# State of AI Report 2024: accelerating progress and a new technological era Author: Fernando Nieto Lobato Original publication: 2024-10-16 Spanish original: https://estrategiabyaleph.substack.com/p/estrategia-55-el-estado-de-la-ia English URL: https://elcontemplador.github.io/estrategia-english/essays/055/ Status: Published translation This is a translation of the original Spanish essay published on 16 October 2024. Its claims, examples and forecasts retain that historical context. English publication: 2026-09-29 *Historical note: this article discusses the October 2024 report and its predictions for the following twelve months. Assessments and forecasts are presented as they stood then.* The [2024 edition of the *State of AI Report*](https://www.stateof.ai/), produced by Nathan Benaich and the Air Street Capital team, was published last week. Published annually since 2018, the report has become a key reference in the artificial intelligence sector. It provides a comprehensive view of advances in AI across research, industry, politics and safety, offering historical analysis, future predictions and practical guidance for professionals and enthusiasts. The report makes very interesting reading, and we recommend it. Even so, we know that its more than 200 pages demand a great deal of time from most of you, so we bring you the following summary of the highlights in each of its four main areas. The image below also shows the authors’ own executive summary: [![State of AI Report 2024 executive summary, with three points each on research, industry, politics and safety; full transcript follows.](https://elcontemplador.github.io/estrategia-english/assets/images/b85fbe04-beed-4c71-a68d-d33fcd02c05a_1600x886.png)](https://substackcdn.com/image/fetch/$s_!kmqF!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb85fbe04-beed-4c71-a68d-d33fcd02c05a_1600x886.png) ### Accessible transcript: the report’s executive summary *Original English text from the accompanying slide, page 8:* > **Research** > > - Frontier lab performance converges, but OpenAI maintains its edge following the launch of o1, as planning and reasoning emerge as a major frontier. > - Foundation models demonstrate their ability to break out of language as multimodal research drives into mathematics, biology, genomics, the physical sciences, and neuroscience. > - US sanctions fail to stop Chinese (V)LLMs rising up community leaderboards. > > **Industry** > > - NVIDIA remains the most powerful company in the world, enjoying a stint in the $3T club, while regulators probe the concentrations of power within GenAI. > - More established GenAI companies bring in billions of dollars in revenue, while start-ups begin to gain traction in sectors like video and audio generation. Although companies begin to make the journey from model to product, long-term questions around pricing and sustainability remain unresolved. > - Driven by a bull run in public markets, AI companies reach $9T in value, while investment levels grow healthily in private companies. > > **Politics** > > - While global governance efforts stall, national and regional AI regulation has continued to advance, with controversial legislation passing in the US and EU. > - The reality of compute requirements forces Big Tech companies to reckon with real-world physical constraints on scaling and their own emissions targets. Meanwhile, governments’ own attempts to build capacity continue to lag. > - Anticipated AI effects on elections, employment and a range of other sensitive areas are yet to be realized at any scale. > > **Safety** > > - A vibe-shift from safety to acceleration takes place as companies that previously warned us about the pending extinction of humanity need to ramp up enterprise sales and usage of their consumer apps. > - Governments around the world emulate the UK in building up state capacity around AI safety, launching institutes and studying critical national infrastructure for potential vulnerabilities. > - Every proposed jailbreaking ‘fix’ has failed, but researchers are increasingly concerned with more sophisticated, long-term attacks. ## 1. Research: competition between giants and converging capabilities The report highlights how frontier AI laboratories are converging in their capabilities, with companies such as OpenAI, Meta and Anthropic leading the way. Despite OpenAI’s initial advantage with GPT-4, the performance gaps between leading models, such as Anthropic’s Claude 3.5, DeepMind’s Grok 2 and Meta’s Llama 3.1, have narrowed considerably, although the recent o1 has somewhat reversed the trend. This has sparked a race to develop new features and functions beyond simply refining language models. Indeed, the launch of models spanning multiple modalities—text, images, video and even robotic interactions—is highlighted as a new turning point in research. *Source note: the attribution of Grok 2 to DeepMind is reproduced from the Spanish article. It remains an editorial verification point.