in this transition.

Applications and Use Cases

Top 3 AI Applications of 2024

1. ChatGPT

2. V7 Go

3. Krea

Special Mentions

1. Heygen

2. Runway

3. Character AI

Tools Overview

The 3 Major AI Figures to Remember in 2024

Top 5 New AI Use Cases in 2024

Watch the demonstration video (6 minutes)

1. Automatic transcription and meeting summaries with Limitless. Time saved per week: 1h

2. Email automation with Lindy. Time saved per week: 1h

3. Creation of custom dashboards with Claude Artefacts. Time saved per week: 2h

4. Automatic competitor reports with V7 Go. Time saved per week: 1h

5. Project creation (or GPTs) for building business proposals in ChatGPT. Time saved per week: 2h

Industries Most Impacted by AI in 2024

In August this year, we took the time to analyze the database of listed jobs (and tasks available on O*NET Online, example here) to understand the industries most impacted by the advent of generative AI. Information exchange, analysis, and intellectual creation-related service industries are the most affected. We share our research and conclusions with you:

Watch the analysis video (4 min)

7 “Wow” AI Applications of 2024

Watch the tools demonstration (8 min)

Want to impress your family? The wow effect here is guaranteed. Our best apps found in 2024, from the simplest to the most complex:

1. Suno: create your own song with a prompt.

2. NotebookLM: add documents or sites, and generate a podcast (in English)

3. Heygen: film a two-minute video of yourself speaking and then generate videos with your “virtual clone”, using your voice or another language.

4. Krea (in “realtime” mode): draw on the interface on the left, and an inspired image of your drawing appears on the right

5. Eleven Labs: add your text to have it read with a magical, comedic, or realistic sounding voice

6. Bolt app or Val Town: quickly create mini sites or web apps with a prompt

7. Replicate (with the flux-dev-lora-trainer model): add a dozen photos of yourself, train a model (for about 2 euros), then generate images of yourself in different contexts (about 6 cents per generation).

8. (Bonus) Glif: Add your project’s logo to any image available on the web

Top 3 AI Flops of 2024

AI products that flopped

Rabbit R1 ($199): The Rabbit R1, despite its innovative design, failed to convince due to a lack of clearly defined practical functionalities and real usefulness for the user.

Humane AI Pin ($499 + $24/month): The Humane AI Pin disappointed due to its high price, limited performance, reliance on internet connection, and user experience that didn't meet expectations and its cost.

Meta Ray-Bans (359€): The Meta Ray-Bans did not achieve the expected success, as the AI features were considered gimmicky and lacked real added value to justify their price.

The 4 Automations to Copy for Your Business

These are the use cases integrated by our bootcamp participants that received the best ROI in 2024

1. Invoice analysis on V7

2. Create your GPT to send personalized quotes

3. Rewrite to optimize hundreds of SEO articles with V7 Go

4. Create a GPT to train selling with ChatGPT’s advanced audio mode

Once a cost or time-saving opportunity is identified and implemented, the adoption of this new use case within the company is rarely assured. The entire company must take the “AI reflex”. Discover Side School bootcamps to get into AI.

Average Costs of AI Integration Projects in Companies in 2024

We've created a calculator for the report to help you better estimate costs:

Industrial Trends

Highlights

From Hype to Reality
The hype wave around AI officially began on November 30, 2022, with the release of ChatGPT, gaining considerable momentum throughout 2023. In 2024, we witness the concrete application of generative AI models, with companies, especially large ones, adopting technologies like RAG (Retrieval Augmented Generation) to enhance efficiency and optimize operations.

Acceleration of Generative AI Adoption

The adoption of generative AI accelerates both within enterprises and among young generations. AI-based products are also starting to prove their ability to retain users. Until recently, these tools struggled to keep paying clients after the initial wow effect and trial periods.

Today, top performers include OpenAI, Grammarly, Anthropic, Midjourney, Otter, and ElevenLabs. For instance, ChatGPT now has 300 million active users weekly.

In companies, AI products are also demonstrating loyalty. GenAI products struggled to retain paying clients beyond their initial wow effect and trial periods. Top performers include OpenAI, Grammarly, Anthropic, Midjourney, Otter, and ElevenLabs. ChatGPT now has 300 million weekly users.

