From AMD cutting 4% of its workers as it continues an AI push to LAMs, AI continues to shake up the technology landscape. While a Reuters report suggests that genAI progress may be cooling, several prominent figures such as Anthropic CEO Dario Amodei and former Google CEO Eric Schmidt have refuted it. OpenAI explores novel training methods, the UAE’s MGX fund commits $500M to OpenAI’s latest round, and Amazon pledges $110M to university research. A piece on R&D World also notes that a major restructuring is underway across sectors, with SAP investing €2 billion annually in AI while cutting up to10,000 jobs, and Klarna’s AI now handling work previously done by 700 employees. Meanwhile, valuations show signs of moderating, with AI video startup Synthesia’s latest round suggesting a cooling from the astronomical multiples seen earlier this year. Special thanks to Frédéric Célerse, Ph.D., Research Scientist in AI for Chemistry at EPFL, for research assistance in compiling this week’s developments in AI.
1. AI companies explore new training techniques amid model scaling challenges
Source: Reuters
Artificial intelligence firms, including OpenAI, are testing novel training methodologies to overcome the limitations encountered with scaling existing models. These approaches aim to emulate human-like cognitive processes, as by OpenAI’s recently introduced o1 model series have shown. This paradigm shift may influence resource requirements and could “impact AI arms race for resources like chips and energy.” The difficulty in finding new data sources from the web has also led some prominent players to make growing use of synthetic data for training.
2. Large Action Models (LAMs) show promise in grasping complexity
Sources: LinkedIn
Aishwarya Naresh Reganti notes on LinkedIn that Large Action Models (LAMs) could represent a significant step beyond traditional AI agents, distinguished by their ability to handle complex, multi-step tasks through deep learning from action-oriented datasets. Unlike conventional AI agents that primarily make decisions based on existing data, LAMs optimize action sequences and continuously adapt through environmental interactions. This means potentially better autonomous decision-making and execution at scale, though some debate exists about the necessity of the new classification.
What’s the difference between LAMs and AI agents? Aishwarya Naresh Reganti has you covered: “AI agents focus on making decisions based on existing data and user interactions. LAMs, on the other hand, are designed to handle more complex, multi-step tasks by learning from massive, action-oriented datasets,” she wrote.
3. UAE’s $100 billion MGX fund emerges as major AI investor
Source: The Information
MGX, a joint venture between UAE’s Mubadala Investment Co. and G42, has emerged as a significant player in AI investment after shelling outroughly $500 million to OpenAI’s recent $6.6 billion funding round. The Abu Dhabi-based fund, which has access to up to $100 billion for AI investments, is targeting U.S. investments in chips, data centers, and software. With Mubadala executive Ahmed Yahia Al Idrissi and UAE’s Sheikh Tahnoon bin Zayed Al Nahyan at the helm, MGX highlights a strategic push by the UAE to become a major force in AI investment.
4. AI drives major R&D job restructuring across several industries in 2024
Source: R&D World
[Adobe Stock]
5. Eric Schmidt weighs in on AI scaling laws
Source: Business Insider
Eric Schmidt
Eric Schmidt, former CEO of Google, remarked that there is “no evidence” indicating a halt in AI scaling laws, despite industry apprehensions about a potential deceleration in AI advancements. He anticipates that large language models will continue to progress over the next five years, though he acknowledges that scaling laws will eventually encounter limitations. “These large models are scaling with an ability that is unprecedented,” Schmidt said on a recent podcast.
6. IBM launches Docling for enhanced document analysis
Source: LinkedIn
IBM has introduced Docling, a RAG-based system designed to enhance document analysis capabilities in large language model applications. The tool enables improved integration of various document types within prompts, according to Vincent Perrin in a French-language LinkedIn post. The new open-source library (MIT-licensed) that PDFs, DOCX, and PPTX files and exports them to Markdown and JSON. It also supports advanced analysis of PDFs and offers seamless integration with LlamaIndex and LangChain.
[Adobe Stock]
7. OpenAI prepares to launch ‘Operator’ AI agent for task automation
Source: Bloomberg
OpenAI is set to launch a new AI agent codenamed “Operator” in January 2024, capable of performing complex tasks like coding and travel booking on behalf of users. The tool will be released as a research preview and through the company’s API for developers, according to internal sources. This development aligns with industry trends, as competitors like Anthropic, Microsoft, and Google are also developing similar agent-based tools. OpenAI CEO Sam Altman previously hinted at this direction on Reddit, suggesting that “agents will feel like the next giant breakthrough” in AI development.
8. Study finds AI poetry rated more creative than human works
Source: The Guardian
A recent study found that participants rated AI-generated poems as more emotional, creative, and beautiful compared to those by poets like Shakespeare and Lord Byron. However, opinions shifted negatively when participants learned the poems were AI-generated. The study suggests that AI’s use of simpler language may contribute to its appeal.
9. AI startup valuations show signs of cooling as Synthesia raises $150M
Source: The Information
AI video startup Synthesia’s latest funding round indicates a potential cooling in AI valuations, with the company raising $150 million at a $2.1 billion valuation—approximately 30 times its forward revenue of $70 million ARR. While still substantial, this multiple is notably lower than the 100x multiples common in early 2024. The deal, led by New Enterprise Associates (NEA), reflects growing competition in the AI video space, where Synthesia, which creates AI-powered business videos using avatars, faces rivals including Pika Labs and HeyGen, as well as offerings from major players like OpenAI and Google.
