NeuralByte's weekly AI rundown - 21th January
AI that can solve geometry at Olympiad level, first humanoid robots at BMW, and AMIE: medical diagnostic model more empathetic than doctors.
Greetings fellow AI enthusiasts!
This week is quite busy. This time it is full of new breakthrough studies. For example, you'll learn about AlphaGeometry, Google's model that really came close to breaking the human record in the Geometry Olympiad. Or that another Google model called AMIE was rated more empathetic and helpful than human doctors in a blind test. And there's so much more!
🧠 Stay curious!
Dear subscribers,
Thanks for reading my newsletter and supporting my work. I have more AI content to share with you soon. Everything is free for now, but if you like my work, please consider becoming a paid subscriber. This will help me create more and better content for you.
Now, let's dive into the AI rundown to keep you in the loop on the latest happenings:
🔵 How AlphaGeometry, solves geometry problems at the Olympiad level
🦾 Microsoft Introduces Copilot Pro: A new era of AI-powered office features
📱 Samsung’s Galaxy S24 line: Camera enhancements and generative AI
☝️ Study finds fingerprints of same person share strong similarities
🚗 BMW to deploy Figure’s humanoid robot at South Carolina plant
🧑⚕️ AMIE outperforms human doctors in bedside manner and accuracy of diagnoses.
🏫 OpenAI forges its first university partnership
💵 How AI can boost the global economy and benefit humanity
⛔ Google’s novel framework ASPIRE improves language model’s answering skills
♾️ Meta to boost AI hardware with Nvidia and AMD GPUs
🧲 Meta unveils MAGNeT: A ne open-source text-to-audio model
And more!
How AlphaGeometry, solves geometry problems at the Olympiad level
DeepMind has recently introduced AlphaGeometry, an AI system that has demonstrated the ability to solve complex geometry problems at a level approaching a human Olympiad gold medalist. This breakthrough in AI performance was revealed in a paper published on January 17, 2024. In a benchmarking test of 30 Olympiad geometry problems, AlphaGeometry solved 25 within the standard Olympiad time limit.
The details:
AlphaGeometry is a neuro-symbolic system made up of a neural language model and a symbolic deduction engine.
It uses a new approach that combines the predictive power of a neural language model with a rule-bound deduction engine.
The system was trained on a vast pool of synthetic training data - 100 million unique examples.
This training method allowed AlphaGeometry to train itself by synthesizing millions of known theorems and proofs with various levels of complexity.
In the benchmarking set of 30 Olympiad geometry problems (IMO-AG-30), compiled from the Olympiads from 2000 to 2022, AlphaGeometry solved 25 problems under competition time limits.
This performance is approaching the average score of human gold medalists on these same problems.
The previous state-of-the-art approach, known as “Wu’s method”, solved 10 of these geometry problems.
Why it’s important:
The introduction of AlphaGeometry represents a significant milestone in the development of AI systems capable of deep mathematical reasoning. Solving Olympiad-level geometry problems is a crucial step towards more advanced and general AI systems. The open-sourcing of the AlphaGeometry code and model could open up new possibilities across mathematics, science, and AI.
If you want to learn more about AlphaGeometry. I’ve written an article about it:
How Google DeepMind’s AI system can solve geometry problems like a math Olympian
Geometry is one of the oldest and most beautiful branches of mathematics, but also one of the most challenging for both humans and machines. Solving complex geometry problems requires not only spatial intuition, but also logical reasoning and creativity.
Microsoft Introduces Copilot Pro: A new era of AI-powered office features
Microsoft has recently launched Copilot Pro, a new subscription offering that enhances the existing Copilot with cutting-edge features. This AI-powered tool is designed to supercharge both creativity and productivity. It offers faster performance and priority access to advanced language models like GPT-4 and GPT-4 Turbo during peak times.
The details:
Copilot Pro is available for a monthly subscription of $20.
It provides faster AI image creation with 100 boosts per day with Designer.
The tool is integrated into select Microsoft 365 apps, requiring a Microsoft 365 Personal or Family subscription.
It provides faster web search than ChatGPT
Why it’s important:
The launch of Copilot Pro represents a significant step forward in the integration of AI into everyday productivity tools. By providing advanced features and faster performance, it has the potential to greatly enhance productivity and creativity for individuals and businesses alike. Furthermore, its integration into Microsoft 365 apps demonstrates the growing trend of AI becoming an integral part of our digital tools.