OpenAI Still Dominant: Despite OpenAI’s strong position, many companies are actively competing to develop their own large-scale language models (LLM). These competitive approaches manifest through various strategic axes: some focus on optimizing open-source model performance, others explore innovative architectures for more specialized functionalities. A competitive dynamic emerges from these pursuits, resulting in increased accessibility and diversity of AI tools for businesses. This race to innovation is essential for breaking the monopoly and ensuring the sustainability of the AI sector.

Ranking of generative AI models by LiveBench, updated on November 25, 2024. The latest version of Gemini 2.0 Flash, announced afterwards, is, in theory, ranked in the top three:

The dice are rolled every week. The largest tech companies are fighting to build the latest “SOTA” (State of the Art) model.

o1 by OpenAI: The first model that “reasons” The rise of these models, capable of reasoning and learning, opens up numerous applications in the professional world. Companies can thus automate complex tasks, improve decision-making, and personalize the client experience in a more sophisticated way. The new reasoning models notably open up the possibility of solving mathematical, scientific, and programming problems at multiple levels where large language models (LLMs) have historically encountered difficulties.

Maxim Lott, an AI enthusiast, has put multiple AI models to the Mensa online IQ tests. The new o1 version from OpenAI has an IQ of 121, which means it is smarter than 92% of the average population. Previous models and competitors are below average:

An interesting test, but to be taken with caution: the “intelligence” of an artificial model is trained on the type of questions from this kind of test.

The o1 model mainly challenges the skill of “prompt engineering.” The models seem to “understand” more advanced queries more easily and no longer require so many back-and-forths to improve the quality of generated responses.

OpenAI’s 5 Levels of AI Revealed in 2024

OpenAI has defined this year five levels of artificial intelligence to assess its progress toward developing systems surpassing human capabilities and has informed its employees during an internal meeting (which reportedly “leaked,” according to Bloomberg). Currently, the company is at the second level (that of simple reasoners) and is nearing the third, the “Agents.” This classification, inspired by other research work, ranges from AI capable of basic conversation to systems that can perform the work of an entire organization, illustrating OpenAI’s vision to create an advanced AGI (Artificial General Intelligence).

Major Employment Changes in 2024

1. Impact of AI on Freelancing (drop in income measured on Fiverr): Freelancers have experienced a decline in income of about 5% since the launch of ChatGPT (source: University of Washington, NYU). Freelance writers are particularly affected, with some reporting drastic drops in income, from $3000 to $1000 per month. This decline is attributed to AI’s ability to perform tasks previously reserved for humans, such as writing, translation, and web design.

2. Recruitment using AI: The use of AI for CV screening and cover letter writing is expanding, transforming the recruitment process. Are we heading towards the end of cover letters, too easily generated?
   The AI Act (discussed below) will significantly impact the use of AI in recruitment: AI systems used for recruitment are classified as "high risk". Applicants will notably have the right to obtain human intervention in the recruitment process. Ban on AI systems assessing sensitive personal characteristics (ethnicity, political or sexual orientation, etc.). (source: Benjamin Greze for CIO).

3. New hybrid roles combining industry expertise and AI skills: This evolution responds to the growing complexity of business innovation and digital transformation challenges. These professionals are capable of:
   1. Understanding technological challenges.
   2. Dialogue with technical teams.
   3. Proposing solutions adapted to digital transformation projects.

“How should I prepare for AI?” It’s a question we receive every day. Beyond monitoring and adopting new tools, we encourage our bootcamp participants to develop their questioning skills:

1. To better question AI.

2. Questioning reality (false information generated by AI, deep fakes) and...

3. Questioning their profession (monitoring by staying updated, internal upskilling).

This new skill development at Side School accompanies the rise of employees on key generative AI tools.

Would you like to discuss it? Let's set up a meeting together.  

