DeepSeek Diving + The News You Missed
For those that are new around here, this is an extension of my weekly newsletter where I highlight new and innovative AI products that are worth exploring.
Hey hey! Happy not-Wednesday, hereâs your news in newsletter! Due to some technical errors This was supposed to land Friday, Jan 31st, and is now in your inboxes Feb 4th đ In this Issue:
The DeepSeek Splash đ
Coming out of stealth mode comes a new LLM Challenger: DeepSeek. Boasting bold benchmark performance, this model was allegedly developed for a fraction of the cost of OpenAIâs models on a per iteration basis. Visit their website here.
Your Weekly Industry News Recap
A new entrant swings hard right out of the gates: New entrants arenât something new to the AI scene - especially not for LLMs. But one that comes in this late in the game swinging this hard is refreshing, and frankly unexpected! The Potential to Shrink Operating Costs vs competitors: Startups and those experimenting with LLMs in their environments are now looking at lower op costs due to DeepSeekâs architecture barely requiring flashy GPUs. See more at Hugging Face.
DeepSeek Proves that Reinforcement Learning has a place in LLMs
Itâs the biggest proof that large models can improve their performance dramatically from pure reinforcement learning. Venture Beat covers the basics nicely of what is and why itâs impressive. Read more on VentureBeat.
Mind you that even when GPT4 released, OpenAI had tried adding a layer of reinforcement learning to tie bow on their modelâs development, and had only noticed minor increases in performance. I ran this LLM through what I could design as a gauntlet. Coding questions, philosophy, rationales behind decisions, detailed science questions, niche local law, etc.. and it all came back pretty clean. I steered clear of politics due to some obvious statements in the Privacy Policy. Diving deeper into some obscure Quebecâs law structures was no problem at all for this model, itâs got link, citation as well as recommendations for further sources. When asked a vague technical question about building a DIY air purifier, I turned on the R1 reasoning features, and I got a stream of consciousness response leading up to my âofficial promptâ response. Even after further pressing the system by diving deeper into my lines of inquiry, the R1 model impresses in the thoroughness of its responses.
LâOrĂ©al & IBMâs AI for Cosmetics: AI is transforming beauty! LâOrĂ©al and IBM introduce generative AI for personalized skincare. Read more here.
The API legend veers into AI: Postmanâs AI Agent Builder Launch: API development is getting smarter! Postman debuts an AI agent builder to automate API processes. Read more here.
The AI Governance Balancing Act (World Bank): How should regulators balance AIâs potential with risks? Read more here.