Samsung’s Galaxy S24 line: Camera enhancements and generative AI
Samsung has recently unveiled its Galaxy S24 series, which includes the Galaxy S24, Galaxy S24+, and Galaxy S24 Ultra. Starting at $800, these new flagships offer brighter screens and a host of new photo editing tools. A key highlight of the announcement is Samsung’s heavy leaning into AI, particularly in the wake of the recent generative AI explosion.
The details:
Samsung is among the first device manufacturers to position generative AI as a software enhancement to its smartphone line.
Innovative Galaxy AI editing tools enable simple edits like erase, recompose, and remaster.
For easier and more efficient optimizations, Edit Suggestion uses Galaxy AI to suggest perfectly suitable tweaks for each photo.
Generative Edit can fill in parts of an image background with generative AI.
Any time the device uses generative AI in the editing process, it will add a digital watermark to the image, as well as the metadata.
Another new feature, Instant Slow-mo, generates additional frames to offer a fuller slow-motion experience.
The S24 will be getting Circle for Search, a feature developed in conjunction with Google.
Why it’s important:
The launch of the Galaxy S24 series represents a significant advancement in the integration of AI into everyday devices. By providing advanced features and faster performance, it has the potential to greatly enhance the user experience. Furthermore, its integration of generative AI demonstrates the growing trend of AI becoming an integral part of our digital tools. This could potentially revolutionize the smartphone industry and the way we interact with our devices.
Study finds fingerprints of the same person share strong similarities
A recent study has challenged the long-standing assumption that no two fingerprints, even from different fingers of the same person, are alike. The research, conducted by Gabe Guo, Aniv Ray, Miles Izydorczak, Judah Goldfeder, Hod Lipson, and Wenyao Xu, demonstrates that fingerprints from different fingers of the same person share very strong similarities. This discovery could have significant implications for digital authentication and forensic science.
The details:
The researchers used deep twin neural networks to extract fingerprint representation vectors.
They found that the networks could identify whether two fingerprints were from the same person with up to 77% accuracy, even if they were from different fingers.
The study suggests that ridge orientation, especially near the fingerprint center, explains a substantial part of this similarity.
Contrary to traditional methods, minutiae used in traditional methods are almost nonpredictive.
The experiments suggest that this relationship can increase forensic investigation efficiency by almost two orders of magnitude.
Why it’s important:
This research challenges the traditional assumption in fingerprint biometrics and opens up new possibilities for digital authentication and forensic science. If fingerprints from different fingers of the same person do share strong similarities, it could enhance the efficiency of forensic investigations and improve the reliability of fingerprint-based authentication systems.
BMW to deploy Figure’s humanoid robot at South Carolina plant
BMW has announced a commercial agreement with Figure, a robotics startup, to deploy its first humanoid robot at a BMW manufacturing facility in South Carolina. The Spartanburg plant, BMW’s only facility in the United States, is known for its high yield among the German manufacturer’s factories worldwide. The specifics of the tasks the robot will perform are not yet disclosed, but Figure confirmed that it would start with an initial five tasks.
The details:
Figure’s humanoid robot will be integrated into BMW’s manufacturing processes.
The robot is expected to perform standard manufacturing tasks such as box moving, pick and place, and pallet unloading and loading. Tasks for which factory owners claim to have difficulty retaining human workers.
Figure expects to ship its first commercial robot within a year.
The initial batch of applications will be largely determined by Figure’s early partners like BMW.
Figure is focused on creating a dexterous, human-like hand for manipulation.
Training will involve a mix of approaches, including reinforcement learning, simulation, and teleoperation.
Figure 01 will be learning on the job, refining its approach during real-world testing.
Why it’s important:
The deployment of Figure’s humanoid robot at BMW’s manufacturing facility represents a significant advancement in the field of robotics and AI. The integration of humanoid robots into manufacturing processes could potentially increase efficiency, perform repetitive tasks, and fill roles where there is a shortage of human workers. This partnership between BMW and Figure could pave the way for further advancements in the field and potentially revolutionize the manufacturing industry.
AMIE outperforms human doctors in bedside manner and accuracy of diagnoses.
Google Research has developed a research AI system, AMIE (Articulate Medical Intelligence Explorer), optimized for diagnostic reasoning and conversations. This system is based on a large language model (LLM) and is designed to be a useful conversational partner to clinicians and patients alike. The goal is to increase the availability, accessibility, quality, and consistency of care.