2024 Insiders' News

• NVIDIA still winner of the current cycle.
  According to last year’s count, NVIDIA was used 19 times more than all its peers combined in AI research articles. This year, this advantage has reduced to 11 times, partly due to the 522% growth of articles using TPUs (the gap is now 34 times with NVIDIA). We also note the 353% growth in the use of Huawei’s Ascend 910, the 61% growth of new competitors in large-scale AI chips, and the newly appearing Apple chips. (Source: Compute Index State of AI report)
  
  
  

• Clusters doubling in size in 2024
  The true growth of large-scale GPU clusters comes from the H100. The largest remains Meta’s with 350k H100, followed by 100k from Al and 35k from Tesla. Meanwhile, Lambda, Oracle, and Google have built large clusters totaling over 72k H100. Companies like Hugging Face, DeepL, Recursion, Photoroom, and Magic have built an H100 capacity of over 10k. Additionally, the first GB200 clusters are going live (e.g., 10,752 at the Swiss Supercomputing Center). (Source: Air Street Capital)
  
  
  

• LLM costs slashed by 100
  100x cheaper now for tokens compared to 2024 (GPT-4 ($60 per million tokens) vs GPT-4o-mini ($0.6 per million tokens)). (Source: Air Street Capital)
  
  
  

• Investment in AI startups continues its rise
  Driven by huge generative AI funding rounds such as OpenAI’s 6-billion-dollar raising, the US private market continues to dominate. Total investment in AI companies reached nearly 100 billion dollars. (Source: Air Street Capital) But where are the revenues? Many of the most visible startups working on generative AI raise record funds, often with triple-digit revenue multiples. While this might indicate investor confidence in future returns, it sets a high bar as many of these companies currently have no identified path to profitability. However, this is not the case for everyone, as the largest model providers see their revenues begin to rise.
  
  
  

Macro AI Regulation Changes in 2024

Top 10 AI Personalities of 2024

Researchers

1. Fei-Fei Li: @drfeifei is a renowned researcher in computer vision, known for developing the imageNet database that became pioneering for AI research through neural networks.

2. Andrew NG: @AndrewYNg is a computer scientist and co-founder of Coursera, best known for his deep learning courses.

3. Ilya Sutskever: @ilyasut is an AI researcher, known for his role at OpenAI, which he left this year. This sparked the viral question “What did Ilya see?” referencing his concern over AI control at OpenAI.

4. Andrej Karpathy: @karpathy an AI researcher, has also worked on developing autonomous vehicles at Tesla and creating the program llm.c.

5. Yann LeCun: @ylecun is known for his skepticism toward generative AI’s progress and is an AI ambassador at Meta.

Entrepreneurs

1. Sam Altman: @sama is a key entrepreneur in AI development, particularly with the launch of ChatGPT.

2. Jensen Huang: @jensenhuang is the CEO of Nvidia, a company that produces essential chips for AI.

3. Demis Hassabis: @demishassabis is the CEO and co-founder of DeepMind (owned by Google), a research company in AI responsible for AlphaGo. A pioneer in generative AI.

4. Elon Musk: @elonmusk in addition to his other activities, is a significant player in AI development through his company xAI and is also known for his critiques of AI risks.

5. Dario Amodei: @darioAmodei is the co-founder of Anthropic, an AI research company competing with OpenAI. He is considered a key figure in the development of ethical AI models.

Top AI Events and Communities in 2024

1. Pause AI: An initiative aimed at raising awareness about AI risks and calling for a moratorium on dangerous developments, with active online presence and stimulating meetups in Paris. Events every month for an audience with advanced AI knowledge.

2. OpenAI Events: Online conferences organized by OpenAI to share their latest technological advancements and research. These events have replaced the buzz of Apple keynotes among geeks. Events every 3 months for an audience with medium AI knowledge.

3. Public Workshops by Side School (Promotion): Open workshops organized by Side School to demystify and teach AI use at work, in French. Events every two weeks for the general public.

4. AI Safety Center: A center for research and expertise in AI safety that organizes Mercred'IA in Paris to raise awareness of risks. Events every month for an audience with advanced AI knowledge.

5. NeurIPS (formerly NIPS): the international conference for researchers, organized since 1987, to explore machine learning advances. You can follow the latest research shared at this conference in this newsletter. Events annually for an audience with expert AI knowledge.

AI Predictions for 2025

These predictions are built on AI technological advancements in 2024. As with any prediction, we encourage you to take them cautiously: 

Winter 2025:

• Release of “Operator” by OpenAI, enabling a ChatGPT-like interface to take actions. This will make AI more proactive, going beyond textual responses.

• Default memory activation on all OpenAI models, avoiding context and style re-entry in every conversation, making AI ever more personalized to the user. ChatGPT-generated responses will seem more relevant to everyone.

• Context size doubled on major AI models. AI will be able to analyze more data, allowing finer analyses. But it will mostly enable new ways to interact with generative AIs.