10. Brightwave introduces AI-powered financial research assistant
Source: Time
Brightwave has launched an AI-powered research assistant designed to process extensive data sources, including SEC filings, earnings call transcripts, and breaking news, to generate financial analysis. The platform serves clients managing a collective total of over $120 billion in assets, aiming to enhance the efficiency and accuracy of financial research.
Elon Musk in 2019 [Wikipedia]
11. Musk expands legal action against OpenAI and Microsoft
Source: The Wall Street Journal
Elon Musk has intensified his legal battle against OpenAI and Microsoft, alleging antitrust violations and efforts to monopolize the AI industry. In an amended lawsuit, Musk accuses both companies of discouraging investors from funding competitors and claims that OpenAI’s CEO engaged in self-dealing. The lawsuit also names venture capitalist Reid Hoffman as a defendant. OpenAI and Microsoft have denied these allegations.
12. AI and stem cell advances target congenital heart defects
Source: New York Post
The Murdoch Children’s Research Institute in Australia, in collaboration with the Gladstone Institutes in San Francisco, is using AI and stem cell technology to address congenital heart defects. This research aims to develop personalized treatments that prevent disease progression and reduce the need for transplants. The approach could potentially treat or cure these conditions before birth.
13. AI enhances road maintenance in Sydney
Source: The Daily Telegraph
In Sydney, local councils are tapping AI to proactively detect potholes. Shoalhaven City Council has implemented AI technology that monitors road conditions through cameras mounted on garbage trucks. This system has enabled the efficient repair of over 10,000 road defects within three months, allowing for better prioritization of repairs and resource allocation.
14. “Real World LLM Engineering” book gets positive reviews
Source: LinkedIn
A new resource for AI practitioners, “Real World LLM Engineering,” is receiving positive attention from the AI community, shared ML engineer and data scientist Pau Labarta Bajo on LinkedIn. The guide, which Bajo describes as ‘a straight-to-the-point-no-BS book,’ offers practical feedback into working with large language models, addressing real-world implementation challenges and best practices. “The authors take you step by step from a concept to a fully working LLM system, using MLOps best practices,” Bajo added.
15. Amazon commits $110M to university AI research with Trainium chips
Source: Amazon
A Trainium board
Amazon has announced a $110 million investment in university AI research through its Build on Trainium program, which will provide compute hours on AWS Trainium UltraClusters containing up to 40,000 Trainium chips. The initiative includes partnerships with institutions like Carnegie Mellon University and UC Berkeley, offering researchers access to high-performance computing resources for AI model development. Amazon will open-source all research advances from the program and has introduced a new Neuron Kernel Interface (NKI) giving researchers direct access to chip instruction sets for optimization experiments.
16. Qwen 2.5 Coder 32B demonstrates efficient resource use
Source: Hugging Face/Qwen AI
The Qwen 2.5 Coder 32B model has achieved a notable technical milestone, supporting up to 128k input while maintaining the ability to run on a single standard GPU in Google Colab. “Qwen 2.5 Coder 32B just matched GPT4o’s performance on coding,” wrote Unsloth co-founder Daniel Han on LinkedIn. The sentiment regarding the model on Reddit was generally positive.
17. U.S. Homeland Security releases AI integration guidelines
Source: AP News
The U.S. Department of Homeland Security has released guidelines for incorporating AI into critical infrastructure sectors, including power grids, water systems, and air travel networks. These directives emphasize assessment of AI’s potentially hazardous risks and protecting user privacy. AI remains something of a moving target — both in terms of definition and capabilities. Thus, Homeland Security Secretary Alejandro Mayorkas noted that “we intend the framework to be, frankly, a living document and to change as developments in the industry change as well.”
18. Insilico Medicine reports promising AI-drug trial results
Source: LinkedIn
Andrii Buvailo, Ph.D. has the scoop on an encouraging Phase IIa trial from Insilico Medicine targeting idiopathic pulmonary fibrosis (IPF). “This novel treatment could transform options for IPF, a disease with very few effective therapies available today,” Buvailo wrote.
19. xAI seeks $6B funding for massive GPU acquisition
Source: CNBC
Elon Musk’s AI venture xAI is raising up to $6 billion at a $50 billion valuation, with plans to acquire 100,000 NVIDIA chips for a new Memphis data center. The funding round, expected to close next week, includes $5 billion from Middle Eastern sovereign funds and $1 billion from other investors. This sizable hardware acquisition signals xAI’s ambitions to compete directly with established genAI players like OpenAI, Google, and Anthropic. The chips will also support Tesla’s Full Self Driving capabilities through the new Memphis supercomputer. The move comes as Musk begins working more closely with President-elect Trump’s administration on AI policy, particularly as Trump plans to repeal Biden’s executive order on AI in favor of what his campaign describes as development “rooted in Free Speech and Human Flourishing.” Trump has also entrusted Elon Musk and Vivek Ramaswamy will to lead a newly-created Department of Government Efficiency body to trim government spending.
Research assistance: Frédéric Célerse, Ph.D., Research Scientist in AI for Chemistry, Ecole Polytechnique Fédérale de Lausanne.