The details:
AMIE is trained on real-world datasets comprising medical reasoning, medical summarization, and real-world clinical conversations.
It uses a novel self-play-based simulated diagnostic dialogue environment with automated feedback mechanisms to enrich and accelerate its learning process.
An inference time chain-of-reasoning strategy is used to improve AMIE’s diagnostic accuracy and conversation quality.
AMIE was tested prospectively in real examples of multi-turn dialogue by simulating consultations with trained actors.
It is optimized for diagnostic conversations, asking questions that help to reduce uncertainty and improve diagnostic accuracy.
AMIE also balances diagnostic accuracy with other requirements of effective clinical communication, such as empathy, fostering a relationship, and providing information clearly.
Why it’s important:
The development of AMIE represents a significant step forward in the field of AI in healthcare. By approximating clinicians’ considerable expertise, AMIE has the potential to greatly enhance the diagnostic process. Its ability to hold rich, empathetic conversations could improve the patient experience, while its diagnostic accuracy could lead to better patient outcomes. Furthermore, its development highlights the potential of AI to transform healthcare by making high-quality care more accessible and consistent.
OpenAI forges its first university partnership
OpenAI, a leading artificial intelligence research lab, has announced its first-ever partnership with a higher education institution. Starting in February, Arizona State University (ASU) will have full access to ChatGPT Enterprise and plans to use it for coursework, tutoring, research, and more. This partnership has been in the works for at least six months.
The details:
ASU plans to build a personalized AI tutor for students, allowing them to create AI avatars for study help.
The university will broaden its prompt engineering course.
ASU will have full access to ChatGPT Enterprise, including access to GPT-4 with no usage caps, and performance up to two times faster than previous versions.
The university will use the tool in ASU’s largest course, Freshman Composition, to offer students writing help.
ASU also plans to use ChatGPT Enterprise to develop AI avatars as a “creative buddy” for studying certain subjects.
The access to ChatGPT Enterprise means students will no longer be limited by usage caps.
OpenAI and ASU’s joint release specified that any prompts the ASU community inputs into ChatGPT “remain secure,” and that OpenAI “does not use this data for its training models”.
Why it’s important:
This partnership represents a significant milestone in the integration of AI into the educational sector. By providing advanced AI tools to a major university, OpenAI is not only enhancing the learning experience for students but also paving the way for further research and development in AI. This could potentially revolutionize the education industry and the way we approach learning and teaching.
How AI can boost the global economy and benefit humanity
Artificial intelligence (AI) is a powerful technology that has the potential to transform the global economy and improve the lives of billions of people. However, it also poses significant challenges and risks, such as displacing workers, increasing inequality, and undermining social cohesion. The question is, how can we ensure that AI serves humanity instead of the other way around?
A new analysis by the IMF suggests that AI will impact almost 40 percent of jobs worldwide, replacing some and complementing others. Furthermore, AI will also affect high-skilled jobs, which means that advanced economies face greater opportunities and threats from AI than emerging and developing economies. As a result, the latter may not have the infrastructure or the skills necessary to harness the benefits of AI and could fall further behind in the global race for innovation and competitiveness.
The IMF report recommends a careful balance of policies to tap the potential of AI while mitigating its negative effects.
The details:
AI will affect almost 40 percent of global employment, with advanced economies facing greater exposure than emerging and developing economies.
AI will impact both routine and non-routine tasks and both low-skilled and high-skilled jobs.
AI could boost productivity and growth, but also increase inequality and polarization within and between countries.
AI could also pose ethical, legal, and social challenges, such as bias, discrimination, accountability, and human dignity.
Policies to leverage AI for the benefit of humanity include investing in education and training, strengthening social protection and safety nets, promoting competition and innovation, and enhancing international cooperation and coordination.
Why it’s important:
AI is a game-changing technology that could reshape the global economy and society in profound ways. It could offer immense opportunities for improving living standards, health, education, and environmental sustainability. But it could also create serious challenges and risks for workers, consumers, businesses, and governments. We need to ensure that AI is aligned with human values and goals and that it is used for good and not evil. As the IMF report states, "AI will transform the global economy. Let’s make sure it benefits humanity."
Google’s novel framework ASPIRE improves language model’s answering skills
Google Research has introduced a novel framework called ASPIRE, aimed at enhancing the selective prediction capabilities of Large Language Models (LLMs). This development is a significant stride in the rapidly evolving landscape of artificial intelligence, particularly in the realm of natural language understanding and generation.