• Launch of Copilot applications, Google AI streaming, and OpenAI on Microsoft and MacOS to continuously share your screen and have a quick model (GPT mini, Gemini Flash, etc.) that provides audio or written feedback. This new way of interacting with AI, in a more intuitive manner, will become prevalent during 2025.

• AI Act challenge: The early implementation of the European AI Act proves more flexible than expected after legislators worry about overstepping their mandates. Balancing innovation and AI regulation will be central to debates, particularly during the AI Action Summit at the Élysée on February 10 and 11, 2025.

Spring 2025:

• A non-technical entrepreneur breaks through thanks to AI. An application or website created solely by someone with no programming skills will go viral (e.g., Top 100 of the App Store). This will be possible thanks to the rise of programming agents and commercial agents (for outreach, content creation, etc.). This story will inspire millions more to launch apps quickly and easily.

• Sora available in the European Union: Deepfakes and cyber intrusions will take a new dimension with the arrival of ultra-realistic video generators (e.g., through a national-scale deepfake scandal), accelerating investment in misinformation training (example in 2024). Fighting misinformation will become a priority, driving investment in training.

• An open source alternative to OpenAI o1 surpasses it on a range of reasoning benchmarks. Competition will make these models more accessible, diverse, and usable at lower costs on one's own servers.

• Accelerationist clubs appear worldwide advocating for tech sector deregulation in AI.

Summer 2025:

• Simplified RAG on ChatGPT. Practically, this means ChatGPT will be able to retrieve and use external information more easily. Imagine an assistant that not only answers your questions but can also quickly extract relevant information from a document without you having to do it manually.

• Release of the boundary model GPT-5. Specifically, one can expect an AI with even more advanced language comprehension and problem-solving capabilities, opening new horizons for practical applications such as complex planning.

• Investment of more than 10 billion dollars from a sovereign state in a major American AI lab triggers a national security review. We will continue to scrutinize the impact of these advances on society and seek ways to use AI more ethically and safely.

• The “uncanny valley” effect is reached on AI-generated videos: we can no longer distinguish between real and artificial videos. New versions of models like Sora, Runway, and Kling are available at a monthly price below 200€ per month. In practice, this could make high-quality video content creation more accessible to all. But it calls for caution in the face of misinformation and video manipulation.

• NVIDIA remains unchallengeable. Challengers fail to significantly impact NVIDIA’s market position. NVIDIA will continue to dominate chip production necessary for AI technologies.

Fall 2025:

• Introduction of generative ad parameterization on Meta, Tiktok, Google. Companies will be able to personalize their ads much more precisely and creatively. We can expect more relevant and targeted ads.

• Adobe announces a new creation interface that is “AI first”, allowing exploration of visual possibilities while creating, with a very fast feedback cycle. This will make visual creation much quicker and intuitive.

• Deployment of virtual employees. In practice, SMEs will use virtual assistants to automate administrative, marketing, and customer service tasks. This solution will improve efficiency and productivity, reducing costs and focusing on key tasks.

Tools to Watch in 2025

• Operator by OpenAI, expected to be released in January, should offer a simple interface to connect with the web and your computer.

• Lindy (and on-premise alternatives like Langflow): “if Zapier and ChatGPT had a child”. These tools to create agents on a web interface are the most advanced currently and are ahead of the competition.

• Recraft: new promising tool for image generation and editing, focused on graphic designers’ tasks

• Agents on Copilot: announced in autumn, the next versions of agents on Copilot (currently like GPTs on ChatGPT) will be able to take actions directly on the Microsoft suite.

• Gemini 2.0: Google catches up and attempts to beat OpenAI by offering its models at low or no costs to rival with its competitor.

The 3 Major AI Challenges in Business for 2025

1. Finding Use Cases, considering automations made possible by the agents. Identifying relevant use cases involves thorough exploration of autonomous agents' potential, capable of optimizing and automating any business process. Their flexibility allows imagining unprecedented innovation scenarios, from multilingual 24/7 customer support to predictive maintenance of critical infrastructures.

2. Driving Adoption: Encouraging AI adoption involves developing a genuine internal culture of change, where each employee learns to leverage the intelligent tools at their disposal. Through training, transparency, and support, a domino effect of innovation and trust is created, naturally accelerating technological transition.

