This Week in NLP #320
Keep up with what happened in NLP in the week ending Friday 3rd January 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
ByteDance and Tencent have each purchased around 230,000 Nvidia Hopper GPUs in 2024, surpassing all buyers except Microsoft's 485,000 units. [TechRadar]
DeepSeek developed DeepSeek-V3, a world-class AI LM, for just US$5.6m. [The Decoder]
Google's CEO Pichai has warned employees about critical challenges in 2025, citing AI competition and regulatory pressures while emphasizing the company’s focus on Gemini development. [Verdict]
OpenAI's o3 model has achieved breakthrough scores on the ARC benchmark through five key innovations, sparking renewed optimism about AI progress despite computational cost concerns. [VentureBeat]
OpenAI plans to become a public benefit corporation in 2025, shifting from non-profit oversight while retaining charitable operations through its original foundation. [Engadget]
Now read on for everything else that happened in NLP this week.
Making News:
What makes this newsletter different?
This Week’s Topics:
If you’re reading this in a desktop web browser, you can access a navigation menu at the left-hand edge of this window.
The Generative AI Wars
Meta has launched AI character creation tools and plans to integrate AI-generated profiles across its platforms. [Fox Business]
Microsoft confirmed that its OpenAI partnership allows integration of multiple AI models into products, including non-OpenAI options. [Yahoo Finance]
Encode, the nonprofit organization that co-sponsored California’s ill-fated SB 1047 AI safety legislation, has filed to support Elon Musk’s injunction against OpenAI's nonprofit-to-for-profit conversion, arguing it would compromise the company’s commitment to AI safety. [TechCrunch]
Here’s OpenAI’s blog post explaining the shift to a public benefit corporation. [OpenAI]
OpenAI’s ChatGPT experienced a major outage lasting nearly 10 hours due to a Microsoft datacenter power issue, affecting multiple AI services. [The Verge]
OpenAI's former head of go-to-market has argued against the single-model future of AI, predicting a multi-model landscape driven by specialization and market fragmentation. [VentureBeat]
Samsung's upcoming Galaxy S25 phones are expected to include a free Gemini Advanced AI subscription worth over $200, according to code discovered in Google's app. [ZDNet]
Hardware
Airoha Technology has launched the AB1595 wireless AI audio chip, integrating multiple functions into one single system-on-chip. [PR Newswire]
ByteDance has planned a US$7B investment in overseas cloud-based GPUs to bypass US chip export restrictions to China through data center rentals in Southeast Asia and the Middle East. [Verdict]
D-Matrix's Corsair processor delivers GPU-free AI inference with 60,000 tokens/second performance, backed by Microsoft and supported by Nvidia partner Micron Technology. [TechRadar]
Google's Trillium TPU, now available for rent, offers 4x better training performance than previous generations. [TechRadar]
LG has expanded its Gram laptop lineup with AI-powered models featuring Intel processors, enhanced displays, and both cloud-based and on-device AI capabilities through Gram Chat. [The Verge]
Nvidia faces market pressure as Chinese startup DeepSeek demonstrates cost-effective AI models, while geopolitical tensions and emerging competitors threaten its market dominance. [Yahoo Finance]
Ori Industries has secured rights to deploy Nvidia's H200 chips in the UK from January 2025, becoming the country’s first company to offer these advanced AI processors. [UKTN]
Sagence AI is challenging Nvidia with a new analog in-memory compute system promising better energy efficiency and lower costs for AI inference processing. [TechRadar]
It’s Only a Model
DeepSeek-V3 frequently misidentifies itself as ChatGPT, likely due to training on datasets containing GPT-4-generated content. [TechCrunch]
EuroLLM-9B, an open-source language model optimized for European languages, outperforms similar-sized models in multilingual tasks using a specialized 128,000-word tokenizer. [InfoQ]
Alibaba’s Qwen has launched QwQ-32B-Preview, a 32B-parameter AI model with 32K context length that excels in mathematical and scientific reasoning tasks. [InfoQ]
Ruliad has launched DeepThought-8B, a compact LLaMA-based model requiring 16GB VRAM that performs step-by-step reasoning tasks while documenting its decision-making process. [InfoQ]
The LLM Ecosystem
The Association for Intelligent Information Management has released a whitepaper and assessment tool to help organizations evaluate their unstructured data readiness for generative AI implementation. [AIIM]
AWS has enhanced Amazon Q Developer with automated documentation generation, code reviews, and unit test creation capabilities across supported IDEs in all regions where Amazon Q is available. [InfoQ]
DoubleVerify has launched its Generative Artificial Intelligence Website Avoidance & Detection solution, which aims to helps advertisers navigate the risks posed by low-quality, AI-generated content. [MarTech Series]
