2024 AI Annual Report Released with Top 10 Predictions: humanoid robots heat up, NVIDIA maintains dominance

AI technology is changing rapidly, with big companies like Netflix already using AI to create special effects, and new AI models are outperforming humans in games and scientific research. In the coming year, we may see people with no programming skills creating hit apps, and even AI-created songs may make it to the top 10 of the music charts.

The momentum of AI development is rapid, from technology, policy to product risk, compared with the former, there are great changes.

Recently, State of AI released the 2024 Industry Report, the seventh annual report released by the research team, which provides an in-depth analysis of the current status and expectations of the AI industry from the perspectives of research progress, industry situation, existing policies, safety issues, and future predictions.

Report Link:https://docs.google.com/presentation/d/1GmZmoWOa2O92BPrncRcTKa15xvQGhq7g4I4hJSNlC0M/edit?usp=sharing

The report focuses on the following:

1. the performance of frontier labs is starting to converge, the advantage of proprietary models is fading, and the gap between GPT-4 and the others is narrowing. openAI's o1 model has once again put the labs at the top of the leaderboards, but for how much longer?

2. Planning and reasoning have taken precedence in large-scale language model (LLM) research, and companies are exploring combining LLM with reinforcement learning, evolutionary algorithms, and self-improvement to unlock future applications for intelligentsia.

3. Foundational models demonstrate their ability to break out of language and support multimodal research across mathematics, biology, genomics, physical sciences and neuroscience.

4. The sanctions imposed by the United States on Chinese laboratories have had a limited impact on their ability to produce high-performance models, as a combination of inventory, approved hardware, smuggling, and cloud access has enabled the construction of high-performance (V)LLMs domestically, but the development of the semiconductor industry in the country remains relatively disorganized.

5. Because publicly traded companies have experienced a bull market in AI exposure, the enterprise value of AI companies has reached $9 trillion. Investment in private AI companies has also increased, but the increase is an order of magnitude smaller, despite the emergence of GenAI mega-funding in the United States.

6. A number of AI companies are beginning to generate significant revenues, including basic model builders and start-ups engaged in video and audio generation. However, questions around long-term sustainability remain unanswered as models become cheaper as part of a corporate grab for market share.

7. Pseudo-acquisitions have become a way out for AI companies, as some have struggled to find a viable business model because of the high cost of staying ahead of the technology.

8. Discussions of existential risks have cooled in the wake of the attempted coup d'état of OpenAI. However, researchers have been deepening the understanding of potential model vulnerabilities and misuse, and suggesting possible fixes and safeguards.

Research progress

1. Frontier labs are converging in performance, but OpenAI has maintained its leadership position since the launch of o1, with planning and reasoning emerging as the main research frontiers.

2. Foundational models have demonstrated their ability to transcend language, and multimodal research has advanced mathematics, biology, genomics, physical sciences and neuroscience.

3. U.S. sanctions have not prevented China's Large Language Model (V) LLM from rising up the community charts.

industry situation

1. NVIDIA remains one of the world's most powerful companies, enjoying a moment of glory as its market capitalization reaches $3 trillion, and regulators are investigating the concentration of power within generative artificial intelligence (GenAI).

2. The more established GenAI companies are bringing in billions of dollars in revenue, while startups are beginning to gain traction in areas such as video and audio generation. While companies are beginning to move from models to products, long-term issues around pricing and sustainability remain unresolved.

3. Driven by a bull market in the public markets, the market capitalization of AI companies has reached $9 trillion, while the level of investment in private companies is also growing at a healthy rate.

Existing policies

1. While global governance efforts have stalled, national and regional AI regulation continues to advance, with several controversial legislative bills passed in the United States and the European Union.

2. The realities of computing demand are forcing large technology companies to confront expanding real-world physical constraints and emissions targets. Meanwhile, governments themselves continue to lag behind in their attempts to build capacity.

3. The expected impact of AI on elections, employment and other sensitive areas has not yet materialized.

security issue

1. With the need to increase the use of enterprise sales and consumer applications by companies that previously warned us of the impending extinction of the human race, the climate for moving from security to acceleration has changed.

