Generative's AI Newsletter
Posts
🇨🇳 New Chinese Models, Perplexity Merge With TikTok and How Good AI Is In History

🇨🇳 New Chinese Models, Perplexity Merge With TikTok and How Good AI Is In History

Plus, don’t miss out on 5 top customer service tools with exclusive GenAI discounts! 🚀

Generative AI Daily Newsletter
January 20, 2025

AI Innovators, Gather Round

Unlock 6 months of Notion AI for free, explore Perplexity AI’s bold bid to merge with TikTok, and learn how the Pentagon is using AI to accelerate its “kill chain.” Meanwhile, a new study highlights AI’s struggles with history, while Chinese AI company MiniMax claims its new models rival the industry’s best.

In today’s knowledge-packed edition:

Unlock 6 Months of Notion AI—Free For Startups!
Perplexity AI Submits Bid To Merge With TikTok
The Pentagon Says AI Is Speeding Up Its ‘kill chain’
AI Isn’t Very Good At History, New Paper Finds
Chinese AI Company MiniMax Releases New Models It Claims Are Competitive With The Industry’s Best
5 Customer Service Tools with Exclusive GenAI Discounts

🆓 Unlock 6 Months of Notion AI—Free For Startups!

Join thousands of startups that trust Notion as their all-in-one workspace!
For a limited time, you can access 6 months of Notion’s Plus plan—including unlimited Notion AI—to power up your productivity.

🤔 Why Notion?

Create and share docs, take notes, manage projects, and organize knowledge, all in one place.

👇How to Claim Your Offer:

Submit your application via this link https://ntn.so/generativeai and choose Generative AI in the partner list.
Use the partner key: STARTUP4110P37330
This exclusive offer is only available through the Beehiive newsletter— Don't Miss Out!

Act now and transform your workflow with Notion!

Perplexity AI Submits Bid To Merge With TikTok

DAVID PAUL MORRIS/

Perplexity AI has submitted a bid to merge with TikTok US, aiming to create a new entity that includes Perplexity, TikTok US, and new equity partners.

The move comes as ByteDance faces a US government deadline to sell TikTok or face a ban starting January 19, President Biden’s last day in office.

While President-elect Trump has indicated he might grant a 90-day extension, TikTok has warned it may have to shut down operations without clear assurances from the current administration.

Perplexity hopes the merger will integrate more video content into its AI search engine.

The Pentagon Says AI Is Speeding Up Its ‘kill chain’

BRYCE DURBIN

Leading AI developers like OpenAI, Anthropic, and Meta are partnering with the Pentagon to enhance military efficiency using AI, focusing on identifying and assessing threats while ensuring these systems do not harm humans.

Generative AI is being deployed to accelerate decision-making in the military’s “kill chain” processes, particularly during planning and strategizing phases. Companies like Anthropic and OpenAI are striking a balance between enabling innovation and adhering to policies that prohibit harmful applications of AI.

Recent collaborations include Meta working with Lockheed Martin, Anthropic teaming up with Palantir, and OpenAI partnering with Anduril. These partnerships highlight the growing role of AI in defense, despite ongoing debates about the ethical implications of autonomous weaponry and life-and-death decisions.

The Pentagon emphasizes human oversight in decision-making processes and aims to use AI as a collaborative tool rather than an independent decision-maker.

While the defense sector increasingly integrates AI, some experts argue that working with militaries is essential to ensure responsible deployment and prevent misuse, recognizing that AI’s involvement in defense is inevitable.

AI Isn’t Very Good At History, New Paper Finds

METAMORWORKS

A new study has revealed that large language models (LLMs) like GPT-4, Meta’s Llama, and Google’s Gemini struggle with advanced historical questions, achieving only 46% accuracy at best. Researchers tested these models using the Hist-LLM benchmark, which draws from the Seshat Global History Databank.

While LLMs excel at tasks like coding, their historical knowledge is limited, often extrapolating from prominent data and failing with obscure or nuanced topics.

For example, GPT-4 incorrectly stated that scale armor existed in ancient Egypt 1,500 years before it actually appeared, highlighting gaps in historical understanding. The study also identified regional biases, with worse performance for areas like sub-Saharan Africa.

Despite these limitations, researchers remain optimistic that refining benchmarks and training data could help LLMs support historians in the future.

Chinese AI Company MiniMax Releases New Models It Claims Are Competitive With The Industry’s Best

Yuchiro Chino

Chinese AI startup MiniMax, backed by Alibaba and Tencent, has unveiled three new models—MiniMax-Text-01, MiniMax-VL-01, and T2A-01-HD—showcasing capabilities that rival leading U.S. AI systems.

MiniMax-Text-01, with 456 billion parameters and a 4-million-token context window, excels at text-based tasks, outperforming models like Google’s Gemini 2.0 Flash on benchmarks such as MMLU. MiniMax-VL-01 supports multimodal understanding, while T2A-01-HD specializes in advanced audio generation, cloning voices in 17 languages from just 10 seconds of input.

While these models are available via GitHub and Hugging Face, licensing restrictions limit their open-source usability. MiniMax’s projects, such as the AI role-playing app Talkie and text-to-video tools, have sparked controversies over training on copyrighted material. The new releases come amid escalating U.S. export restrictions on AI technologies for Chinese ventures, tightening access to advanced chips and models necessary for cutting-edge AI development.