TLDR AI 2025-01-21

Your users trust you. Let's keep it that way.

DEEPSEEK'S NEW REASONING MODEL SURPASSES OPENAI'S O1

DeepSeek says that its latest reasoning-focused AI model outperforms OpenAI's O1 model on specific benchmarks, citing improvements in logical inference and factual accuracy.

OPENAI QUIETLY FUNDED INDEPENDENT MATH BENCHMARK BEFORE SETTING RECORD WITH O3

OpenAI's record success with the o3 model on the FrontierMath benchmark revealed its undisclosed funding of the project. Despite excellence in solving complex math problems, a lapse in transparency was admitted by benchmark developer Epoch AI. Future collaboration promises improved openness, particularly regarding funding and data access.

PERPLEXITY AI SUBMITS BID TO MERGE WITH TIKTOK

Perplexity AI has submitted a bid to merge with TikTok US as a TikTok ban approaches. The proposal aims to combine Perplexity, TikTok US, and new equity partners while allowing ByteDance investors to retain their equity. Perplexity's merger offer seeks to integrate more video content into its AI search operations.

3D RECONSTRUCTION OF MICROSTRUCTURES

OpticFusion is a multimodal neural implicit 3D reconstruction of microstructures that fuses white light interferometry and optical microscopy.

DEEPSEEK R1 PAPER

DeepSeek has released a paper along with its strong MIT licensed reasoning models. Importantly, the paper shows that there is no need for extremely fancy process reward models. DeepSeek simply used accuracy of answer, a formatting reward, and large scale RL to encourage reasoning in the model. It found that distilling this reasoning ability via SFT to smaller models gave an incredible boost in final performance.

EVOLVING DEEPER LLM THINKING

Contrary to the direct R1 approach, this work from DeepMind uses an evolutionary strategy, guided by a language model, to modify and propose new responses. This leads to strong reasoning ability, specifically in planning tasks that require some exploration.

here

FAST 4K MESH GENERATION

CaPa can generate high-quality 4K textured mesh under only 30 seconds. It can produce 3D assets for commercial applications such as games, movies, and VR/AR.

GENERATIVE MODEL OF UNBOUNDED 4D CITIES

CityDreamer4D is a framework for creating high-fidelity 4D simulations of urban environments, enabling advanced planning and analysis in smart city applications.

HARDWARE ACCELERATOR DESIGN LEARNING

AIRCHITECT V2 is a learning-based design space exploration tool developed to navigate the challenges of optimizing hardware accelerator designs for deep neural networks.

LAION'S BUD-E AI ASSISTED EDUCATION

Bud-E is a system developed by Laion that can assist with educational queries in a more empathic and personalized way.

ELLIQ MAKER INTUITION ROBOTICS LAUNCHES AI SYSTEM TO SUPPORT CAREGIVERS WHO HELP OLDER ADULTS

Intuition Robotics has launched a new ElliQ solution to assist caregivers of older adults. Featuring an AI-powered companion robot and a smartphone app, the system provides remote care management and helps alleviate caregiver stress by enabling caregivers to track health changes and set care goals. This innovation aims to enhance independence for older adults while transforming caregiving into a more rewarding experience.

META FACES BACKLASH OVER AI PROFILES

Meta plans to replace fact checkers with a community notes model and AI for content moderation. GPTZero introduces a new feature to fact-check AI-generated content, while AI startup Anthropic seeks $2 billion at a $60 billion valuation. Jeff Bezos continues to invest in AI startups, focusing on robotics companies.

DEEPSEEK R1 WEIGHTS (HUGGING FACE HUB)

Weights to the full R1 model, which is 600B+ parameters.

AI CAN WRITE IMPROVED CODE, BUT YOU HAVE TO KNOW HOW TO ASK

LLMs can write better code with iterative prompting.

NVIDIA UNVEILS $3,000 DESKTOP AI COMPUTER FOR HOME RESEARCHERS

Nvidia's Project DIGITS, a $3,000 desktop device debuting at CES 2025, allows users to run AI models with up to 200 billion parameters locally.