2. Governments around the world have mimicked the United Kingdom's efforts to build national capacity in AI security by establishing research institutes and studying potential vulnerabilities in critical national infrastructure.

3. Each of the proposed cracking 'fixes' failed, and researchers began to focus on more sophisticated, long-term attacks.

Top 10 Predictions for 2025

The predictions cover all aspects of the AI space, from large investments to the rise of no-code apps to the implementation of data privacy and AI regulations, and even the competition for open-source technologies, challenges for market leaders, and trends in humanoid robotics.

1. A sovereign nation invests more than $10 billion in a major U.S. artificial intelligence lab, triggering national security scrutiny.

2. an app or website created by someone with absolutely no coding skills goes viral, e.g. into the top 100 of the Apple App Store.

3. The Frontier Laboratory experienced a significant shift in the way it practiced data collection after the commencement of trials in cases.

4. Early implementation of the EU AI bill turned out to be more lenient than expected, and lawmakers may worry that they went too far.

5. Open source OpenAI o1 alternatives outperform in a series of inference benchmarks.

6. The challenger failed to make any meaningful dent in NVIDIA's market position.

7. The level of investment in humanoid robots will decline as companies struggle to achieve product market fit.

8. Apple's research on devices has yielded strong results, accelerating the momentum around artificial intelligence on personal devices.

9. A research paper generated by an AI scientist is accepted at a major machine learning conference or workshop.

10. A video game that interacts with elements of "generative artificial intelligence" will achieve breakthrough status.

2023 Annual Report Forecast Results

correct forecast

1. Hollywood-level production companies are starting to use generative AI for visual effects.

The proof: while most of the effects aren't ideal yet, generative AI visual effects are already appearing in Netflix and HBO productions.

2. The U.S. Federal Trade Commission (FTC) or the U.K. Competition and Markets Authority (CMA) investigated the Microsoft/OpenAI transaction on competitive grounds.

Evidence: both regulators are investigating the partnership.

3. There has been very limited progress on global AI governance, which would go beyond high-level voluntary commitments.

EVIDENCE: Yes, the commitment from Bletchley and the Seoul Summit remains voluntary and high level.

4. A song written by an AI enters the Billboard Hot 100 top 10 or the Spotify 2024 Hot 100.

Evidence: the song 'Heart on My Sleeve' is an example of this, and it can also be seen that a song written by an AI made it to #27 in Germany and stayed there for a few days.

5. As reasoning workloads and costs have grown significantly, a large AI company (e.g., OpenAI) has acquired or established an AI chip company focused on reasoning.

The evidence: Sam Altman is reportedly raising a lot of money to make it happen, while companies like Google, Amazon, Meta, and Microsoft continue to build and improve their own AI chips.

false forecast

1. There are generative AI media companies under investigation for their abusive practices during the 2024 US elections.

Explanation: not yet, but let the bullets fly for a while.

2. Self-improving AI intelligences exceed the highest levels of current technology in complex environments (e.g., AAA gaming, tool use, scientific exploration).

Explanation: not yet, although there is some more promising work on openness, such as a strong showing on games.

3. The tech IPO market thawed and saw at least one AI-focused company go public (e.g. DBRX).

Explanation: while the Magnificent Seven has enjoyed strong growth, the private company is still waiting for the market to stabilize; artificial intelligence chip company Cerebras has filed for an IPO.

4. In the scaling boom of generative AI, one team spent over a billion dollars to train a large-scale model.

Explanation: it's not quite there yet, so you can wait until next year to see.

5. Financial institutions launch GPU debt funds to replace venture capital equity dollars for computing funding.

Explanation: not yet, some VC funds are rumored to offer GPUs in exchange for equity, but haven't seen anyone go the debt route yet.

References:

https://x.com/nathanbenaich/status/1844263448831758767

https://www.stateof.ai/

Report Download:

Link: https://pan.baidu.com/s/12_xWsqSWD-aUXF9r-6c4VA?pwd=j4t5