In a groundbreaking collaboration, researchers from the Massachusetts Institute of Technology (MIT) and Harvard University have unveiled a pioneering open-source framework, FAn, to revolutionize real-time object detection, tracking, and following. The team’s paper, titled “Follow Anything: Open-set detection, tracking, and following in real-time,” showcases a system that promises to eliminate the limitations of existing robotic object-following systems.
The core challenge addressed by FAn is the adaptability of robotic systems to new objects. Conventional systems are confined by a closed-set structure, only capable of handling a predefined range of object categories. FAn defies this constraint, introducing an open-set approach that can detect, segment, track, and follow any object in real-time. Notably, it can dynamically adapt to new objects through inputs such as text, images, or click queries. — Read More
Read the Paper
Monthly Archives: August 2023
Meta’s Next Big Open Source AI Dump Will Reportedly Be a Code-Generating Bot
Meta’s language-centric LlaMA AI will soon find itself in the company of a nerdier, coding wiz brother. The company’s next AI release will reportedly be a big coding machine meant to compete against the proprietary software from the likes of OpenAI and Google. The model could see a release as soon as next week.
According to The Information who spoke to two anonymous sources with direct knowledge of the AI, this new model dubbed “Code Llama” will be open source and available free online. This is consistent with the company’s strategy so far of releasing widely available AI software that makes developing new customizable AI models much easier for companies who don’t want to pay OpenAI or others for the privilege. — Read More
Largest genetic study of brain structure identifies how the brain is organised
The largest ever study of the genetics of the brain – encompassing some 36,000 brain scans – has identified more than 4,000 genetic variants linked to brain structure. The results of the study, led by researchers at the University of Cambridge, are published in Nature Genetics today.
Our brains are very complex organs, with huge variety between individuals in terms of the overall volume of the brain, how it is folded and how thick these folds are. Little is known about how our genetic make-up shapes the development of the brain.
… [F]indings have allowed researchers to confirm and, in some cases, identify, how different properties of the brain are genetically linked to each other. — Read More
Arthur AI tested top AI models in math, hallucinations. Here are the results.
Arthur, a platform for monitoring machine learning models, has released new research gauging how top large language models perform in areas like mathematics, so-called “hedging,” and their knowledge of U.S. presidents.
What the numbers say: According to Arthur, OpenAI’s GPT-4 performed best on questions involving combinatorial (counting) mathematics and probability, followed by Anthropic’s Claude 2. Cohere’s model performed the worst in math with zero correct answers and 18 hallucinations, which occur when models generate inaccurate or nonsensical information. — Read More
How to Prevent an AI Catastrophe
In April 2023, a group of academics at Carnegie Mellon University set out to test the chemistry powers of artificial intelligence. To do so, they connected an AI system to a hypothetical laboratory. Then they asked it to produce various substances. With just two words of guidance—“synthesize ibuprofen”—the chemists got the system to identify the steps necessary for laboratory machines to manufacture the painkiller. The AI, as it turned out, knew both the recipe for ibuprofen and how to produce it.
Unfortunately, the researchers quickly discovered that their AI tool would synthesize chemicals far more dangerous than Advil. The program was happy to craft instruction to produce a World War I–era chemical weapon and a common date-rape drug. It almost agreed to synthesize sarin, the notoriously lethal nerve gas, until it Googled the compound’s dark history. The researchers found this safeguard to be cold comfort. “The search function,” they wrote, “can be easily manipulated by altering the terminology.” AI, the chemists concluded, can make devastating weapons. — Read More
Tips for Taking Advantage of Open Large Language Models
Prompting? Few-Shot? Fine-Tuning? Pretraining from scratch? Open LLMs mean more options for developers.
An increasing variety of large language models (LLMs) are open source, or close to it. The proliferation of models with relatively permissive licenses gives developers more options for building applications. — Read More
Open challenges in LLM research
Never before in my life had I seen so many smart people working on the same goal: making LLMs better. After talking to many people working in both industry and academia, I noticed the 10 major research directions that emerged. The first two directions, hallucinations and context learning, are probably the most talked about today. I’m the most excited about numbers 3 (multimodality), 5 (new architecture), and 6 (GPU alternatives).
Open challenges in LLM research
1. Reduce and measure hallucinations
2. Optimize context length and context construction
3. Incorporate other data modalities
4. Make LLMs faster and cheaper
5. Design a new model architecture
6. Develop GPU alternatives
7. Make agents usable
8. Improve learning from human preference
9. Improve the efficiency of the chat interface
10. Build LLMs for non-English languages
Using GPT-4 for content moderation
We use GPT-4 for content policy development and content moderation decisions, enabling more consistent labeling, a faster feedback loop for policy refinement, and less involvement from human moderators.
Content moderation plays a crucial role in sustaining the health of digital platforms. A content moderation system using GPT-4 results in much faster iteration on policy changes, reducing the cycle from months to hours. GPT-4 is also able to interpret rules and nuances in long content policy documentation and adapt instantly to policy updates, resulting in more consistent labeling. We believe this offers a more positive vision of the future of digital platforms, where AI can help moderate online traffic according to platform-specific policy and relieve the mental burden of a large number of human moderators. Anyone with OpenAI API access can implement this approach to create their own AI-assisted moderation system. — Read More
The AI Power Paradox
Can States Learn to Govern Artificial Intelligence—Before It’s Too Late?
It’s 2035, and artificial intelligence is everywhere. AI systems run hospitals, operate airlines, and battle each other in the courtroom. Productivity has spiked to unprecedented levels, and countless previously unimaginable businesses have scaled at blistering speed, generating immense advances in well-being. New products, cures, and innovations hit the market daily, as science and technology kick into overdrive. And yet the world is growing both more unpredictable and more fragile, as terrorists find new ways to menace societies with intelligent, evolving cyberweapons and white-collar workers lose their jobs en masse.
Just a year ago, that scenario would have seemed purely fictional; today, it seems nearly inevitable. Generative AI systems can already write more clearly and persuasively than most humans and can produce original images, art, and even computer code based on simple language prompts. And generative AI is only the tip of the iceberg. Its arrival marks a Big Bang moment, the beginning of a world-changing technological revolution that will remake politics, economies, and societies.
Like past technological waves, AI will pair extraordinary growth and opportunity with immense disruption and risk. But unlike previous waves, it will also initiate a seismic shift in the structure and balance of global power as it threatens the status of nation-states as the world’s primary geopolitical actors. — Read More
Zoom Contradicts Its Own Policy About Training AI on Your Data
Zoom’s Terms of Service say it can train its AI on your calls, videos, and other data. The company says you don’t have to worry about that.
Zoom updated its Terms of Service on Monday after a controversy over the company’s policies about training AI on user data. Although the policy literally says that Zoom reserves the right to train AI on your calls without your explicit permission, the Terms of Service now include an additional line which says, essentially, we promise not to do that.
The company’s Terms of Service call your video, audio, and chat transcripts “Customer Content.” When you click through Zoom’s terms, you agree to give Zoom “perpetual, worldwide, non-exclusive, royalty-free, sublicensable, and transferable license and all other rights” to use that Customer Content for “machine learning, artificial intelligence, training, testing,” and a variety of other product development purposes. The company reserves similar rights for “Service Generated Data,” which includes telemetry data, product usage data, diagnostic data, and other information it gets from analyzing your content and behavior. — Read More
Rick's Cafe AI 