Researchers from Northeastern University, MIT, and University of Glasgow found that parrots can learn to make video calls with other parrots on Facebook Messenger using tablets and smartphones.
The birds initiated calls freely and seemed to understand that a real fellow parrot was on the other end, resulting in positive experiences for the birds.
The parrots engaged in most calls for the maximum allowed time and formed strong preferences for certain parrots, suggesting a reciprocal dynamic similar to human socialization.
The findings suggest that video calls can improve a pet parrot's quality of life, especially for those who cannot physically be close to other birds due to various reasons.
The study highlights the cognitive complexity of parrots and their ability to express themselves through vocalizations and interactions in video calls.
Parrots have been shown in a study to be able to learn and use video calls for communication with other parrots.
Parrots have faster response times and four primaries in their eyes, which allow them to notice differences in hues, but they may find low frame rates and PWM-driven backlight irritating.
Video calls can benefit the quality of life of pet parrots, but there are concerns about keeping animals in captivity and the concept of owning pets.
Sundar Pichai received $226M compensation, boosted by a triennial stock grant
Sundar Pichai's compensation package for 2020 was valued at $226 million.
The majority of this compensation was from a triennial stock grant worth $200 million.
Pichai's pay increase was reportedly due to the company's strong financial performance.
Some are criticizing the compensation as excessive, especially in light of recent controversies at Google.
Pichai's compensation package is significantly higher than that of other tech industry CEOs, including Tim Cook of Apple and Mark Zuckerberg of Facebook.
Sundar Pichai's compensation has surged to $226 million due to a triennial stock grant.
The debate on Pichai's performance as CEO of Google is ongoing, with some praising his ability to build consensus while others criticize his failure to set bold visions.
The discussion on Hacker News also covers Google's lack of innovation, decline as a "dream employer," and issues regarding its mission statement.
Jaron Lanier argues that the term "artificial intelligence" is misleading and suggests that A.I. should be viewed as a tool for social collaboration.
A.I. needs to be designed with transparency to prevent potential harm to humans and avoid misuse.
The concept of data dignity is needed to trace and acknowledge the most unique and influential contributors to big AI models, prevent displacement of workers, and expand the models into new frontiers.
Technical research and policy innovation are required to implement data dignity, and intermediate rights organizations could play a role.
It's important to increase human awareness, responsibility, and participation in automation to make it safe and prevent risks like Deepfakes.
Debate over AI models that emulate human cognition versus those that are created with no internal resemblance to humans, and the importance or limitations of language models in understanding intelligence.
Discussion on the lack of a clear definition for terms such as AGI, consciousness, and intelligence, and the potential for machines to out-reason humans in some tasks.
Debate on the cultural and societal shift towards virtual entities and the potential risks of AI-dominated futures, as well as the importance of regulating corporations' use of AI to protect consumers.
Consideration of the impact of increasing automation on employment and the inherent flaws of capitalism, and suggestions for consumer protection through technological means rather than laws, with calls for government intervention to fund development of such solutions.
Writer expresses empathy for those priced out of housing markets, but insists their own house should be exempt from price reduction
Writer believes that new housing should be affordable, as long as it doesn't affect the high selling price of their current home
Writer acknowledges the high demand for housing, but believes developers should not reap the rewards, instead insisting on personal profits from selling their own home
Writer is willing to support building new housing to lower prices, but only after their own sale clears escrow
Writer hopes that all housing will be affordable in the future, excluding their own backyard.
Rumors suggest iOS 17 app sideloading may only be available in Europe and not the US
App sideloading allows installing apps from outside an official app store, like on macOS or Android
Apple is likely adding sideloading due to regulatory pressure from the EU and may introduce the feature quietly without fanfare
Developers may need to pay extra fees to have apps available outside the iOS App Store
Apple has previously been under pressure from EU regulators over the issue of the App Store being perceived as a monopoly run by Apple, which could be why they are allowing apps to be installed from other sources.
Discussions on app sideloading continue, with arguments for and against the practice
Some believe that iPhones should be considered general computing devices and offer more customization options
Concerns arise over liability and potential security risks with sideloaded apps, but the EU's proposal increases individual freedom at the cost of corporate freedom.
Why the brain's connections to the body are crisscrossed
The article delves into why the brain's connections to the body are crisscrossed, and there is a debate on the reason for this structure and whether it's only for visual processing.
The conversation revolves around the structure of the human brain and its wiring, including cross-wiring between hemispheres, the use of 2D mapping for 3D phenomena, effects of damage to neural connections, and the possibility of alternative wiring methods for biology and robotics.
Split-brain patients confabulate excuses for their behavior due to the separation of the left and right hemispheres, and there are theories about why the brain wiring avoids a single non-split capture.
Fine-tuning large language models has become increasingly important in the field of AI
The article discusses three conventional feature-based and fine-tuning approaches for LLMs along with real-world results on a movie review classifier
Parameter-efficient fine-tuning techniques can minimize computational costs and resources while achieving high modeling performance
Reinforcement learning with human feedback uses human ranking or rating to train a reward model that guides the language model's adaptation to human preferences
The author invites readers to check out their Substack Notes or books and support their work by becoming a free or paid subscriber.
The article explores fine-tuning Large Language Models (LLMs) like GPT-3 for specialist tasks, limitations for factual recall, and opinions on the feasibility of fine-tuning for knowledge recall.
Commenters discuss transformer adapters, SuperCOT, and Tabnine, and the debate around the efficacy of fine-tuning for documentation.
LLMs are viewed as platforms with great potential, with considerations such as context size and tradeoffs between fine-tuning and few-shot prompting for different tasks.