3. Implementing On-Premise AI Systems (on their own servers). Implementing AI solutions within internal infrastructures offers increased control over security, confidentiality, and data sovereignty. This approach allows adapting resources and performance to real needs, creating a custom intelligent ecosystem, responsive and robust against future challenges.

Questions of AI in 2025

In 2025, French companies will face three major AI-related challenges. Each of these challenges deserves an entire report for proper addressing. Here, however, are the main issues:

1. Environmental Impact: Google's energy consumption has increased by 50% this year for training its new AI models. An AI chatbot query consumes much more than a classic web request. And even if these costs are divided by 10 each year, we inexorably advance toward ever-increasing use of large data centers. 

2. Managing Sensitive Data: Strengthening regulatory frameworks and transparency expectations will require companies to better secure their data and strictly comply with data protection laws (like GDPR). Unless the EU changes its strategy to help European companies’ competitiveness… Internal processes will need to adapt to ensure confidentiality and traceability, thus, partially, fostering consumer and authority trust.

3. Ethical Question: Companies will be pushed to prevent algorithmic biases and make AI decisions explainable. But current systems remain quite unpredictable… Discussions on the governance of these systems will continue to grow. Anne Alombert, lecturer in contemporary French philosophy at Université Paris 8, expresses these ethical questions very well. 

AI Training in 2025

The year 2025 will be defined by AI. It’s no longer just a fad; it’s the new reality of how businesses function. But navigating this landscape presents key challenges that, if not met, can leave your business behind:

Challenge #1: Finding the right AI use cases. Are you struggling to see how AI can truly benefit your daily operations? It’s not enough to know AI exists. You need practical applications that yield tangible results.

Challenge #2: Stimulating AI adoption. You may have the tools, but how to engage your team? Resistance to new technologies, fear of the unknown, and lack of understanding can all slow progress.

Challenge #3: Choosing the right format for AI training. Classroom training is often too time-consuming and hard to quantify. Video tutorial libraries on AI, meanwhile, do not deeply change usages. The challenges of AI are urgent and important; a better way to train is needed. 

Challenge #4: Leveraging AI agents for automation. Do you know how to harness AI agents to automate repetitive tasks and significantly improve your operational efficiency? These allow achieving unprecedented levels of automation, freeing up time for high-value activities. AI automation will soon become the norm for all. Are you ready for it? 

Side School: Transform Your Team with Our AI Bootcamps

Side School bootcamps are designed to tackle these challenges head-on, providing your team with the practical knowledge and skills they need to succeed in the AI world of 2025. Here's what you'll gain:

1. Master practical AI tools: Learn which AI tools will transform your business and how to use them for tangible results. It won't just be theories, but concrete tools implemented in your team.

2. Develop an “AI-First” mindset: Adopt new habits and reflexes to integrate AI and be comfortable with the new “AI reflex”. You will no longer feel overwhelmed and be ready to face all challenges that arise.

3. Stay at the forefront of technology: No longer feel left behind, stay constantly at the forefront of the latest AI advancements. Our bootcamps ensure a continuous improvement of your knowledge.

Our bootcamps include:

• Four weeks of online training

• An affordable investment of 1400€ excl. VAT per participant

• Practical workshops, concrete challenges, access to a peer group, a dedicated platform, and free tools

Take the next step: Don’t let the AI revolution pass you by. Give your team the skills to optimize processes in 2025. Join Side School and start transforming your business today. Discover Side School

Discover the testimonials

Conclusions

• AI in 2024: From hype to concrete application. There is an increasing adoption of generative AI in companies, with well-defined use cases and increasingly effective tools but still generally limited understanding.

• The year was marked by significant advancements. Scientific, innovative, and application progress continues to emerge at a bewildering speed.

• Still a general misunderstanding about how generative AI works, and notably the consequences and risks related to data, security, and misinformation.

• Growing importance of “AI agents” and automation. AI automation is no longer a concept, but a reality impacting daily business life. Upcoming AI agents will be capable of making decisions and taking actions, opening new productivity perspectives.

• Faced with this evolution, it is crucial to train and adapt. The job market is undergoing major changes, and it is vital to develop AI skills to remain relevant and not be outpaced by technology.

  
  

“AI Report 2024”, published by Side School.
Written by Ben Issen. Reviewed by Zineb Salamat and Olivier Martin.