* One of the most promising emerging areas is multimodal generative models, capable of applying AI to fields as diverse as mathematics, biology and the physical sciences. This multimodal approach is enabling significant advances in fields such as genomics and neuroscience, paving the way for applications that go far beyond the natural language processing that has dominated AI development in recent years. The report also highlights the impact of open-source models such as Meta’s Llama 3, which is levelling the playing field with powerful proprietary models. It warns, however, that the term “open source” is being used ambiguously, creating confusion over what “open” really means in terms of access to data, models and weights. ## 2. Industry: centralisation and an AI-driven economic boom AI’s commercial impact continues to accelerate, with NVIDIA emerging as one of the world’s most influential companies thanks to its control of the graphics processing unit (GPU) market. GPU infrastructure is essential for training large-scale AI models. This has made NVIDIA a dominant player in the AI ecosystem, briefly reaching a market valuation of $3 trillion. However, the report notes growing concern about power being concentrated in a handful of technology giants, raising questions about fairness and long-term sustainability. Meanwhile, start-ups in sectors such as AI-generated video and audio are beginning to earn substantial revenues, indicating that commercial AI adoption is moving ahead at full speed. Despite this optimism, doubts remain about the long-term viability of business models, with many start-ups facing the challenge of turning technological advances into sustainable products. The total market value of AI companies reached $9 trillion, reflecting strong investment growth in both public and private companies. Yet it remains to be seen how the ecosystem will evolve, particularly regarding the cost of training and maintaining these increasingly complex models. ## 3. Politics: national regulation versus global efforts The AI regulatory landscape remains a critical issue, and the report highlights the divergence between national approaches and efforts at global governance. Although key legislation has passed in the United States and the European Union, such as the EU AI Act, international efforts to establish unified rules have largely failed, leaving only high-level voluntary commitments. One issue that has gained prominence is AI’s energy demand and associated carbon emissions. The report notes that major technology companies are encountering real physical limits to scaling their infrastructure because of compute requirements, calling their sustainability targets into question. As governments try to build their own AI capacity, barriers to entry remain high, resulting in a growing gap between the capabilities of public and private actors. The report also warns of AI’s potential effects in sensitive areas such as elections and employment. However, these effects have not yet materialised at scale, and there is some uncertainty about how to manage these transformations. ## 4. Safety: from caution to acceleration One of the report’s most controversial points is the shift in emphasis among leading AI companies, from warning about AI’s existential risks to focusing on maximising sales and business adoption. This shift has drawn criticism from the AI safety community, which fears that concerns about long-term catastrophic risks will be neglected in favour of immediate economic gains. Governments, for their part, are beginning to build AI safety capacity through initiatives such as the establishment of dedicated AI safety institutes in the United Kingdom. These efforts aim to identify potential vulnerabilities in critical infrastructure that could be exploited by advanced AI systems. ## Predictions for 2025 One of the report’s most interesting features is its annual predictions. An early slide records the extent to which its 2023 predictions for 2024 came true; we include the image below. Another, towards the end, sets out predictions for 2025. Alongside the image, we also provide them in translation. Broadly speaking, they strike us as fairly sound and plausible, although we do not agree with all of them. We will probably attempt our own AI predictions for 2025 towards the end of the year. [![Ten 2023 predictions assessed in the 2024 report with yes, no or intermediate status and evidence; an accessible table follows.](https://elcontemplador.github.io/estrategia-english/assets/images/c383c6ea-8bd3-4f92-9024-54cba877575d_1600x891.png)](https://substackcdn.com/image/fetch/$s_!6xh-!