Graphlit has launched its Agent Tools Library on GitHub, offering developers a comprehensive RAG-as-a-Service platform for building AI agents with streamlined data handling capabilities. [PR Newswire]
Hugging Face has launched Smolagents, a simple library enabling language models to perform agentic tasks through code-based actions rather than JSON snippets. [Hugging Face]
Microsoft has launched serverless GPU capabilities in Azure Container Apps, enabling pay-per-second Nvidia GPU usage for AI workloads without infrastructure management. [InfoQ]
Other LLM Sightings
BrandRep has launched its proprietary AI-powered keyword generator designed to help small and mid-sized businesses supercharge their online visibility. [MarTech Series]
Seoul National University of Science and Technology has developed PV2DOC, a tool that converts presentation videos into concise PDF documents with summaries and images. [PR Newswire]
Total Neural Enterprises’ first product, Persuasion, analyzes a company’s spreadsheets, databases, documents, and more to develop recommendations. [GeekWire]
Risks and Responses
Forrester advises organizations to proactively address AI regulation compliance across multiple jurisdictions, rather than waiting for specific legislation to be enforced. [Computer Weekly]
IBM reports that 56% of businesses are delaying AI investments due to ethical concerns. [ZDNet]
OpenAI has delayed its promised Media Manager tool, which was announced in May to help creators control their works’ inclusion in AI training data. [TechCrunch]
Conversational AI
Cerence AI has partnered with JLR in a multi-year agreement to develop next-generation AI-powered in-car experiences for luxury vehicles worldwide. [GlobeNewswire]
LXRGuide has announced the January 2025 launch of Intellichat, an AI-powered conversational marketing assistant designed to help small online businesses compete with larger brands. [PR Newswire]
SoundHound AI has established itself as a leader in enterprise-focused conversational AI, differentiating from Big Tech competitors through its Speech-to-Meaning technology and white-label solutions. [Yahoo Finance]
Wilby.ai has launched ‘Simon Says’, an AI-powered multilingual assistant that handles voice commands in 165 languages for tasks like calls, bookings, and translations. [AiThority]
Voice News
Apple is developing a redesigned Magic Mouse with AI features and voice command capabilities. [ZDNet]
HitPaw has released VoicePea V2.4.0, adding 30+ new voices and improved AI effects to its voice-changing software. [PR Newswire]
And HitPaw has released Edimakor V3.4.0, featuring a new AI voice changer with 50+ voice options and enhanced AI tools for video editing. [PR Newswire]
iWallet has launched what it says is the world’s first PCI-compliant AI-based voice payment system for telephone orders. [PR Newswire]
SoundHound AI has partnered with Lucid to launch the Lucid Assistant, a multilingual voice-control system powered by generative AI for their electric vehicles. [Business Wire]
Viaim has announced its RecDot AI earbuds and NoteKit desktop recorder, offering AI-powered transcription, translation, and meeting assistance features. [PR Newswire]
Legal Tech
AI.Law has launched an AI-powered platform enabling individuals to draft professional lawsuits without attorneys, making legal action more accessible and affordable. [AiThority]
Funding
Across AI launched with US$5.7m seed funding to develop next-generation agentic AI software for enterprise workflows, targeting B2B sales processes. [CRN]
Nodepay has raised US$7m to expand its decentralized AI platform that converts unused internet bandwidth into real-time data for AI training. [PR Newswire]
Nvidia has acquired Run:ai for a reported US$700m and announced plans to open-source the GPU cloud orchestration software company’s platform. [VentureBeat]
There’s More
AI progress in 2025 will be ‘even more dramatic’, says Anthropic co-founder Jack Clark. [The Decoder]
Meanwhile, Meta's AI chief Yann LeCun dismisses AGI hype, arguing that human-like AI requires more than language models, including sensory learning, emotions, and world-modeling capabilities. [The Decoder]
AI medical scribes have transformed healthcare documentation by automating transcription of doctor-patient conversations, reducing physician burnout and improving patient care quality. [Deepgram]
OpenAI is tripling the size of its Washington DC policy team. [Politico]
The MIT Technology Review looks at the biggest AI flops of 2024. [MIT Technology Review]
The Back Cover
Here’s Keith Jarrett with a cover of Harold Arlen’s Somewhere Over the Rainbow.
If you know of a great cover version of a back-catalogue classic, drop an email to us at backcover@language-technology.com and we’ll consider it for inclusion here.
Got this from someone else?
Did you find this newsletter useful? if so, you might forward it to a friend; or you could email us at news@language-technology.com to tell us what you want more of.
Did you hate this newsletter? if so, you could forward it to an enemy; or you could email us at news@language-technology.com to tell us why – make it better!