ASPIRE fine-tunes LLMs on question-answering tasks through parameter-efficient fine-tuning, training them to evaluate the correctness of their generated answers. The framework allows LLMs to output an answer along with a confidence score for that answer. ASPIRE is designed to enhance the selective prediction capabilities of LLMs.
The framework fine-tunes LLMs on question-answering tasks and trains them to evaluate the correctness of their generated answers.
ASPIRE allows LLMs to output an answer along with a confidence score for that answer.
The experimental results demonstrate that ASPIRE significantly outperforms state-of-the-art selective prediction methods on a variety of question-answering datasets.
ASPIRE is meticulously designed to enable LLMs to not only answer questions but also evaluate those answers.
Why it’s important:
The introduction of ASPIRE is a significant advancement in the field of AI, particularly for LLMs. By enabling LLMs to evaluate the correctness of their own responses and provide a confidence score, ASPIRE enhances the reliability of these models. This is especially crucial for high-stakes decision-making applications where the inherent uncertainty of model predictions can pose challenges. With ASPIRE, users can better understand the reliability of LLMs deployed in a variety of applications, making it a valuable tool in the ongoing pursuit of improving AI systems.
Meta to Boost AI Hardware with Nvidia and AMD GPUs
Meta, the company formerly known as Facebook, is investing heavily in artificial intelligence (AI) and the hardware that powers it. According to Mark Zuckerberg, Meta will have AI performance equivalent to 600,000 Nvidia H100 GPUs by the end of 2024, using a mix of Nvidia’s Hopper and Blackwell products, as well as AMD’s Instinct MI300 processors. Meta’s goal is to build artificial general intelligence (AGI) that can reason and learn like humans, and to create new AI-centric computing devices such as smart glasses.
Meta unveils MAGNeT: A new open-source text-to-audio model
Meta AI has recently introduced a new model for text-to-audio generation, called MAGNeT, that can create realistic and diverse sounds from text input. MAGNeT is a non-autoregressive transformer model that operates on multiple audio token streams, enabling rapid and efficient audio generation with a single-stage approach. It also uses a novel rescoring method to refine the predictions and enhance the audio quality. MAGNeT promises to revolutionize the fields of music production, sound design, and accessibility, as it can synthesize audio seven times faster than autoregressive baselines. Meta AI has also made the model available to the public through a user-friendly Gradio demo.
Quick news
RunwayML launched a new feature called “Multi Motion Brush” (link)
Complex photo editing in just a few taps with Pixel8 and Pixel 8 Pro (link)
Real time AI translation of Milei’s 2024 Davos speech in his original accent by HeyGen. (link)
FDA cleared the first AI device detecting all major skin cancer by DermoSensor. (link)
Rabbit CEO Jesse Lyu: The first 100,000 Rabbit R1 purchases will also come with one year of its Perplexity Pro subscription. (link)
Microsoft Reading Coach app is now free for anyone with a Microsoft account. (link)
For daily news from the AI and Tech world follow me on:
Be better with AI
In this section, we will provide you with comprehensive tutorials, practical tips, ingenious tricks, and insightful strategies for effectively employing a diverse range of AI tools.
Training costume models to change garments on your character with Scenario
This tutorial was made by Halim Alrasihi (link)
Make your own clothes with ChatGPT. Tell it what you want and get different pictures of it from different sides.
Train a SD LoRa model on Scenario with your clothes. Use a special word (vsvs) to label your pictures.
Use AI Canvas to put your clothes on any model. Use the same word and the inpainting brush to make it look real.
Have fun with your new images
Thanks to Halim Alrasihi (https://twitter.com/HalimAlrasihi)
Tools
🗣️ Gotalk.ai - Transform text into AI voiceovers (link)
🖼️ AI Picasso - Animate AI-generated dances from photos (link)
📝 AI Assist - Advanced spreadsheet helper (link)
🎙️ Podsqueeze 2.0 - Multifunctional podcast content repurposing (link)
🗣️ Byrdhouse AI 2.0 - Multilingual video call AI interpreter (link)
🌈 mentalport - Streamline daily mental wellness (link)
📈 Lume - Automate data mappings with AI (link)
We hope you enjoy this newsletter!
Please feel free to share it with your friends and colleagues and follow me on socials.
what s up with mixing bytedance news with ASU and OpenAI news ? Also never mentioning what even ASU stands for ? are you using some AI text generator that scraps internet for AI news and just collects and publishes here ?
Yay neurosymbolic is the way!