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc383c6ea-8bd3-4f92-9024-54cba877575d_1600x891.png) ### Accessible transcript: the report’s assessment of its 2023 predictions *The following reproduces the original English slide, page 10, including the report authors’ own status labels and evidence as of 2024. A tilde (~) is the intermediate status displayed on the slide.* | Our 2023 prediction | Status shown | Evidence | |---|---|---| | A Hollywood-grade production makes use of generative AI for visual effects. | YES | Largely badly, but GenAI AI visual effects have been seen in Netflix and HBO productions. | | A generative AI media company is investigated for its misuse during in the 2024 US election circuit. | ~ | Not yet, but there’s still time. | | Self-improving AI agents crush SOTA in a complex environment (e.g. AAA game, tool use, science). | NO | Not yet, despite promising work on open-endedness, including strong game performance. | | Tech IPO markets unthaw and we see at least one major listing for an AI-focused company (e.g. DBRX). | ~ | While the Magnificent Seven have enjoyed strong gains, private companies are hanging on until markets settle. However, AI chip company Cerebras has filed to IPO. | | The GenAI scaling craze sees a group spend >$1B to train a single large-scale model. | NO | Not quite yet - let’s give it another year. | | The US’s FTC or UK’s CMA investigate the Microsoft/OpenAI deal on competition grounds. | YES | Both regulators are investigating this partnership. | | We see limited progress on global AI governance beyond high-level voluntary commitments. | YES | The commitments from Bletchley and Seoul summits remain voluntary and high-level. | | Financial institutions launch GPU debt funds to replace VC equity dollars for compute funding. | NO | Some VC funds are rumored to be offering GPUs for equity, but we’re yet to see anyone go down the debt route. | | An AI-generated song breaks into the Billboard Hot 100 Top 10 or the Spotify Top Hits 2024. | YES | It turns out this had already happened last year with “Heart on My Sleeve”, but we’ve also seen an AI-generated song reach #27 in Germany and spend several days in the Top 50. | | As inference workloads and costs grow significantly, a large AI company (e.g. OpenAI) acquires or builds an inference-focused AI chip company. | YES | Sam Altman is reportedly raising huge sums of money to do this, while each of Google, Amazon, Meta and Microsoft continue to build and improve their owned AI silicon. | ### Ten predictions for the next twelve months *Translation note: the Spanish article translated these predictions from the report. Here their original English wording is restored from the accompanying slide, page 205. In prediction 1 the slide specifies a national security “review”, rendered more broadly as “intervention” in the Spanish article; in prediction 8 it explicitly says “on-device”.* 1. A $10B+ investment from a sovereign state into a US large AI lab invokes national security review. 2. An app or website created solely by someone with no coding ability will go viral (e.g. App Store Top-100). 3. Frontier labs implement meaningful changes to data collection practices after cases begin reaching trial. 4. Early EU AI Act implementation ends up softer than anticipated after lawmakers worry they’ve overreached. 5. An open source alternative to OpenAI o1 surpasses it across a range of reasoning benchmarks. 6. Challengers fail to make any meaningful dent in NVIDIA’s market position. 7. Levels of investment in humanoids will trail off, as companies struggle to achieve product-market fit. 8. Strong results from Apple’s on-device research accelerates momentum around personal on-device AI. 9. A research paper generated by an AI Scientist is accepted at a major ML conference or workshop. 10. A video game based around interacting with GenAI-based elements will achieve break-out status. [![Ten State of AI Report predictions for the twelve months after October 2024; their original English wording is transcribed above.](https://elcontemplador.github.io/estrategia-english/assets/images/1ae20c26-750d-48f8-bd5b-1d58302e4840_1600x894.png)](https://substackcdn.com/image/fetch/$s_!end6!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1ae20c26-750d-48f8-bd5b-1d58302e4840_1600x894.png) *The State of AI Report’s original English slide of ten predictions, transcribed above.* Fernando Nieto Lobato *Director of Digital Innovation at Institución Educativa ALEPH*